.. _config_reference: Algorithm Configuration Reference ================================= .. note:: This page is **auto-generated** at documentation build time from the (pydantic) JSON schemas of the built-in algorithms' configs -- do not edit it by hand; the generator is ``docs/gen_config_reference.py``. All server and client configs accept additional keyword arguments beyond the listed fields (``extra="allow"``); extras are stored as attributes and are available to the algorithm implementation. Positional construction maps arguments onto the fields in their declaration order, i.e. the order of the tables below; a ``*`` in the Default column marks a required field. .. _config-apfl: APFL ---- **Server config:** ``APFLServerConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``num_iters`` - int - * * - ``num_clients`` - Optional[int] - None * - ``clients_sample_ratio`` - float - * * - ``log_dir`` - Optional[str] - None * - ``txt_logger`` - bool - True * - ``json_logger`` - bool - True * - ``eval_every`` - int - 1 * - ``visiable_gpus`` - Optional[list[int]] - None * - ``extra_observes`` - list[str] - (factory) * - ``seed`` - int - 0 * - ``tag`` - Optional[str] - None * - ``verbose`` - int - 1 * - ``gpu_proportion`` - float - 0.2 * - ``algorithm`` - str - "" * - ``tau`` - int - 1 .. dropdown:: ``APFLServerConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Server config for the APFL algorithm.\n\nParameters\n----------\nnum_iters : int\n The number of (outer) iterations.\nnum_clients : int\n The number of clients.\nclients_sample_ratio : float\n The ratio of clients to sample for each iteration.\ntau : int, default 1,\n synchronization gap.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``log_dir`` : str or pathlib.Path, optional\n The log directory.\n If not specified, will use the default log directory.\n If not absolute, will be relative to the default log directory.\n - ``txt_logger`` : bool, default True\n Whether to use txt logger.\n - ``json_logger`` : bool, default True\n Whether to use json logger.\n - ``eval_every`` : int, default 1\n The number of iterations to evaluate the model.\n - ``visiable_gpus`` : Sequence[int], optional\n Visable GPU IDs for allocating devices for clients.\n Defaults to use all GPUs if available.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training.\n - ``seed`` : int, default 0\n The random seed.\n - ``tag`` : str, optional\n The tag of the experiment.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``gpu_proportion`` : float, default 0.2\n The proportion of clients to use GPU.\n Used to similate the system heterogeneity of the clients.\n Not used in the current version, reserved for future use.", "properties": { "num_iters": { "title": "Num Iters", "type": "integer" }, "num_clients": { "anyOf": [ { "type": "integer" }, { "type": "null" } ], "default": null, "title": "Num Clients" }, "clients_sample_ratio": { "title": "Clients Sample Ratio", "type": "number" }, "log_dir": { "anyOf": [ { "type": "string" }, { "format": "path", "type": "string" }, { "type": "null" } ], "default": null, "title": "Log Dir" }, "txt_logger": { "default": true, "title": "Txt Logger", "type": "boolean" }, "json_logger": { "default": true, "title": "Json Logger", "type": "boolean" }, "eval_every": { "default": 1, "title": "Eval Every", "type": "integer" }, "visiable_gpus": { "anyOf": [ { "items": { "type": "integer" }, "type": "array" }, { "type": "null" } ], "default": null, "title": "Visiable Gpus" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "seed": { "default": 0, "title": "Seed", "type": "integer" }, "tag": { "anyOf": [ { "type": "string" }, { "type": "null" } ], "default": null, "title": "Tag" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "gpu_proportion": { "default": 0.2, "title": "Gpu Proportion", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "tau": { "default": 1, "title": "Tau", "type": "integer" } }, "required": [ "num_iters", "clients_sample_ratio" ], "title": "APFLServerConfig", "type": "object" } **Client config:** ``APFLClientConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``batch_size`` - int - * * - ``num_epochs`` - int - * * - ``lr`` - float - 0.01 * - ``optimizer`` - str - "SGD" * - ``scheduler`` - Optional[dict] - None * - ``extra_observes`` - list[str] - (factory) * - ``verbose`` - int - 1 * - ``latency`` - float - 0.0 * - ``algorithm`` - str - "" * - ``mixture_weight`` - float - 0.5 .. dropdown:: ``APFLClientConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Client config for the APFL algorithm.\n\nParameters\n----------\nbatch_size : int\n The batch size.\nnum_epochs : int\n The number of epochs.\noptimizer : str, default \"SGD\"\n The name of the optimizer to solve the local (inner) problem.\nlr : float, default 1e-2\n The learning rate.\nmixture_weight : float, default 0.5\n The weight of the local model parameters and the global model parameters.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``scheduler`` : dict, optional\n The scheduler config.\n None for no scheduler, using constant learning rate.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training,\n which would be recorded in evaluated metrics,\n sent to the server, and written to the log file.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``latency`` : float, default 0.0\n The latency of the client.\n Not used in the current version, reserved for future use.", "properties": { "batch_size": { "title": "Batch Size", "type": "integer" }, "num_epochs": { "title": "Num Epochs", "type": "integer" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "optimizer": { "default": "SGD", "title": "Optimizer", "type": "string" }, "scheduler": { "anyOf": [ { "additionalProperties": true, "type": "object" }, { "type": "null" } ], "default": null, "title": "Scheduler" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "latency": { "default": 0.0, "title": "Latency", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "mixture_weight": { "default": 0.5, "title": "Mixture Weight", "type": "number" } }, "required": [ "batch_size", "num_epochs" ], "title": "APFLClientConfig", "type": "object" } .. _config-ditto: Ditto ----- **Server config:** ``DittoServerConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``num_iters`` - int - * * - ``num_clients`` - Optional[int] - None * - ``clients_sample_ratio`` - float - * * - ``log_dir`` - Optional[str] - None * - ``txt_logger`` - bool - True * - ``json_logger`` - bool - True * - ``eval_every`` - int - 1 * - ``visiable_gpus`` - Optional[list[int]] - None * - ``extra_observes`` - list[str] - (factory) * - ``seed`` - int - 0 * - ``tag`` - Optional[str] - None * - ``verbose`` - int - 1 * - ``gpu_proportion`` - float - 0.2 * - ``algorithm`` - str - "" .. dropdown:: ``DittoServerConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Server config for the Ditto algorithm.\n\nParameters\n----------\nnum_iters : int\n The number of (outer) iterations.\nnum_clients : int\n The number of clients.\nclients_sample_ratio : float\n The ratio of clients to sample for each iteration.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``log_dir`` : str or pathlib.Path, optional\n The log directory.\n If not specified, will use the default log directory.\n If not absolute, will be relative to the default log directory.\n - ``txt_logger`` : bool, default True\n Whether to use txt logger.\n - ``json_logger`` : bool, default True\n Whether to use json logger.\n - ``eval_every`` : int, default 1\n The number of iterations to evaluate the model.\n - ``visiable_gpus`` : Sequence[int], optional\n Visable GPU IDs for allocating devices for clients.\n Defaults to use all GPUs if available.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training.\n - ``seed`` : int, default 0\n The random seed.\n - ``tag`` : str, optional\n The tag of the experiment.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``gpu_proportion`` : float, default 0.2\n The proportion of clients to use GPU.\n Used to similate the system heterogeneity of the clients.\n Not used in the current version, reserved for future use.", "properties": { "num_iters": { "title": "Num Iters", "type": "integer" }, "num_clients": { "anyOf": [ { "type": "integer" }, { "type": "null" } ], "default": null, "title": "Num Clients" }, "clients_sample_ratio": { "title": "Clients Sample Ratio", "type": "number" }, "log_dir": { "anyOf": [ { "type": "string" }, { "format": "path", "type": "string" }, { "type": "null" } ], "default": null, "title": "Log Dir" }, "txt_logger": { "default": true, "title": "Txt Logger", "type": "boolean" }, "json_logger": { "default": true, "title": "Json Logger", "type": "boolean" }, "eval_every": { "default": 1, "title": "Eval Every", "type": "integer" }, "visiable_gpus": { "anyOf": [ { "items": { "type": "integer" }, "type": "array" }, { "type": "null" } ], "default": null, "title": "Visiable Gpus" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "seed": { "default": 0, "title": "Seed", "type": "integer" }, "tag": { "anyOf": [ { "type": "string" }, { "type": "null" } ], "default": null, "title": "Tag" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "gpu_proportion": { "default": 0.2, "title": "Gpu Proportion", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" } }, "required": [ "num_iters", "clients_sample_ratio" ], "title": "DittoServerConfig", "type": "object" } **Client config:** ``DittoClientConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``batch_size`` - int - * * - ``num_epochs`` - int - * * - ``lr`` - float - 0.01 * - ``optimizer`` - str - "SGD" * - ``scheduler`` - Optional[dict] - None * - ``extra_observes`` - list[str] - (factory) * - ``verbose`` - int - 1 * - ``latency`` - float - 0.0 * - ``algorithm`` - str - "" * - ``optimizer_per`` - str - "ProxSGD" * - ``prox`` - float - 0.01 * - ``lr_per`` - Optional[float] - None .. dropdown:: ``DittoClientConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Client config for the Ditto algorithm.\n\nParameters\n----------\nbatch_size : int\n The batch size.\nnum_epochs : int\n The number of epochs.\noptimizer : str, default \"SGD\"\n The name of the optimizer to solve the local (inner) problem.\noptimizer_per : str, default \"ProxSGD\"\n The name of the optimizer to solve the personalization problem.\nprox : float, default 0.01\n Coefficient of the proximal term.\nlr : float, default 1e-2\n The learning rate.\nlr_per : float, optional\n The learning rate for personalization.\n If not specified, ``lr_per`` will be set to ``lr``.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``scheduler`` : dict, optional\n The scheduler config.\n None for no scheduler, using constant learning rate.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training,\n which would be recorded in evaluated metrics,\n sent to the server, and written to the log file.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``latency`` : float, default 0.0\n The latency of the client.\n Not used in the current version, reserved for future use.", "properties": { "batch_size": { "title": "Batch Size", "type": "integer" }, "num_epochs": { "title": "Num Epochs", "type": "integer" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "optimizer": { "default": "SGD", "title": "Optimizer", "type": "string" }, "scheduler": { "anyOf": [ { "additionalProperties": true, "type": "object" }, { "type": "null" } ], "default": null, "title": "Scheduler" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "latency": { "default": 0.0, "title": "Latency", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "optimizer_per": { "default": "ProxSGD", "title": "Optimizer Per", "type": "string" }, "prox": { "default": 0.01, "title": "Prox", "type": "number" }, "lr_per": { "anyOf": [ { "type": "number" }, { "type": "null" } ], "default": null, "title": "Lr Per" } }, "required": [ "batch_size", "num_epochs" ], "title": "DittoClientConfig", "type": "object" } .. _config-fpfc: FPFC ---- **Server config:** ``FPFCServerConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``num_iters`` - int - * * - ``num_clients`` - Optional[int] - None * - ``clients_sample_ratio`` - float - * * - ``log_dir`` - Optional[str] - None * - ``txt_logger`` - bool - True * - ``json_logger`` - bool - True * - ``eval_every`` - int - 1 * - ``visiable_gpus`` - Optional[list[int]] - None * - ``extra_observes`` - list[str] - (factory) * - ``seed`` - int - 0 * - ``tag`` - Optional[str] - None * - ``verbose`` - int - 1 * - ``gpu_proportion`` - float - 0.2 * - ``algorithm`` - str - "" * - ``optimizer`` - str - "Avg" * - ``lr`` - float - 1 * - ``betas`` - list - [0, 1] * - ``tau`` - float - 1 .. dropdown:: ``FPFCServerConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Server config for the FPFC algorithm.\n\nParameters\n----------\nnum_iters : int\n The number of (outer) iterations.\nnum_clients : int\n The number of clients.\nclients_sample_ratio : float, default 1\n The ratio of clients to participate in each round.\nrho : float, default 1.0\n The penalty parameter :math:`\\rho` of the augmented Lagrangian (ADMM).\nxi : float, default 1e-4\n The smoothing parameter :math:`\\xi` of the (smoothed SCAD-type) fusion penalty.\nlam : float, default 1.0\n The penalty coefficient :math:`\\lambda` of the pairwise fusion penalty.\na : float, default 3.7\n The parameter :math:`a` of the fusion penalty, should be larger than 1.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``log_dir`` : str or pathlib.Path, optional\n The log directory.\n If not specified, will use the default log directory.\n If not absolute, will be relative to the default log directory.\n - ``txt_logger`` : bool, default True\n Whether to use txt logger.\n - ``json_logger`` : bool, default True\n Whether to use json logger.\n - ``eval_every`` : int, default 1\n The number of iterations to evaluate the model.\n - ``visiable_gpus`` : Sequence[int], optional\n Visable GPU IDs for allocating devices for clients.\n Defaults to use all GPUs if available.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training.\n - ``seed`` : int, default 0\n The random seed.\n - ``tag`` : str, optional\n The tag of the experiment.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``gpu_proportion`` : float, default 0.2\n The proportion of clients to use GPU.\n Used to similate the system heterogeneity of the clients.\n Not used in the current version, reserved for future use.", "properties": { "num_iters": { "title": "Num Iters", "type": "integer" }, "num_clients": { "anyOf": [ { "type": "integer" }, { "type": "null" } ], "default": null, "title": "Num Clients" }, "clients_sample_ratio": { "title": "Clients Sample Ratio", "type": "number" }, "log_dir": { "anyOf": [ { "type": "string" }, { "format": "path", "type": "string" }, { "type": "null" } ], "default": null, "title": "Log Dir" }, "txt_logger": { "default": true, "title": "Txt Logger", "type": "boolean" }, "json_logger": { "default": true, "title": "Json Logger", "type": "boolean" }, "eval_every": { "default": 1, "title": "Eval Every", "type": "integer" }, "visiable_gpus": { "anyOf": [ { "items": { "type": "integer" }, "type": "array" }, { "type": "null" } ], "default": null, "title": "Visiable Gpus" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "seed": { "default": 0, "title": "Seed", "type": "integer" }, "tag": { "anyOf": [ { "type": "string" }, { "type": "null" } ], "default": null, "title": "Tag" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "gpu_proportion": { "default": 0.2, "title": "Gpu Proportion", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "optimizer": { "default": "Avg", "title": "Optimizer", "type": "string" }, "lr": { "default": 1, "title": "Lr", "type": "number" }, "betas": { "default": [ 0, 1 ], "maxItems": 2, "minItems": 2, "prefixItems": [ { "type": "number" }, { "type": "number" } ], "title": "Betas", "type": "array" }, "tau": { "default": 1, "title": "Tau", "type": "number" } }, "required": [ "num_iters", "clients_sample_ratio" ], "title": "FPFCServerConfig", "type": "object" } **Client config:** ``FPFCClientConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``batch_size`` - int - * * - ``num_epochs`` - int - * * - ``lr`` - float - 0.01 * - ``optimizer`` - str - "SGD" * - ``scheduler`` - Optional[dict] - None * - ``extra_observes`` - list[str] - (factory) * - ``verbose`` - int - 1 * - ``latency`` - float - 0.0 * - ``algorithm`` - str - "" .. dropdown:: ``FPFCClientConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Client config for the FPFC algorithm.\n\nParameters\n----------\nbatch_size : int\n The batch size.\nnum_epochs : int\n The number of (local) epochs, i.e. :math:`T_i` in the paper.\nlr : float, default 1e-2\n The learning rate, i.e. the stepsize :math:`\\alpha` in the paper.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``scheduler`` : dict, optional\n The scheduler config.\n None for no scheduler, using constant learning rate.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training,\n which would be recorded in evaluated metrics,\n sent to the server, and written to the log file.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``latency`` : float, default 0.0\n The latency of the client.\n Not used in the current version, reserved for future use.", "properties": { "batch_size": { "title": "Batch Size", "type": "integer" }, "num_epochs": { "title": "Num Epochs", "type": "integer" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "optimizer": { "default": "SGD", "title": "Optimizer", "type": "string" }, "scheduler": { "anyOf": [ { "additionalProperties": true, "type": "object" }, { "type": "null" } ], "default": null, "title": "Scheduler" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "latency": { "default": 0.0, "title": "Latency", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" } }, "required": [ "batch_size", "num_epochs" ], "title": "FPFCClientConfig", "type": "object" } .. _config-fedadagrad: FedAdagrad ---------- **Server config:** ``FedAdagradServerConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``num_iters`` - int - * * - ``num_clients`` - Optional[int] - None * - ``clients_sample_ratio`` - float - * * - ``log_dir`` - Optional[str] - None * - ``txt_logger`` - bool - True * - ``json_logger`` - bool - True * - ``eval_every`` - int - 1 * - ``visiable_gpus`` - Optional[list[int]] - None * - ``extra_observes`` - list[str] - (factory) * - ``seed`` - int - 0 * - ``tag`` - Optional[str] - None * - ``verbose`` - int - 1 * - ``gpu_proportion`` - float - 0.2 * - ``algorithm`` - str - "" * - ``optimizer`` - str - "Adagrad" * - ``lr`` - float - 0.01 * - ``betas`` - list - [0.0, 0.99] * - ``tau`` - float - 1e-05 .. dropdown:: ``FedAdagradServerConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Server config for the FedOpt algorithm.\n\nParameters\n----------\nnum_iters : int\n The number of (outer) iterations.\nnum_clients : int\n The number of clients.\nclients_sample_ratio : float\n The ratio of clients to sample for each iteration.\nlr : float, default 1e-2\n The learning rate.\nbetas : Sequence[float], default (0.9, 0.99)\n The beta values for the optimizer.\ntau : float, default 1e-5\n The tau value for the optimizer,\n which controls the degree of adaptivity of the algorithm.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``log_dir`` : str or pathlib.Path, optional\n The log directory.\n If not specified, will use the default log directory.\n If not absolute, will be relative to the default log directory.\n - ``txt_logger`` : bool, default True\n Whether to use txt logger.\n - ``json_logger`` : bool, default True\n Whether to use json logger.\n - ``eval_every`` : int, default 1\n The number of iterations to evaluate the model.\n - ``visiable_gpus`` : Sequence[int], optional\n Visable GPU IDs for allocating devices for clients.\n Defaults to use all GPUs if available.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training.\n - ``seed`` : int, default 0\n The random seed.\n - ``tag`` : str, optional\n The tag of the experiment.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``gpu_proportion`` : float, default 0.2\n The proportion of clients to use GPU.\n Used to similate the system heterogeneity of the clients.\n Not used in the current version, reserved for future use.", "properties": { "num_iters": { "title": "Num Iters", "type": "integer" }, "num_clients": { "anyOf": [ { "type": "integer" }, { "type": "null" } ], "default": null, "title": "Num Clients" }, "clients_sample_ratio": { "title": "Clients Sample Ratio", "type": "number" }, "log_dir": { "anyOf": [ { "type": "string" }, { "format": "path", "type": "string" }, { "type": "null" } ], "default": null, "title": "Log Dir" }, "txt_logger": { "default": true, "title": "Txt Logger", "type": "boolean" }, "json_logger": { "default": true, "title": "Json Logger", "type": "boolean" }, "eval_every": { "default": 1, "title": "Eval Every", "type": "integer" }, "visiable_gpus": { "anyOf": [ { "items": { "type": "integer" }, "type": "array" }, { "type": "null" } ], "default": null, "title": "Visiable Gpus" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "seed": { "default": 0, "title": "Seed", "type": "integer" }, "tag": { "anyOf": [ { "type": "string" }, { "type": "null" } ], "default": null, "title": "Tag" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "gpu_proportion": { "default": 0.2, "title": "Gpu Proportion", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "optimizer": { "default": "Adagrad", "title": "Optimizer", "type": "string" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "betas": { "default": [ 0.0, 0.99 ], "maxItems": 2, "minItems": 2, "prefixItems": [ { "type": "number" }, { "type": "number" } ], "title": "Betas", "type": "array" }, "tau": { "default": 1e-05, "title": "Tau", "type": "number" } }, "required": [ "num_iters", "clients_sample_ratio" ], "title": "FedAdagradServerConfig", "type": "object" } **Client config:** ``FedAdagradClientConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``batch_size`` - int - * * - ``num_epochs`` - int - * * - ``lr`` - float - 0.01 * - ``optimizer`` - str - "SGD" * - ``scheduler`` - Optional[dict] - None * - ``extra_observes`` - list[str] - (factory) * - ``verbose`` - int - 1 * - ``latency`` - float - 0.0 * - ``algorithm`` - str - "" .. dropdown:: ``FedAdagradClientConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Client config for the FedAdagrad algorithm.\n\nParameters\n----------\nbatch_size : int\n The batch size.\nnum_epochs : int\n The number of epochs.\nlr : float, default 1e-2\n The learning rate.\noptimizer : str, default \"SGD\"\n The name of the optimizer to solve the local (inner) problem.\n**kwargs : dict, optional\n Additional keyword arguments for specific algorithms\n (FedAvg, FedAdagrad, FedYogi, FedAdam). And\n\n - ``scheduler`` : dict, optional\n The scheduler config.\n None for no scheduler, using constant learning rate.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training,\n which would be recorded in evaluated metrics,\n sent to the server, and written to the log file.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``latency`` : float, default 0.0\n The latency of the client.\n Not used in the current version, reserved for future use.", "properties": { "batch_size": { "title": "Batch Size", "type": "integer" }, "num_epochs": { "title": "Num Epochs", "type": "integer" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "optimizer": { "default": "SGD", "title": "Optimizer", "type": "string" }, "scheduler": { "anyOf": [ { "additionalProperties": true, "type": "object" }, { "type": "null" } ], "default": null, "title": "Scheduler" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "latency": { "default": 0.0, "title": "Latency", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" } }, "required": [ "batch_size", "num_epochs" ], "title": "FedAdagradClientConfig", "type": "object" } .. _config-fedadam: FedAdam ------- **Server config:** ``FedAdamServerConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``num_iters`` - int - * * - ``num_clients`` - Optional[int] - None * - ``clients_sample_ratio`` - float - * * - ``log_dir`` - Optional[str] - None * - ``txt_logger`` - bool - True * - ``json_logger`` - bool - True * - ``eval_every`` - int - 1 * - ``visiable_gpus`` - Optional[list[int]] - None * - ``extra_observes`` - list[str] - (factory) * - ``seed`` - int - 0 * - ``tag`` - Optional[str] - None * - ``verbose`` - int - 1 * - ``gpu_proportion`` - float - 0.2 * - ``algorithm`` - str - "" * - ``optimizer`` - str - "Adam" * - ``lr`` - float - 0.01 * - ``betas`` - list - [0.9, 0.99] * - ``tau`` - float - 1e-05 .. dropdown:: ``FedAdamServerConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Server config for the FedOpt algorithm.\n\nParameters\n----------\nnum_iters : int\n The number of (outer) iterations.\nnum_clients : int\n The number of clients.\nclients_sample_ratio : float\n The ratio of clients to sample for each iteration.\nlr : float, default 1e-2\n The learning rate.\nbetas : Sequence[float], default (0.9, 0.99)\n The beta values for the optimizer.\ntau : float, default 1e-5\n The tau value for the optimizer,\n which controls the degree of adaptivity of the algorithm.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``log_dir`` : str or pathlib.Path, optional\n The log directory.\n If not specified, will use the default log directory.\n If not absolute, will be relative to the default log directory.\n - ``txt_logger`` : bool, default True\n Whether to use txt logger.\n - ``json_logger`` : bool, default True\n Whether to use json logger.\n - ``eval_every`` : int, default 1\n The number of iterations to evaluate the model.\n - ``visiable_gpus`` : Sequence[int], optional\n Visable GPU IDs for allocating devices for clients.\n Defaults to use all GPUs if available.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training.\n - ``seed`` : int, default 0\n The random seed.\n - ``tag`` : str, optional\n The tag of the experiment.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``gpu_proportion`` : float, default 0.2\n The proportion of clients to use GPU.\n Used to similate the system heterogeneity of the clients.\n Not used in the current version, reserved for future use.", "properties": { "num_iters": { "title": "Num Iters", "type": "integer" }, "num_clients": { "anyOf": [ { "type": "integer" }, { "type": "null" } ], "default": null, "title": "Num Clients" }, "clients_sample_ratio": { "title": "Clients Sample Ratio", "type": "number" }, "log_dir": { "anyOf": [ { "type": "string" }, { "format": "path", "type": "string" }, { "type": "null" } ], "default": null, "title": "Log Dir" }, "txt_logger": { "default": true, "title": "Txt Logger", "type": "boolean" }, "json_logger": { "default": true, "title": "Json Logger", "type": "boolean" }, "eval_every": { "default": 1, "title": "Eval Every", "type": "integer" }, "visiable_gpus": { "anyOf": [ { "items": { "type": "integer" }, "type": "array" }, { "type": "null" } ], "default": null, "title": "Visiable Gpus" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "seed": { "default": 0, "title": "Seed", "type": "integer" }, "tag": { "anyOf": [ { "type": "string" }, { "type": "null" } ], "default": null, "title": "Tag" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "gpu_proportion": { "default": 0.2, "title": "Gpu Proportion", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "optimizer": { "default": "Adam", "title": "Optimizer", "type": "string" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "betas": { "default": [ 0.9, 0.99 ], "maxItems": 2, "minItems": 2, "prefixItems": [ { "type": "number" }, { "type": "number" } ], "title": "Betas", "type": "array" }, "tau": { "default": 1e-05, "title": "Tau", "type": "number" } }, "required": [ "num_iters", "clients_sample_ratio" ], "title": "FedAdamServerConfig", "type": "object" } **Client config:** ``FedAdamClientConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``batch_size`` - int - * * - ``num_epochs`` - int - * * - ``lr`` - float - 0.01 * - ``optimizer`` - str - "SGD" * - ``scheduler`` - Optional[dict] - None * - ``extra_observes`` - list[str] - (factory) * - ``verbose`` - int - 1 * - ``latency`` - float - 0.0 * - ``algorithm`` - str - "" .. dropdown:: ``FedAdamClientConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Client config for the FedAdam algorithm.\n\nParameters\n----------\nbatch_size : int\n The batch size.\nnum_epochs : int\n The number of epochs.\nlr : float, default 1e-2\n The learning rate.\noptimizer : str, default \"SGD\"\n The name of the optimizer to solve the local (inner) problem.\n**kwargs : dict, optional\n Additional keyword arguments for specific algorithms\n (FedAvg, FedAdagrad, FedYogi, FedAdam). And\n\n - ``scheduler`` : dict, optional\n The scheduler config.\n None for no scheduler, using constant learning rate.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training,\n which would be recorded in evaluated metrics,\n sent to the server, and written to the log file.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``latency`` : float, default 0.0\n The latency of the client.\n Not used in the current version, reserved for future use.", "properties": { "batch_size": { "title": "Batch Size", "type": "integer" }, "num_epochs": { "title": "Num Epochs", "type": "integer" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "optimizer": { "default": "SGD", "title": "Optimizer", "type": "string" }, "scheduler": { "anyOf": [ { "additionalProperties": true, "type": "object" }, { "type": "null" } ], "default": null, "title": "Scheduler" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "latency": { "default": 0.0, "title": "Latency", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" } }, "required": [ "batch_size", "num_epochs" ], "title": "FedAdamClientConfig", "type": "object" } .. _config-fedavg: FedAvg ------ **Server config:** ``FedAvgServerConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``num_iters`` - int - * * - ``num_clients`` - Optional[int] - None * - ``clients_sample_ratio`` - float - * * - ``log_dir`` - Optional[str] - None * - ``txt_logger`` - bool - True * - ``json_logger`` - bool - True * - ``eval_every`` - int - 1 * - ``visiable_gpus`` - Optional[list[int]] - None * - ``extra_observes`` - list[str] - (factory) * - ``seed`` - int - 0 * - ``tag`` - Optional[str] - None * - ``verbose`` - int - 1 * - ``gpu_proportion`` - float - 0.2 * - ``algorithm`` - str - "" * - ``optimizer`` - str - "Avg" * - ``lr`` - float - 1 * - ``betas`` - list - [0, 1] * - ``tau`` - float - 1 .. dropdown:: ``FedAvgServerConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Server config for the FedOpt algorithm.\n\nParameters\n----------\nnum_iters : int\n The number of (outer) iterations.\nnum_clients : int\n The number of clients.\nclients_sample_ratio : float\n The ratio of clients to sample for each iteration.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``log_dir`` : str or pathlib.Path, optional\n The log directory.\n If not specified, will use the default log directory.\n If not absolute, will be relative to the default log directory.\n - ``txt_logger`` : bool, default True\n Whether to use txt logger.\n - ``json_logger`` : bool, default True\n Whether to use json logger.\n - ``eval_every`` : int, default 1\n The number of iterations to evaluate the model.\n - ``visiable_gpus`` : Sequence[int], optional\n Visable GPU IDs for allocating devices for clients.\n Defaults to use all GPUs if available.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training.\n - ``seed`` : int, default 0\n The random seed.\n - ``tag`` : str, optional\n The tag of the experiment.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``gpu_proportion`` : float, default 0.2\n The proportion of clients to use GPU.\n Used to similate the system heterogeneity of the clients.\n Not used in the current version, reserved for future use.", "properties": { "num_iters": { "title": "Num Iters", "type": "integer" }, "num_clients": { "anyOf": [ { "type": "integer" }, { "type": "null" } ], "default": null, "title": "Num Clients" }, "clients_sample_ratio": { "title": "Clients Sample Ratio", "type": "number" }, "log_dir": { "anyOf": [ { "type": "string" }, { "format": "path", "type": "string" }, { "type": "null" } ], "default": null, "title": "Log Dir" }, "txt_logger": { "default": true, "title": "Txt Logger", "type": "boolean" }, "json_logger": { "default": true, "title": "Json Logger", "type": "boolean" }, "eval_every": { "default": 1, "title": "Eval Every", "type": "integer" }, "visiable_gpus": { "anyOf": [ { "items": { "type": "integer" }, "type": "array" }, { "type": "null" } ], "default": null, "title": "Visiable Gpus" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "seed": { "default": 0, "title": "Seed", "type": "integer" }, "tag": { "anyOf": [ { "type": "string" }, { "type": "null" } ], "default": null, "title": "Tag" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "gpu_proportion": { "default": 0.2, "title": "Gpu Proportion", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "optimizer": { "default": "Avg", "title": "Optimizer", "type": "string" }, "lr": { "default": 1, "title": "Lr", "type": "number" }, "betas": { "default": [ 0, 1 ], "maxItems": 2, "minItems": 2, "prefixItems": [ { "type": "number" }, { "type": "number" } ], "title": "Betas", "type": "array" }, "tau": { "default": 1, "title": "Tau", "type": "number" } }, "required": [ "num_iters", "clients_sample_ratio" ], "title": "FedAvgServerConfig", "type": "object" } **Client config:** ``FedAvgClientConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``batch_size`` - int - * * - ``num_epochs`` - int - * * - ``lr`` - float - 0.01 * - ``optimizer`` - str - "SGD" * - ``scheduler`` - Optional[dict] - None * - ``extra_observes`` - list[str] - (factory) * - ``verbose`` - int - 1 * - ``latency`` - float - 0.0 * - ``algorithm`` - str - "" .. dropdown:: ``FedAvgClientConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Client config for the FedAvg algorithm.\n\nParameters\n----------\nbatch_size : int\n The batch size.\nnum_epochs : int\n The number of epochs.\nlr : float, default 1e-2\n The learning rate.\noptimizer : str, default \"SGD\"\n The name of the optimizer to solve the local (inner) problem.\n**kwargs : dict, optional\n Additional keyword arguments for specific algorithms\n (FedAvg, FedAdagrad, FedYogi, FedAdam). And\n\n - ``scheduler`` : dict, optional\n The scheduler config.\n None for no scheduler, using constant learning rate.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training,\n which would be recorded in evaluated metrics,\n sent to the server, and written to the log file.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``latency`` : float, default 0.0\n The latency of the client.\n Not used in the current version, reserved for future use.", "properties": { "batch_size": { "title": "Batch Size", "type": "integer" }, "num_epochs": { "title": "Num Epochs", "type": "integer" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "optimizer": { "default": "SGD", "title": "Optimizer", "type": "string" }, "scheduler": { "anyOf": [ { "additionalProperties": true, "type": "object" }, { "type": "null" } ], "default": null, "title": "Scheduler" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "latency": { "default": 0.0, "title": "Latency", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" } }, "required": [ "batch_size", "num_epochs" ], "title": "FedAvgClientConfig", "type": "object" } .. _config-fedcr: FedCR ----- **Server config:** ``FedCRServerConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``num_iters`` - int - * * - ``num_clients`` - Optional[int] - None * - ``clients_sample_ratio`` - float - 1 * - ``log_dir`` - Optional[str] - None * - ``txt_logger`` - bool - True * - ``json_logger`` - bool - True * - ``eval_every`` - int - 1 * - ``visiable_gpus`` - Optional[list[int]] - None * - ``extra_observes`` - list[str] - (factory) * - ``seed`` - int - 0 * - ``tag`` - Optional[str] - None * - ``verbose`` - int - 1 * - ``gpu_proportion`` - float - 0.2 * - ``algorithm`` - str - "" * - ``optimizer`` - str - "Avg" * - ``lr`` - float - 1 * - ``betas`` - list - [0, 1] * - ``tau`` - float - 1 * - ``beta`` - float - 0.005 .. dropdown:: ``FedCRServerConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Server config for the FedCR algorithm.\n\nParameters\n----------\nnum_iters : int\n The number of (outer) iterations.\nnum_clients : int\n The number of clients.\nclients_sample_ratio : float, default 1\n The ratio of clients to participate in each round.\nbeta : float, default 0.005\n The coefficient :math:`\\beta` of the CMI (KL alignment) regularizer.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``log_dir`` : str or pathlib.Path, optional\n The log directory.\n If not specified, will use the default log directory.\n If not absolute, will be relative to the default log directory.\n - ``txt_logger`` : bool, default True\n Whether to use txt logger.\n - ``json_logger`` : bool, default True\n Whether to use json logger.\n - ``eval_every`` : int, default 1\n The number of iterations to evaluate the model.\n - ``visiable_gpus`` : Sequence[int], optional\n Visable GPU IDs for allocating devices for clients.\n Defaults to use all GPUs if available.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training.\n - ``seed`` : int, default 0\n The random seed.\n - ``tag`` : str, optional\n The tag of the experiment.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``gpu_proportion`` : float, default 0.2\n The proportion of clients to use GPU.\n Used to similate the system heterogeneity of the clients.\n Not used in the current version, reserved for future use.", "properties": { "num_iters": { "title": "Num Iters", "type": "integer" }, "num_clients": { "anyOf": [ { "type": "integer" }, { "type": "null" } ], "default": null, "title": "Num Clients" }, "clients_sample_ratio": { "default": 1, "title": "Clients Sample Ratio", "type": "number" }, "log_dir": { "anyOf": [ { "type": "string" }, { "format": "path", "type": "string" }, { "type": "null" } ], "default": null, "title": "Log Dir" }, "txt_logger": { "default": true, "title": "Txt Logger", "type": "boolean" }, "json_logger": { "default": true, "title": "Json Logger", "type": "boolean" }, "eval_every": { "default": 1, "title": "Eval Every", "type": "integer" }, "visiable_gpus": { "anyOf": [ { "items": { "type": "integer" }, "type": "array" }, { "type": "null" } ], "default": null, "title": "Visiable Gpus" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "seed": { "default": 0, "title": "Seed", "type": "integer" }, "tag": { "anyOf": [ { "type": "string" }, { "type": "null" } ], "default": null, "title": "Tag" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "gpu_proportion": { "default": 0.2, "title": "Gpu Proportion", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "optimizer": { "default": "Avg", "title": "Optimizer", "type": "string" }, "lr": { "default": 1, "title": "Lr", "type": "number" }, "betas": { "default": [ 0, 1 ], "maxItems": 2, "minItems": 2, "prefixItems": [ { "type": "number" }, { "type": "number" } ], "title": "Betas", "type": "array" }, "tau": { "default": 1, "title": "Tau", "type": "number" }, "beta": { "default": 0.005, "title": "Beta", "type": "number" } }, "required": [ "num_iters" ], "title": "FedCRServerConfig", "type": "object" } **Client config:** ``FedCRClientConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``batch_size`` - int - * * - ``num_epochs`` - int - * * - ``lr`` - float - 0.01 * - ``optimizer`` - str - "SGD" * - ``scheduler`` - Optional[dict] - None * - ``extra_observes`` - list[str] - (factory) * - ``verbose`` - int - 1 * - ``latency`` - float - 0.0 * - ``algorithm`` - str - "" .. dropdown:: ``FedCRClientConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Client config for the FedCR algorithm.\n\nParameters\n----------\nbatch_size : int\n The batch size.\nnum_epochs : int\n The number of (local) epochs.\nlr : float, default 1e-2\n The learning rate.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``scheduler`` : dict, optional\n The scheduler config.\n None for no scheduler, using constant learning rate.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training,\n which would be recorded in evaluated metrics,\n sent to the server, and written to the log file.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``latency`` : float, default 0.0\n The latency of the client.\n Not used in the current version, reserved for future use.", "properties": { "batch_size": { "title": "Batch Size", "type": "integer" }, "num_epochs": { "title": "Num Epochs", "type": "integer" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "optimizer": { "default": "SGD", "title": "Optimizer", "type": "string" }, "scheduler": { "anyOf": [ { "additionalProperties": true, "type": "object" }, { "type": "null" } ], "default": null, "title": "Scheduler" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "latency": { "default": 0.0, "title": "Latency", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" } }, "required": [ "batch_size", "num_epochs" ], "title": "FedCRClientConfig", "type": "object" } .. _config-feddc: FedDC ----- **Server config:** ``FedDCServerConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``num_iters`` - int - * * - ``num_clients`` - Optional[int] - None * - ``clients_sample_ratio`` - float - 1 * - ``log_dir`` - Optional[str] - None * - ``txt_logger`` - bool - True * - ``json_logger`` - bool - True * - ``eval_every`` - int - 1 * - ``visiable_gpus`` - Optional[list[int]] - None * - ``extra_observes`` - list[str] - (factory) * - ``seed`` - int - 0 * - ``tag`` - Optional[str] - None * - ``verbose`` - int - 1 * - ``gpu_proportion`` - float - 0.2 * - ``algorithm`` - str - "" * - ``optimizer`` - str - "Avg" * - ``lr`` - float - 1 * - ``betas`` - list - [0, 1] * - ``tau`` - float - 1 * - ``alpha`` - float - 0.01 .. dropdown:: ``FedDCServerConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Server config for the FedDC algorithm.\n\nParameters\n----------\nnum_iters : int\n The number of (outer) iterations.\nnum_clients : int\n The number of clients.\nclients_sample_ratio : float, default 1\n The ratio of clients to participate in each round.\nalpha : float, default 0.01\n The coefficient :math:`\\alpha` of the parameter correction\n (penalized) term :math:`\\frac{\\alpha}{2} \\lVert \\theta_i - (w - h_i) \\rVert^2`.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``log_dir`` : str or pathlib.Path, optional\n The log directory.\n If not specified, will use the default log directory.\n If not absolute, will be relative to the default log directory.\n - ``txt_logger`` : bool, default True\n Whether to use txt logger.\n - ``json_logger`` : bool, default True\n Whether to use json logger.\n - ``eval_every`` : int, default 1\n The number of iterations to evaluate the model.\n - ``visiable_gpus`` : Sequence[int], optional\n Visable GPU IDs for allocating devices for clients.\n Defaults to use all GPUs if available.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training.\n - ``seed`` : int, default 0\n The random seed.\n - ``tag`` : str, optional\n The tag of the experiment.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``gpu_proportion`` : float, default 0.2\n The proportion of clients to use GPU.\n Used to similate the system heterogeneity of the clients.\n Not used in the current version, reserved for future use.", "properties": { "num_iters": { "title": "Num Iters", "type": "integer" }, "num_clients": { "anyOf": [ { "type": "integer" }, { "type": "null" } ], "default": null, "title": "Num Clients" }, "clients_sample_ratio": { "default": 1, "title": "Clients Sample Ratio", "type": "number" }, "log_dir": { "anyOf": [ { "type": "string" }, { "format": "path", "type": "string" }, { "type": "null" } ], "default": null, "title": "Log Dir" }, "txt_logger": { "default": true, "title": "Txt Logger", "type": "boolean" }, "json_logger": { "default": true, "title": "Json Logger", "type": "boolean" }, "eval_every": { "default": 1, "title": "Eval Every", "type": "integer" }, "visiable_gpus": { "anyOf": [ { "items": { "type": "integer" }, "type": "array" }, { "type": "null" } ], "default": null, "title": "Visiable Gpus" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "seed": { "default": 0, "title": "Seed", "type": "integer" }, "tag": { "anyOf": [ { "type": "string" }, { "type": "null" } ], "default": null, "title": "Tag" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "gpu_proportion": { "default": 0.2, "title": "Gpu Proportion", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "optimizer": { "default": "Avg", "title": "Optimizer", "type": "string" }, "lr": { "default": 1, "title": "Lr", "type": "number" }, "betas": { "default": [ 0, 1 ], "maxItems": 2, "minItems": 2, "prefixItems": [ { "type": "number" }, { "type": "number" } ], "title": "Betas", "type": "array" }, "tau": { "default": 1, "title": "Tau", "type": "number" }, "alpha": { "default": 0.01, "title": "Alpha", "type": "number" } }, "required": [ "num_iters" ], "title": "FedDCServerConfig", "type": "object" } **Client config:** ``FedDCClientConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``batch_size`` - int - * * - ``num_epochs`` - int - * * - ``lr`` - float - 0.01 * - ``optimizer`` - str - "SGD" * - ``scheduler`` - Optional[dict] - None * - ``extra_observes`` - list[str] - (factory) * - ``verbose`` - int - 1 * - ``latency`` - float - 0.0 * - ``algorithm`` - str - "" .. dropdown:: ``FedDCClientConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Client config for the FedDC algorithm.\n\nParameters\n----------\nbatch_size : int\n The batch size.\nnum_epochs : int\n The number of (local) epochs.\nlr : float, default 1e-2\n The learning rate :math:`\\eta`.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``scheduler`` : dict, optional\n The scheduler config.\n None for no scheduler, using constant learning rate.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training,\n which would be recorded in evaluated metrics,\n sent to the server, and written to the log file.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``latency`` : float, default 0.0\n The latency of the client.\n Not used in the current version, reserved for future use.", "properties": { "batch_size": { "title": "Batch Size", "type": "integer" }, "num_epochs": { "title": "Num Epochs", "type": "integer" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "optimizer": { "default": "SGD", "title": "Optimizer", "type": "string" }, "scheduler": { "anyOf": [ { "additionalProperties": true, "type": "object" }, { "type": "null" } ], "default": null, "title": "Scheduler" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "latency": { "default": 0.0, "title": "Latency", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" } }, "required": [ "batch_size", "num_epochs" ], "title": "FedDCClientConfig", "type": "object" } .. _config-feddr: FedDR ----- **Server config:** ``FedDRServerConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``num_iters`` - int - * * - ``num_clients`` - Optional[int] - None * - ``clients_sample_ratio`` - float - * * - ``log_dir`` - Optional[str] - None * - ``txt_logger`` - bool - True * - ``json_logger`` - bool - True * - ``eval_every`` - int - 1 * - ``visiable_gpus`` - Optional[list[int]] - None * - ``extra_observes`` - list[str] - (factory) * - ``seed`` - int - 0 * - ``tag`` - Optional[str] - None * - ``verbose`` - int - 1 * - ``gpu_proportion`` - float - 0.2 * - ``algorithm`` - str - "" * - ``eta`` - float - 1.0 * - ``alpha`` - float - 1.9 * - ``reg_type`` - str - "l1_norm" .. dropdown:: ``FedDRServerConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Server config for the FedDR algorithm.\n\nParameters\n----------\nnum_iters : int\n The number of (outer) iterations.\nnum_clients : int\n The number of clients.\nclients_sample_ratio : float\n The ratio of clients to sample for each iteration.\neta : float, default 1.0\n The eta (regularization) parameter for the FedDR algorithm.\nalpha : float, default 1.9\n The alpha parameter for the FedDR algorithm.\nreg_type : str, default \"l1_norm\"\n The type of regularizer to use for the FedDR algorithm.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``log_dir`` : str or pathlib.Path, optional\n The log directory.\n If not specified, will use the default log directory.\n If not absolute, will be relative to the default log directory.\n - ``txt_logger`` : bool, default True\n Whether to use txt logger.\n - ``json_logger`` : bool, default True\n Whether to use json logger.\n - ``eval_every`` : int, default 1\n The number of iterations to evaluate the model.\n - ``visiable_gpus`` : Sequence[int], optional\n Visable GPU IDs for allocating devices for clients.\n Defaults to use all GPUs if available.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training.\n - ``seed`` : int, default 0\n The random seed.\n - ``tag`` : str, optional\n The tag of the experiment.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``gpu_proportion`` : float, default 0.2\n The proportion of clients to use GPU.\n Used to similate the system heterogeneity of the clients.\n Not used in the current version, reserved for future use.", "properties": { "num_iters": { "title": "Num Iters", "type": "integer" }, "num_clients": { "anyOf": [ { "type": "integer" }, { "type": "null" } ], "default": null, "title": "Num Clients" }, "clients_sample_ratio": { "title": "Clients Sample Ratio", "type": "number" }, "log_dir": { "anyOf": [ { "type": "string" }, { "format": "path", "type": "string" }, { "type": "null" } ], "default": null, "title": "Log Dir" }, "txt_logger": { "default": true, "title": "Txt Logger", "type": "boolean" }, "json_logger": { "default": true, "title": "Json Logger", "type": "boolean" }, "eval_every": { "default": 1, "title": "Eval Every", "type": "integer" }, "visiable_gpus": { "anyOf": [ { "items": { "type": "integer" }, "type": "array" }, { "type": "null" } ], "default": null, "title": "Visiable Gpus" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "seed": { "default": 0, "title": "Seed", "type": "integer" }, "tag": { "anyOf": [ { "type": "string" }, { "type": "null" } ], "default": null, "title": "Tag" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "gpu_proportion": { "default": 0.2, "title": "Gpu Proportion", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "eta": { "default": 1.0, "title": "Eta", "type": "number" }, "alpha": { "default": 1.9, "title": "Alpha", "type": "number" }, "reg_type": { "default": "l1_norm", "title": "Reg Type", "type": "string" } }, "required": [ "num_iters", "clients_sample_ratio" ], "title": "FedDRServerConfig", "type": "object" } **Client config:** ``FedDRClientConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``batch_size`` - int - * * - ``num_epochs`` - int - * * - ``lr`` - float - 0.01 * - ``optimizer`` - str - "FedDR" * - ``scheduler`` - Optional[dict] - None * - ``extra_observes`` - list[str] - (factory) * - ``verbose`` - int - 1 * - ``latency`` - float - 0.0 * - ``algorithm`` - str - "" * - ``eta`` - float - 1.0 * - ``alpha`` - float - 1.9 .. dropdown:: ``FedDRClientConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Client config for the FedDR algorithm.\n\nParameters\n----------\nbatch_size : int\n The batch size.\nnum_epochs : int\n The number of epochs.\nlr : float, default 1e-2\n The learning rate.\neta : float, default 1.0\n The eta (regularization) parameter for the FedDR algorithm.\nalpha : float, default 1.9\n The alpha parameter for the FedDR algorithm.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``scheduler`` : dict, optional\n The scheduler config.\n None for no scheduler, using constant learning rate.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training,\n which would be recorded in evaluated metrics,\n sent to the server, and written to the log file.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``latency`` : float, default 0.0\n The latency of the client.\n Not used in the current version, reserved for future use.", "properties": { "batch_size": { "title": "Batch Size", "type": "integer" }, "num_epochs": { "title": "Num Epochs", "type": "integer" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "optimizer": { "default": "FedDR", "title": "Optimizer", "type": "string" }, "scheduler": { "anyOf": [ { "additionalProperties": true, "type": "object" }, { "type": "null" } ], "default": null, "title": "Scheduler" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "latency": { "default": 0.0, "title": "Latency", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "eta": { "default": 1.0, "title": "Eta", "type": "number" }, "alpha": { "default": 1.9, "title": "Alpha", "type": "number" } }, "required": [ "batch_size", "num_epochs" ], "title": "FedDRClientConfig", "type": "object" } .. _config-feddyn: FedDyn ------ **Server config:** ``FedDynServerConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``num_iters`` - int - * * - ``num_clients`` - Optional[int] - None * - ``clients_sample_ratio`` - float - * * - ``log_dir`` - Optional[str] - None * - ``txt_logger`` - bool - True * - ``json_logger`` - bool - True * - ``eval_every`` - int - 1 * - ``visiable_gpus`` - Optional[list[int]] - None * - ``extra_observes`` - list[str] - (factory) * - ``seed`` - int - 0 * - ``tag`` - Optional[str] - None * - ``verbose`` - int - 1 * - ``gpu_proportion`` - float - 0.2 * - ``algorithm`` - str - "" * - ``mu`` - float - 0.1 .. dropdown:: ``FedDynServerConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Server config for the FedDyn algorithm.\n\nParameters\n----------\nnum_iters : int\n The number of (outer) iterations.\nnum_clients : int\n The number of clients.\nclients_sample_ratio : float\n The ratio of clients to sample for each iteration.\nmu : float, default 1 / 10\n The coefficient of the \"proximal\" term.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``log_dir`` : str or pathlib.Path, optional\n The log directory.\n If not specified, will use the default log directory.\n If not absolute, will be relative to the default log directory.\n - ``txt_logger`` : bool, default True\n Whether to use txt logger.\n - ``json_logger`` : bool, default True\n Whether to use json logger.\n - ``eval_every`` : int, default 1\n The number of iterations to evaluate the model.\n - ``visiable_gpus`` : Sequence[int], optional\n Visable GPU IDs for allocating devices for clients.\n Defaults to use all GPUs if available.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training.\n - ``seed`` : int, default 0\n The random seed.\n - ``tag`` : str, optional\n The tag of the experiment.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``gpu_proportion`` : float, default 0.2\n The proportion of clients to use GPU.\n Used to similate the system heterogeneity of the clients.\n Not used in the current version, reserved for future use.", "properties": { "num_iters": { "title": "Num Iters", "type": "integer" }, "num_clients": { "anyOf": [ { "type": "integer" }, { "type": "null" } ], "default": null, "title": "Num Clients" }, "clients_sample_ratio": { "title": "Clients Sample Ratio", "type": "number" }, "log_dir": { "anyOf": [ { "type": "string" }, { "format": "path", "type": "string" }, { "type": "null" } ], "default": null, "title": "Log Dir" }, "txt_logger": { "default": true, "title": "Txt Logger", "type": "boolean" }, "json_logger": { "default": true, "title": "Json Logger", "type": "boolean" }, "eval_every": { "default": 1, "title": "Eval Every", "type": "integer" }, "visiable_gpus": { "anyOf": [ { "items": { "type": "integer" }, "type": "array" }, { "type": "null" } ], "default": null, "title": "Visiable Gpus" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "seed": { "default": 0, "title": "Seed", "type": "integer" }, "tag": { "anyOf": [ { "type": "string" }, { "type": "null" } ], "default": null, "title": "Tag" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "gpu_proportion": { "default": 0.2, "title": "Gpu Proportion", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "mu": { "default": 0.1, "title": "Mu", "type": "number" } }, "required": [ "num_iters", "clients_sample_ratio" ], "title": "FedDynServerConfig", "type": "object" } **Client config:** ``FedDynClientConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``batch_size`` - int - * * - ``num_epochs`` - int - * * - ``lr`` - float - 0.01 * - ``optimizer`` - str - "" * - ``scheduler`` - Optional[dict] - None * - ``extra_observes`` - list[str] - (factory) * - ``verbose`` - int - 1 * - ``latency`` - float - 0.0 * - ``algorithm`` - str - "" .. dropdown:: ``FedDynClientConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Client config for the FedDyn algorithm.\n\nParameters\n----------\nbatch_size : int\n The batch size.\nnum_epochs : int\n The number of epochs.\nlr : float, default 1e-2\n The learning rate.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``scheduler`` : dict, optional\n The scheduler config.\n None for no scheduler, using constant learning rate.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training,\n which would be recorded in evaluated metrics,\n sent to the server, and written to the log file.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``latency`` : float, default 0.0\n The latency of the client.\n Not used in the current version, reserved for future use.", "properties": { "batch_size": { "title": "Batch Size", "type": "integer" }, "num_epochs": { "title": "Num Epochs", "type": "integer" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "optimizer": { "default": "", "title": "Optimizer", "type": "string" }, "scheduler": { "anyOf": [ { "additionalProperties": true, "type": "object" }, { "type": "null" } ], "default": null, "title": "Scheduler" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "latency": { "default": 0.0, "title": "Latency", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" } }, "required": [ "batch_size", "num_epochs" ], "title": "FedDynClientConfig", "type": "object" } .. _config-fedopt: FedOpt ------ **Server config:** ``FedOptServerConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``num_iters`` - int - * * - ``num_clients`` - Optional[int] - None * - ``clients_sample_ratio`` - float - * * - ``log_dir`` - Optional[str] - None * - ``txt_logger`` - bool - True * - ``json_logger`` - bool - True * - ``eval_every`` - int - 1 * - ``visiable_gpus`` - Optional[list[int]] - None * - ``extra_observes`` - list[str] - (factory) * - ``seed`` - int - 0 * - ``tag`` - Optional[str] - None * - ``verbose`` - int - 1 * - ``gpu_proportion`` - float - 0.2 * - ``algorithm`` - str - "" * - ``optimizer`` - str - "Adam" * - ``lr`` - float - 0.01 * - ``betas`` - list - [0.9, 0.99] * - ``tau`` - float - 1e-05 .. dropdown:: ``FedOptServerConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Server config for the FedOpt algorithm.\n\nParameters\n----------\nnum_iters : int\n The number of (outer) iterations.\nnum_clients : int\n The number of clients.\nclients_sample_ratio : float\n The ratio of clients to sample for each iteration.\noptimizer : {\"SGD\", \"Adam\", \"Adagrad\", \"Yogi\"}, default \"Adam\"\n The optimizer to use, case insensitive.\nlr : float, default 1e-2\n The learning rate.\nbetas : Sequence[float], default (0.9, 0.99)\n The beta values for the optimizer.\ntau : float, default 1e-5\n The tau value for the optimizer,\n which controls the degree of adaptivity of the algorithm.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``log_dir`` : str or pathlib.Path, optional\n The log directory.\n If not specified, will use the default log directory.\n If not absolute, will be relative to the default log directory.\n - ``txt_logger`` : bool, default True\n Whether to use txt logger.\n - ``json_logger`` : bool, default True\n Whether to use json logger.\n - ``eval_every`` : int, default 1\n The number of iterations to evaluate the model.\n - ``visiable_gpus`` : Sequence[int], optional\n Visable GPU IDs for allocating devices for clients.\n Defaults to use all GPUs if available.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training.\n - ``seed`` : int, default 0\n The random seed.\n - ``tag`` : str, optional\n The tag of the experiment.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``gpu_proportion`` : float, default 0.2\n The proportion of clients to use GPU.\n Used to similate the system heterogeneity of the clients.\n Not used in the current version, reserved for future use.", "properties": { "num_iters": { "title": "Num Iters", "type": "integer" }, "num_clients": { "anyOf": [ { "type": "integer" }, { "type": "null" } ], "default": null, "title": "Num Clients" }, "clients_sample_ratio": { "title": "Clients Sample Ratio", "type": "number" }, "log_dir": { "anyOf": [ { "type": "string" }, { "format": "path", "type": "string" }, { "type": "null" } ], "default": null, "title": "Log Dir" }, "txt_logger": { "default": true, "title": "Txt Logger", "type": "boolean" }, "json_logger": { "default": true, "title": "Json Logger", "type": "boolean" }, "eval_every": { "default": 1, "title": "Eval Every", "type": "integer" }, "visiable_gpus": { "anyOf": [ { "items": { "type": "integer" }, "type": "array" }, { "type": "null" } ], "default": null, "title": "Visiable Gpus" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "seed": { "default": 0, "title": "Seed", "type": "integer" }, "tag": { "anyOf": [ { "type": "string" }, { "type": "null" } ], "default": null, "title": "Tag" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "gpu_proportion": { "default": 0.2, "title": "Gpu Proportion", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "optimizer": { "default": "Adam", "title": "Optimizer", "type": "string" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "betas": { "default": [ 0.9, 0.99 ], "maxItems": 2, "minItems": 2, "prefixItems": [ { "type": "number" }, { "type": "number" } ], "title": "Betas", "type": "array" }, "tau": { "default": 1e-05, "title": "Tau", "type": "number" } }, "required": [ "num_iters", "clients_sample_ratio" ], "title": "FedOptServerConfig", "type": "object" } **Client config:** ``FedOptClientConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``batch_size`` - int - * * - ``num_epochs`` - int - * * - ``lr`` - float - 0.01 * - ``optimizer`` - str - "SGD" * - ``scheduler`` - Optional[dict] - None * - ``extra_observes`` - list[str] - (factory) * - ``verbose`` - int - 1 * - ``latency`` - float - 0.0 * - ``algorithm`` - str - "" .. dropdown:: ``FedOptClientConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Client config for the FedOpt algorithm.\n\nParameters\n----------\nbatch_size : int\n The batch size.\nnum_epochs : int\n The number of epochs.\nlr : float, default 1e-2\n The learning rate.\noptimizer : str, default \"SGD\"\n The name of the optimizer to solve the local (inner) problem.\n**kwargs : dict, optional\n Additional keyword arguments for specific algorithms\n (FedAvg, FedAdagrad, FedYogi, FedAdam). And\n\n - ``scheduler`` : dict, optional\n The scheduler config.\n None for no scheduler, using constant learning rate.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training,\n which would be recorded in evaluated metrics,\n sent to the server, and written to the log file.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``latency`` : float, default 0.0\n The latency of the client.\n Not used in the current version, reserved for future use.", "properties": { "batch_size": { "title": "Batch Size", "type": "integer" }, "num_epochs": { "title": "Num Epochs", "type": "integer" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "optimizer": { "default": "SGD", "title": "Optimizer", "type": "string" }, "scheduler": { "anyOf": [ { "additionalProperties": true, "type": "object" }, { "type": "null" } ], "default": null, "title": "Scheduler" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "latency": { "default": 0.0, "title": "Latency", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" } }, "required": [ "batch_size", "num_epochs" ], "title": "FedOptClientConfig", "type": "object" } .. _config-fedpd: FedPD ----- **Server config:** ``FedPDServerConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``num_iters`` - int - * * - ``num_clients`` - Optional[int] - None * - ``clients_sample_ratio`` - float - 1 * - ``log_dir`` - Optional[str] - None * - ``txt_logger`` - bool - True * - ``json_logger`` - bool - True * - ``eval_every`` - int - 1 * - ``visiable_gpus`` - Optional[list[int]] - None * - ``extra_observes`` - list[str] - (factory) * - ``seed`` - int - 0 * - ``tag`` - Optional[str] - None * - ``verbose`` - int - 1 * - ``gpu_proportion`` - float - 0.2 * - ``algorithm`` - str - "" * - ``p`` - float - * * - ``stochastic`` - bool - False * - ``vr`` - bool - False .. dropdown:: ``FedPDServerConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Server config for the FedPD algorithm.\n\nParameters\n----------\nnum_iters : int\n The number of (outer) iterations.\nnum_clients : int\n The number of clients.\np : float\n Probability of skipping communication.\nstochastic : bool, default False\n Skip communication in a stochastic manner or not.\nvr : bool, default False\n Whether to use variance reduction.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``log_dir`` : str or pathlib.Path, optional\n The log directory.\n If not specified, will use the default log directory.\n If not absolute, will be relative to the default log directory.\n - ``txt_logger`` : bool, default True\n Whether to use txt logger.\n - ``json_logger`` : bool, default True\n Whether to use json logger.\n - ``eval_every`` : int, default 1\n The number of iterations to evaluate the model.\n - ``visiable_gpus`` : Sequence[int], optional\n Visable GPU IDs for allocating devices for clients.\n Defaults to use all GPUs if available.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training.\n - ``seed`` : int, default 0\n The random seed.\n - ``tag`` : str, optional\n The tag of the experiment.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``gpu_proportion`` : float, default 0.2\n The proportion of clients to use GPU.\n Used to similate the system heterogeneity of the clients.\n Not used in the current version, reserved for future use.", "properties": { "num_iters": { "title": "Num Iters", "type": "integer" }, "num_clients": { "anyOf": [ { "type": "integer" }, { "type": "null" } ], "default": null, "title": "Num Clients" }, "clients_sample_ratio": { "default": 1, "title": "Clients Sample Ratio", "type": "number" }, "log_dir": { "anyOf": [ { "type": "string" }, { "format": "path", "type": "string" }, { "type": "null" } ], "default": null, "title": "Log Dir" }, "txt_logger": { "default": true, "title": "Txt Logger", "type": "boolean" }, "json_logger": { "default": true, "title": "Json Logger", "type": "boolean" }, "eval_every": { "default": 1, "title": "Eval Every", "type": "integer" }, "visiable_gpus": { "anyOf": [ { "items": { "type": "integer" }, "type": "array" }, { "type": "null" } ], "default": null, "title": "Visiable Gpus" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "seed": { "default": 0, "title": "Seed", "type": "integer" }, "tag": { "anyOf": [ { "type": "string" }, { "type": "null" } ], "default": null, "title": "Tag" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "gpu_proportion": { "default": 0.2, "title": "Gpu Proportion", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "p": { "title": "P", "type": "number" }, "stochastic": { "default": false, "title": "Stochastic", "type": "boolean" }, "vr": { "default": false, "title": "Vr", "type": "boolean" } }, "required": [ "num_iters", "p" ], "title": "FedPDServerConfig", "type": "object" } **Client config:** ``FedPDClientConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``batch_size`` - int - * * - ``num_epochs`` - int - * * - ``lr`` - float - 0.01 * - ``optimizer`` - str - "" * - ``scheduler`` - Optional[dict] - None * - ``extra_observes`` - list[str] - (factory) * - ``verbose`` - int - 1 * - ``latency`` - float - 0.0 * - ``algorithm`` - str - "" * - ``mu`` - float - 0.1 * - ``vr`` - bool - False * - ``dual_rand_init`` - bool - False .. dropdown:: ``FedPDClientConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Client config for the FedPD algorithm.\n\nParameters\n----------\nbatch_size : int\n The batch size.\nnum_epochs : int\n The number of epochs.\nlr : float, default 1e-2\n The learning rate.\nmu : float, default 1 / 10\n The coefficient of the proximal term.\nvr : bool, default False\n Whether to use variance reduction.\ndual_rand_init : bool, default False\n Whether to use random initialization for dual variables.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``scheduler`` : dict, optional\n The scheduler config.\n None for no scheduler, using constant learning rate.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training,\n which would be recorded in evaluated metrics,\n sent to the server, and written to the log file.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``latency`` : float, default 0.0\n The latency of the client.\n Not used in the current version, reserved for future use.", "properties": { "batch_size": { "title": "Batch Size", "type": "integer" }, "num_epochs": { "title": "Num Epochs", "type": "integer" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "optimizer": { "default": "", "title": "Optimizer", "type": "string" }, "scheduler": { "anyOf": [ { "additionalProperties": true, "type": "object" }, { "type": "null" } ], "default": null, "title": "Scheduler" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "latency": { "default": 0.0, "title": "Latency", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "mu": { "default": 0.1, "title": "Mu", "type": "number" }, "vr": { "default": false, "title": "Vr", "type": "boolean" }, "dual_rand_init": { "default": false, "title": "Dual Rand Init", "type": "boolean" } }, "required": [ "batch_size", "num_epochs" ], "title": "FedPDClientConfig", "type": "object" } .. _config-fedprox: FedProx ------- **Server config:** ``FedProxServerConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``num_iters`` - int - * * - ``num_clients`` - Optional[int] - None * - ``clients_sample_ratio`` - float - * * - ``log_dir`` - Optional[str] - None * - ``txt_logger`` - bool - True * - ``json_logger`` - bool - True * - ``eval_every`` - int - 1 * - ``visiable_gpus`` - Optional[list[int]] - None * - ``extra_observes`` - list[str] - (factory) * - ``seed`` - int - 0 * - ``tag`` - Optional[str] - None * - ``verbose`` - int - 1 * - ``gpu_proportion`` - float - 0.2 * - ``algorithm`` - str - "" * - ``vr`` - bool - False .. dropdown:: ``FedProxServerConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Server config for the FedProx algorithm.\n\nParameters\n----------\nnum_iters : int\n The number of (outer) iterations.\nnum_clients : int\n The number of clients.\nclients_sample_ratio : float\n The ratio of clients to sample for each iteration.\nvr : bool, default False\n Whether to use variance reduction.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``log_dir`` : str or pathlib.Path, optional\n The log directory.\n If not specified, will use the default log directory.\n If not absolute, will be relative to the default log directory.\n - ``txt_logger`` : bool, default True\n Whether to use txt logger.\n - ``json_logger`` : bool, default True\n Whether to use json logger.\n - ``eval_every`` : int, default 1\n The number of iterations to evaluate the model.\n - ``visiable_gpus`` : Sequence[int], optional\n Visable GPU IDs for allocating devices for clients.\n Defaults to use all GPUs if available.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training.\n - ``seed`` : int, default 0\n The random seed.\n - ``tag`` : str, optional\n The tag of the experiment.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``gpu_proportion`` : float, default 0.2\n The proportion of clients to use GPU.\n Used to similate the system heterogeneity of the clients.\n Not used in the current version, reserved for future use.", "properties": { "num_iters": { "title": "Num Iters", "type": "integer" }, "num_clients": { "anyOf": [ { "type": "integer" }, { "type": "null" } ], "default": null, "title": "Num Clients" }, "clients_sample_ratio": { "title": "Clients Sample Ratio", "type": "number" }, "log_dir": { "anyOf": [ { "type": "string" }, { "format": "path", "type": "string" }, { "type": "null" } ], "default": null, "title": "Log Dir" }, "txt_logger": { "default": true, "title": "Txt Logger", "type": "boolean" }, "json_logger": { "default": true, "title": "Json Logger", "type": "boolean" }, "eval_every": { "default": 1, "title": "Eval Every", "type": "integer" }, "visiable_gpus": { "anyOf": [ { "items": { "type": "integer" }, "type": "array" }, { "type": "null" } ], "default": null, "title": "Visiable Gpus" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "seed": { "default": 0, "title": "Seed", "type": "integer" }, "tag": { "anyOf": [ { "type": "string" }, { "type": "null" } ], "default": null, "title": "Tag" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "gpu_proportion": { "default": 0.2, "title": "Gpu Proportion", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "vr": { "default": false, "title": "Vr", "type": "boolean" } }, "required": [ "num_iters", "clients_sample_ratio" ], "title": "FedProxServerConfig", "type": "object" } **Client config:** ``FedProxClientConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``batch_size`` - int - * * - ``num_epochs`` - int - * * - ``lr`` - float - 0.01 * - ``optimizer`` - str - "" * - ``scheduler`` - Optional[dict] - None * - ``extra_observes`` - list[str] - (factory) * - ``verbose`` - int - 1 * - ``latency`` - float - 0.0 * - ``algorithm`` - str - "" * - ``mu`` - float - 0.01 * - ``vr`` - bool - False .. dropdown:: ``FedProxClientConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Client config for the FedProx algorithm.\n\nParameters\n----------\nbatch_size : int\n The batch size.\nnum_epochs : int\n The number of epochs.\nlr : float, default 1e-2\n The learning rate.\nmu : float, default 0.01\n Coefficient for the proximal term.\nvr : bool, default False\n Whether to use variance reduction.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``scheduler`` : dict, optional\n The scheduler config.\n None for no scheduler, using constant learning rate.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training,\n which would be recorded in evaluated metrics,\n sent to the server, and written to the log file.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``latency`` : float, default 0.0\n The latency of the client.\n Not used in the current version, reserved for future use.", "properties": { "batch_size": { "title": "Batch Size", "type": "integer" }, "num_epochs": { "title": "Num Epochs", "type": "integer" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "optimizer": { "default": "", "title": "Optimizer", "type": "string" }, "scheduler": { "anyOf": [ { "additionalProperties": true, "type": "object" }, { "type": "null" } ], "default": null, "title": "Scheduler" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "latency": { "default": 0.0, "title": "Latency", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "mu": { "default": 0.01, "title": "Mu", "type": "number" }, "vr": { "default": false, "title": "Vr", "type": "boolean" } }, "required": [ "batch_size", "num_epochs" ], "title": "FedProxClientConfig", "type": "object" } .. _config-fedsplit: FedSplit -------- **Server config:** ``FedSplitServerConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``num_iters`` - int - * * - ``num_clients`` - Optional[int] - None * - ``clients_sample_ratio`` - float - * * - ``log_dir`` - Optional[str] - None * - ``txt_logger`` - bool - True * - ``json_logger`` - bool - True * - ``eval_every`` - int - 1 * - ``visiable_gpus`` - Optional[list[int]] - None * - ``extra_observes`` - list[str] - (factory) * - ``seed`` - int - 0 * - ``tag`` - Optional[str] - None * - ``verbose`` - int - 1 * - ``gpu_proportion`` - float - 0.2 * - ``algorithm`` - str - "" .. dropdown:: ``FedSplitServerConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Server config for the FedSplit algorithm.\n\nParameters\n----------\nnum_iters : int\n The number of (outer) iterations.\nnum_clients : int\n The number of clients.\nclients_sample_ratio : float\n The ratio of clients to sample for each iteration.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``log_dir`` : str or pathlib.Path, optional\n The log directory.\n If not specified, will use the default log directory.\n If not absolute, will be relative to the default log directory.\n - ``txt_logger`` : bool, default True\n Whether to use txt logger.\n - ``json_logger`` : bool, default True\n Whether to use json logger.\n - ``eval_every`` : int, default 1\n The number of iterations to evaluate the model.\n - ``visiable_gpus`` : Sequence[int], optional\n Visable GPU IDs for allocating devices for clients.\n Defaults to use all GPUs if available.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training.\n - ``seed`` : int, default 0\n The random seed.\n - ``tag`` : str, optional\n The tag of the experiment.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``gpu_proportion`` : float, default 0.2\n The proportion of clients to use GPU.\n Used to similate the system heterogeneity of the clients.\n Not used in the current version, reserved for future use.", "properties": { "num_iters": { "title": "Num Iters", "type": "integer" }, "num_clients": { "anyOf": [ { "type": "integer" }, { "type": "null" } ], "default": null, "title": "Num Clients" }, "clients_sample_ratio": { "title": "Clients Sample Ratio", "type": "number" }, "log_dir": { "anyOf": [ { "type": "string" }, { "format": "path", "type": "string" }, { "type": "null" } ], "default": null, "title": "Log Dir" }, "txt_logger": { "default": true, "title": "Txt Logger", "type": "boolean" }, "json_logger": { "default": true, "title": "Json Logger", "type": "boolean" }, "eval_every": { "default": 1, "title": "Eval Every", "type": "integer" }, "visiable_gpus": { "anyOf": [ { "items": { "type": "integer" }, "type": "array" }, { "type": "null" } ], "default": null, "title": "Visiable Gpus" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "seed": { "default": 0, "title": "Seed", "type": "integer" }, "tag": { "anyOf": [ { "type": "string" }, { "type": "null" } ], "default": null, "title": "Tag" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "gpu_proportion": { "default": 0.2, "title": "Gpu Proportion", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" } }, "required": [ "num_iters", "clients_sample_ratio" ], "title": "FedSplitServerConfig", "type": "object" } **Client config:** ``FedSplitClientConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``batch_size`` - int - * * - ``num_epochs`` - int - * * - ``lr`` - float - 0.01 * - ``optimizer`` - str - "ProxSGD" * - ``scheduler`` - Optional[dict] - None * - ``extra_observes`` - list[str] - (factory) * - ``verbose`` - int - 1 * - ``latency`` - float - 0.0 * - ``algorithm`` - str - "" * - ``s`` - float - 10.0 .. dropdown:: ``FedSplitClientConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Client config for the FedSplit algorithm.\n\nParameters\n----------\nbatch_size : int\n The batch size.\nnum_epochs : int\n The number of epochs.\nlr : float, default 1e-2\n The learning rate.\ns : float, default 10.0\n Reciprocal of the proximal parameter.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``scheduler`` : dict, optional\n The scheduler config.\n None for no scheduler, using constant learning rate.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training,\n which would be recorded in evaluated metrics,\n sent to the server, and written to the log file.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``latency`` : float, default 0.0\n The latency of the client.\n Not used in the current version, reserved for future use.", "properties": { "batch_size": { "title": "Batch Size", "type": "integer" }, "num_epochs": { "title": "Num Epochs", "type": "integer" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "optimizer": { "default": "ProxSGD", "title": "Optimizer", "type": "string" }, "scheduler": { "anyOf": [ { "additionalProperties": true, "type": "object" }, { "type": "null" } ], "default": null, "title": "Scheduler" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "latency": { "default": 0.0, "title": "Latency", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "s": { "default": 10.0, "title": "S", "type": "number" } }, "required": [ "batch_size", "num_epochs" ], "title": "FedSplitClientConfig", "type": "object" } .. _config-fedyogi: FedYogi ------- **Server config:** ``FedYogiServerConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``num_iters`` - int - * * - ``num_clients`` - Optional[int] - None * - ``clients_sample_ratio`` - float - * * - ``log_dir`` - Optional[str] - None * - ``txt_logger`` - bool - True * - ``json_logger`` - bool - True * - ``eval_every`` - int - 1 * - ``visiable_gpus`` - Optional[list[int]] - None * - ``extra_observes`` - list[str] - (factory) * - ``seed`` - int - 0 * - ``tag`` - Optional[str] - None * - ``verbose`` - int - 1 * - ``gpu_proportion`` - float - 0.2 * - ``algorithm`` - str - "" * - ``optimizer`` - str - "Yogi" * - ``lr`` - float - 0.01 * - ``betas`` - list - [0.9, 0.99] * - ``tau`` - float - 1e-05 .. dropdown:: ``FedYogiServerConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Server config for the FedOpt algorithm.\n\nParameters\n----------\nnum_iters : int\n The number of (outer) iterations.\nnum_clients : int\n The number of clients.\nclients_sample_ratio : float\n The ratio of clients to sample for each iteration.\nlr : float, default 1e-2\n The learning rate.\nbetas : Sequence[float], default (0.9, 0.99)\n The beta values for the optimizer.\ntau : float, default 1e-5\n The tau value for the optimizer,\n which controls the degree of adaptivity of the algorithm.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``log_dir`` : str or pathlib.Path, optional\n The log directory.\n If not specified, will use the default log directory.\n If not absolute, will be relative to the default log directory.\n - ``txt_logger`` : bool, default True\n Whether to use txt logger.\n - ``json_logger`` : bool, default True\n Whether to use json logger.\n - ``eval_every`` : int, default 1\n The number of iterations to evaluate the model.\n - ``visiable_gpus`` : Sequence[int], optional\n Visable GPU IDs for allocating devices for clients.\n Defaults to use all GPUs if available.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training.\n - ``seed`` : int, default 0\n The random seed.\n - ``tag`` : str, optional\n The tag of the experiment.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``gpu_proportion`` : float, default 0.2\n The proportion of clients to use GPU.\n Used to similate the system heterogeneity of the clients.\n Not used in the current version, reserved for future use.", "properties": { "num_iters": { "title": "Num Iters", "type": "integer" }, "num_clients": { "anyOf": [ { "type": "integer" }, { "type": "null" } ], "default": null, "title": "Num Clients" }, "clients_sample_ratio": { "title": "Clients Sample Ratio", "type": "number" }, "log_dir": { "anyOf": [ { "type": "string" }, { "format": "path", "type": "string" }, { "type": "null" } ], "default": null, "title": "Log Dir" }, "txt_logger": { "default": true, "title": "Txt Logger", "type": "boolean" }, "json_logger": { "default": true, "title": "Json Logger", "type": "boolean" }, "eval_every": { "default": 1, "title": "Eval Every", "type": "integer" }, "visiable_gpus": { "anyOf": [ { "items": { "type": "integer" }, "type": "array" }, { "type": "null" } ], "default": null, "title": "Visiable Gpus" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "seed": { "default": 0, "title": "Seed", "type": "integer" }, "tag": { "anyOf": [ { "type": "string" }, { "type": "null" } ], "default": null, "title": "Tag" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "gpu_proportion": { "default": 0.2, "title": "Gpu Proportion", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "optimizer": { "default": "Yogi", "title": "Optimizer", "type": "string" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "betas": { "default": [ 0.9, 0.99 ], "maxItems": 2, "minItems": 2, "prefixItems": [ { "type": "number" }, { "type": "number" } ], "title": "Betas", "type": "array" }, "tau": { "default": 1e-05, "title": "Tau", "type": "number" } }, "required": [ "num_iters", "clients_sample_ratio" ], "title": "FedYogiServerConfig", "type": "object" } **Client config:** ``FedYogiClientConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``batch_size`` - int - * * - ``num_epochs`` - int - * * - ``lr`` - float - 0.01 * - ``optimizer`` - str - "SGD" * - ``scheduler`` - Optional[dict] - None * - ``extra_observes`` - list[str] - (factory) * - ``verbose`` - int - 1 * - ``latency`` - float - 0.0 * - ``algorithm`` - str - "" .. dropdown:: ``FedYogiClientConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Client config for the FedYogi algorithm.\n\nParameters\n----------\nbatch_size : int\n The batch size.\nnum_epochs : int\n The number of epochs.\nlr : float, default 1e-2\n The learning rate.\noptimizer : str, default \"SGD\"\n The name of the optimizer to solve the local (inner) problem.\n**kwargs : dict, optional\n Additional keyword arguments for specific algorithms\n (FedAvg, FedAdagrad, FedYogi, FedAdam). And\n\n - ``scheduler`` : dict, optional\n The scheduler config.\n None for no scheduler, using constant learning rate.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training,\n which would be recorded in evaluated metrics,\n sent to the server, and written to the log file.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``latency`` : float, default 0.0\n The latency of the client.\n Not used in the current version, reserved for future use.", "properties": { "batch_size": { "title": "Batch Size", "type": "integer" }, "num_epochs": { "title": "Num Epochs", "type": "integer" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "optimizer": { "default": "SGD", "title": "Optimizer", "type": "string" }, "scheduler": { "anyOf": [ { "additionalProperties": true, "type": "object" }, { "type": "null" } ], "default": null, "title": "Scheduler" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "latency": { "default": 0.0, "title": "Latency", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" } }, "required": [ "batch_size", "num_epochs" ], "title": "FedYogiClientConfig", "type": "object" } .. _config-ifca: IFCA ---- **Server config:** ``IFCAServerConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``num_iters`` - int - * * - ``num_clients`` - Optional[int] - None * - ``clients_sample_ratio`` - float - 1 * - ``log_dir`` - Optional[str] - None * - ``txt_logger`` - bool - True * - ``json_logger`` - bool - True * - ``eval_every`` - int - 1 * - ``visiable_gpus`` - Optional[list[int]] - None * - ``extra_observes`` - list[str] - (factory) * - ``seed`` - int - 0 * - ``tag`` - Optional[str] - None * - ``verbose`` - int - 1 * - ``gpu_proportion`` - float - 0.2 * - ``algorithm`` - str - "" * - ``optimizer`` - str - "Avg" * - ``lr`` - float - 1 * - ``betas`` - list - [0, 1] * - ``tau`` - float - 1 * - ``num_clusters`` - int - * .. dropdown:: ``IFCAServerConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Server config for the IFCA algorithm.\n\nParameters\n----------\nnum_clusters : int\n The number of clusters.\nnum_iters : int\n The number of (outer) iterations.\nnum_clients : int\n The number of clients.\nclients_sample_ratio : float, default 1\n The ratio of clients to participate in each round.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``log_dir`` : str or pathlib.Path, optional\n The log directory.\n If not specified, will use the default log directory.\n If not absolute, will be relative to the default log directory.\n - ``txt_logger`` : bool, default True\n Whether to use txt logger.\n - ``json_logger`` : bool, default True\n Whether to use json logger.\n - ``eval_every`` : int, default 1\n The number of iterations to evaluate the model.\n - ``visiable_gpus`` : Sequence[int], optional\n Visable GPU IDs for allocating devices for clients.\n Defaults to use all GPUs if available.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training.\n - ``seed`` : int, default 0\n The random seed.\n - ``tag`` : str, optional\n The tag of the experiment.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``gpu_proportion`` : float, default 0.2\n The proportion of clients to use GPU.\n Used to similate the system heterogeneity of the clients.\n Not used in the current version, reserved for future use.", "properties": { "num_iters": { "title": "Num Iters", "type": "integer" }, "num_clients": { "anyOf": [ { "type": "integer" }, { "type": "null" } ], "default": null, "title": "Num Clients" }, "clients_sample_ratio": { "default": 1, "title": "Clients Sample Ratio", "type": "number" }, "log_dir": { "anyOf": [ { "type": "string" }, { "format": "path", "type": "string" }, { "type": "null" } ], "default": null, "title": "Log Dir" }, "txt_logger": { "default": true, "title": "Txt Logger", "type": "boolean" }, "json_logger": { "default": true, "title": "Json Logger", "type": "boolean" }, "eval_every": { "default": 1, "title": "Eval Every", "type": "integer" }, "visiable_gpus": { "anyOf": [ { "items": { "type": "integer" }, "type": "array" }, { "type": "null" } ], "default": null, "title": "Visiable Gpus" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "seed": { "default": 0, "title": "Seed", "type": "integer" }, "tag": { "anyOf": [ { "type": "string" }, { "type": "null" } ], "default": null, "title": "Tag" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "gpu_proportion": { "default": 0.2, "title": "Gpu Proportion", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "optimizer": { "default": "Avg", "title": "Optimizer", "type": "string" }, "lr": { "default": 1, "title": "Lr", "type": "number" }, "betas": { "default": [ 0, 1 ], "maxItems": 2, "minItems": 2, "prefixItems": [ { "type": "number" }, { "type": "number" } ], "title": "Betas", "type": "array" }, "tau": { "default": 1, "title": "Tau", "type": "number" }, "num_clusters": { "title": "Num Clusters", "type": "integer" } }, "required": [ "num_iters", "num_clusters" ], "title": "IFCAServerConfig", "type": "object" } **Client config:** ``IFCAClientConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``batch_size`` - int - * * - ``num_epochs`` - int - * * - ``lr`` - float - 0.01 * - ``optimizer`` - str - "SGD" * - ``scheduler`` - Optional[dict] - None * - ``extra_observes`` - list[str] - (factory) * - ``verbose`` - int - 1 * - ``latency`` - float - 0.0 * - ``algorithm`` - str - "" .. dropdown:: ``IFCAClientConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Client config for the IFCA algorithm.\n\nParameters\n----------\nbatch_size : int\n The batch size.\nnum_epochs : int\n The number of epochs.\nlr : float, default 1e-2\n The learning rate.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``scheduler`` : dict, optional\n The scheduler config.\n None for no scheduler, using constant learning rate.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training,\n which would be recorded in evaluated metrics,\n sent to the server, and written to the log file.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``latency`` : float, default 0.0\n The latency of the client.\n Not used in the current version, reserved for future use.", "properties": { "batch_size": { "title": "Batch Size", "type": "integer" }, "num_epochs": { "title": "Num Epochs", "type": "integer" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "optimizer": { "default": "SGD", "title": "Optimizer", "type": "string" }, "scheduler": { "anyOf": [ { "additionalProperties": true, "type": "object" }, { "type": "null" } ], "default": null, "title": "Scheduler" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "latency": { "default": 0.0, "title": "Latency", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" } }, "required": [ "batch_size", "num_epochs" ], "title": "IFCAClientConfig", "type": "object" } .. _config-proxskip: ProxSkip -------- **Server config:** ``ProxSkipServerConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``num_iters`` - int - * * - ``num_clients`` - Optional[int] - None * - ``clients_sample_ratio`` - float - 1 * - ``log_dir`` - Optional[str] - None * - ``txt_logger`` - bool - True * - ``json_logger`` - bool - True * - ``eval_every`` - int - 1 * - ``visiable_gpus`` - Optional[list[int]] - None * - ``extra_observes`` - list[str] - (factory) * - ``seed`` - int - 0 * - ``tag`` - Optional[str] - None * - ``verbose`` - int - 1 * - ``gpu_proportion`` - float - 0.2 * - ``algorithm`` - str - "" * - ``p`` - float - * * - ``vr`` - bool - False .. dropdown:: ``ProxSkipServerConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Server config for the ProxSkip algorithm.\n\nParameters\n----------\nnum_iters : int\n The number of (outer) iterations.\nnum_clients : int\n The number of clients.\np : float\n Probability of skipping communication.\nvr : bool, default False\n Whether to use variance reduction.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``log_dir`` : str or pathlib.Path, optional\n The log directory.\n If not specified, will use the default log directory.\n If not absolute, will be relative to the default log directory.\n - ``txt_logger`` : bool, default True\n Whether to use txt logger.\n - ``json_logger`` : bool, default True\n Whether to use json logger.\n - ``eval_every`` : int, default 1\n The number of iterations to evaluate the model.\n - ``visiable_gpus`` : Sequence[int], optional\n Visable GPU IDs for allocating devices for clients.\n Defaults to use all GPUs if available.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training.\n - ``seed`` : int, default 0\n The random seed.\n - ``tag`` : str, optional\n The tag of the experiment.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``gpu_proportion`` : float, default 0.2\n The proportion of clients to use GPU.\n Used to similate the system heterogeneity of the clients.\n Not used in the current version, reserved for future use.", "properties": { "num_iters": { "title": "Num Iters", "type": "integer" }, "num_clients": { "anyOf": [ { "type": "integer" }, { "type": "null" } ], "default": null, "title": "Num Clients" }, "clients_sample_ratio": { "default": 1, "title": "Clients Sample Ratio", "type": "number" }, "log_dir": { "anyOf": [ { "type": "string" }, { "format": "path", "type": "string" }, { "type": "null" } ], "default": null, "title": "Log Dir" }, "txt_logger": { "default": true, "title": "Txt Logger", "type": "boolean" }, "json_logger": { "default": true, "title": "Json Logger", "type": "boolean" }, "eval_every": { "default": 1, "title": "Eval Every", "type": "integer" }, "visiable_gpus": { "anyOf": [ { "items": { "type": "integer" }, "type": "array" }, { "type": "null" } ], "default": null, "title": "Visiable Gpus" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "seed": { "default": 0, "title": "Seed", "type": "integer" }, "tag": { "anyOf": [ { "type": "string" }, { "type": "null" } ], "default": null, "title": "Tag" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "gpu_proportion": { "default": 0.2, "title": "Gpu Proportion", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "p": { "title": "P", "type": "number" }, "vr": { "default": false, "title": "Vr", "type": "boolean" } }, "required": [ "num_iters", "p" ], "title": "ProxSkipServerConfig", "type": "object" } **Client config:** ``ProxSkipClientConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``batch_size`` - int - * * - ``num_epochs`` - int - * * - ``lr`` - float - 0.01 * - ``optimizer`` - str - "SCAFFOLD" * - ``scheduler`` - Optional[dict] - None * - ``extra_observes`` - list[str] - (factory) * - ``verbose`` - int - 1 * - ``latency`` - float - 0.0 * - ``algorithm`` - str - "" * - ``vr`` - bool - False .. dropdown:: ``ProxSkipClientConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Client config for the ProxSkip algorithm.\n\nParameters\n----------\nbatch_size : int\n The batch size.\nnum_epochs : int\n The number of epochs.\nlr : float, default 1e-2\n The learning rate.\nvr : bool, default False\n Whether to use variance reduction.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``scheduler`` : dict, optional\n The scheduler config.\n None for no scheduler, using constant learning rate.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training,\n which would be recorded in evaluated metrics,\n sent to the server, and written to the log file.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``latency`` : float, default 0.0\n The latency of the client.\n Not used in the current version, reserved for future use.", "properties": { "batch_size": { "title": "Batch Size", "type": "integer" }, "num_epochs": { "title": "Num Epochs", "type": "integer" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "optimizer": { "default": "SCAFFOLD", "title": "Optimizer", "type": "string" }, "scheduler": { "anyOf": [ { "additionalProperties": true, "type": "object" }, { "type": "null" } ], "default": null, "title": "Scheduler" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "latency": { "default": 0.0, "title": "Latency", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "vr": { "default": false, "title": "Vr", "type": "boolean" } }, "required": [ "batch_size", "num_epochs" ], "title": "ProxSkipClientConfig", "type": "object" } .. _config-scaffold: SCAFFOLD -------- **Server config:** ``SCAFFOLDServerConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``num_iters`` - int - * * - ``num_clients`` - Optional[int] - None * - ``clients_sample_ratio`` - float - * * - ``log_dir`` - Optional[str] - None * - ``txt_logger`` - bool - True * - ``json_logger`` - bool - True * - ``eval_every`` - int - 1 * - ``visiable_gpus`` - Optional[list[int]] - None * - ``extra_observes`` - list[str] - (factory) * - ``seed`` - int - 0 * - ``tag`` - Optional[str] - None * - ``verbose`` - int - 1 * - ``gpu_proportion`` - float - 0.2 * - ``algorithm`` - str - "" * - ``lr`` - float - * .. dropdown:: ``SCAFFOLDServerConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Server config for the SCAFFOLD algorithm.\n\nParameters\n----------\nnum_iters : int\n The number of (outer) iterations.\nnum_clients : int\n The number of clients.\nclients_sample_ratio : float\n The ratio of clients to sample for each iteration.\nlr : float\n The learning rate.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``log_dir`` : str or pathlib.Path, optional\n The log directory.\n If not specified, will use the default log directory.\n If not absolute, will be relative to the default log directory.\n - ``txt_logger`` : bool, default True\n Whether to use txt logger.\n - ``json_logger`` : bool, default True\n Whether to use json logger.\n - ``eval_every`` : int, default 1\n The number of iterations to evaluate the model.\n - ``visiable_gpus`` : Sequence[int], optional\n Visable GPU IDs for allocating devices for clients.\n Defaults to use all GPUs if available.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training.\n - ``seed`` : int, default 0\n The random seed.\n - ``tag`` : str, optional\n The tag of the experiment.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``gpu_proportion`` : float, default 0.2\n The proportion of clients to use GPU.\n Used to similate the system heterogeneity of the clients.\n Not used in the current version, reserved for future use.", "properties": { "num_iters": { "title": "Num Iters", "type": "integer" }, "num_clients": { "anyOf": [ { "type": "integer" }, { "type": "null" } ], "default": null, "title": "Num Clients" }, "clients_sample_ratio": { "title": "Clients Sample Ratio", "type": "number" }, "log_dir": { "anyOf": [ { "type": "string" }, { "format": "path", "type": "string" }, { "type": "null" } ], "default": null, "title": "Log Dir" }, "txt_logger": { "default": true, "title": "Txt Logger", "type": "boolean" }, "json_logger": { "default": true, "title": "Json Logger", "type": "boolean" }, "eval_every": { "default": 1, "title": "Eval Every", "type": "integer" }, "visiable_gpus": { "anyOf": [ { "items": { "type": "integer" }, "type": "array" }, { "type": "null" } ], "default": null, "title": "Visiable Gpus" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "seed": { "default": 0, "title": "Seed", "type": "integer" }, "tag": { "anyOf": [ { "type": "string" }, { "type": "null" } ], "default": null, "title": "Tag" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "gpu_proportion": { "default": 0.2, "title": "Gpu Proportion", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "lr": { "title": "Lr", "type": "number" } }, "required": [ "num_iters", "clients_sample_ratio", "lr" ], "title": "SCAFFOLDServerConfig", "type": "object" } **Client config:** ``SCAFFOLDClientConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``batch_size`` - int - * * - ``num_epochs`` - int - * * - ``lr`` - float - 0.01 * - ``optimizer`` - str - "SGD" * - ``scheduler`` - Optional[dict] - None * - ``extra_observes`` - list[str] - (factory) * - ``verbose`` - int - 1 * - ``latency`` - float - 0.0 * - ``algorithm`` - str - "" * - ``control_variate_update_rule`` - int - 1 .. dropdown:: ``SCAFFOLDClientConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Client config for the SCAFFOLD algorithm.\n\nParameters\n----------\nbatch_size : int\n The batch size.\nnum_epochs : int\n The number of epochs.\nlr : float, default 1e-2\n The learning rate.\ncontrol_variate_update_rule : int, default 1\n The update rule for the control variates.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``scheduler`` : dict, optional\n The scheduler config.\n None for no scheduler, using constant learning rate.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training,\n which would be recorded in evaluated metrics,\n sent to the server, and written to the log file.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``latency`` : float, default 0.0\n The latency of the client.\n Not used in the current version, reserved for future use.", "properties": { "batch_size": { "title": "Batch Size", "type": "integer" }, "num_epochs": { "title": "Num Epochs", "type": "integer" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "optimizer": { "default": "SGD", "title": "Optimizer", "type": "string" }, "scheduler": { "anyOf": [ { "additionalProperties": true, "type": "object" }, { "type": "null" } ], "default": null, "title": "Scheduler" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "latency": { "default": 0.0, "title": "Latency", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "control_variate_update_rule": { "default": 1, "title": "Control Variate Update Rule", "type": "integer" } }, "required": [ "batch_size", "num_epochs" ], "title": "SCAFFOLDClientConfig", "type": "object" } .. _config-pfedmac: pFedMac ------- **Server config:** ``pFedMacServerConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``num_iters`` - int - * * - ``num_clients`` - Optional[int] - None * - ``clients_sample_ratio`` - float - * * - ``log_dir`` - Optional[str] - None * - ``txt_logger`` - bool - True * - ``json_logger`` - bool - True * - ``eval_every`` - int - 1 * - ``visiable_gpus`` - Optional[list[int]] - None * - ``extra_observes`` - list[str] - (factory) * - ``seed`` - int - 0 * - ``tag`` - Optional[str] - None * - ``verbose`` - int - 1 * - ``gpu_proportion`` - float - 0.2 * - ``algorithm`` - str - "" * - ``beta`` - float - 1.0 * - ``vr`` - bool - False .. dropdown:: ``pFedMacServerConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Server config for the pFedMac algorithm.\n\nParameters\n----------\nnum_iters : int\n The number of (outer) iterations.\nnum_clients : int\n The number of clients.\nclients_sample_ratio : float\n The ratio of clients to sample for each iteration.\nbeta : float, default 1.0\n The beta (inertia) parameter for pFedMac.\nvr : bool, default False\n Whether to use variance reduction.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``log_dir`` : str or pathlib.Path, optional\n The log directory.\n If not specified, will use the default log directory.\n If not absolute, will be relative to the default log directory.\n - ``txt_logger`` : bool, default True\n Whether to use txt logger.\n - ``json_logger`` : bool, default True\n Whether to use json logger.\n - ``eval_every`` : int, default 1\n The number of iterations to evaluate the model.\n - ``visiable_gpus`` : Sequence[int], optional\n Visable GPU IDs for allocating devices for clients.\n Defaults to use all GPUs if available.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training.\n - ``seed`` : int, default 0\n The random seed.\n - ``tag`` : str, optional\n The tag of the experiment.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``gpu_proportion`` : float, default 0.2\n The proportion of clients to use GPU.\n Used to similate the system heterogeneity of the clients.\n Not used in the current version, reserved for future use.", "properties": { "num_iters": { "title": "Num Iters", "type": "integer" }, "num_clients": { "anyOf": [ { "type": "integer" }, { "type": "null" } ], "default": null, "title": "Num Clients" }, "clients_sample_ratio": { "title": "Clients Sample Ratio", "type": "number" }, "log_dir": { "anyOf": [ { "type": "string" }, { "format": "path", "type": "string" }, { "type": "null" } ], "default": null, "title": "Log Dir" }, "txt_logger": { "default": true, "title": "Txt Logger", "type": "boolean" }, "json_logger": { "default": true, "title": "Json Logger", "type": "boolean" }, "eval_every": { "default": 1, "title": "Eval Every", "type": "integer" }, "visiable_gpus": { "anyOf": [ { "items": { "type": "integer" }, "type": "array" }, { "type": "null" } ], "default": null, "title": "Visiable Gpus" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "seed": { "default": 0, "title": "Seed", "type": "integer" }, "tag": { "anyOf": [ { "type": "string" }, { "type": "null" } ], "default": null, "title": "Tag" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "gpu_proportion": { "default": 0.2, "title": "Gpu Proportion", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "beta": { "default": 1.0, "title": "Beta", "type": "number" }, "vr": { "default": false, "title": "Vr", "type": "boolean" } }, "required": [ "num_iters", "clients_sample_ratio" ], "title": "pFedMacServerConfig", "type": "object" } **Client config:** ``pFedMacClientConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``batch_size`` - int - * * - ``num_epochs`` - int - * * - ``lr`` - float - 0.01 * - ``optimizer`` - str - "pFedMac" * - ``scheduler`` - Optional[dict] - None * - ``extra_observes`` - list[str] - (factory) * - ``verbose`` - int - 1 * - ``latency`` - float - 0.0 * - ``algorithm`` - str - "" * - ``lam`` - float - 15.0 * - ``vr`` - bool - False .. dropdown:: ``pFedMacClientConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Client config for the pFedMac algorithm.\n\nParameters\n----------\nbatch_size : int\n The batch size.\nnum_epochs : int\n The number of epochs.\nlr : float, default 1e-2\n The learning rate.\nlam : float, default 15.0\n The lambda parameter for pFedMac.\nvr : bool, default False\n Whether to use variance reduction.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``scheduler`` : dict, optional\n The scheduler config.\n None for no scheduler, using constant learning rate.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training,\n which would be recorded in evaluated metrics,\n sent to the server, and written to the log file.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``latency`` : float, default 0.0\n The latency of the client.\n Not used in the current version, reserved for future use.", "properties": { "batch_size": { "title": "Batch Size", "type": "integer" }, "num_epochs": { "title": "Num Epochs", "type": "integer" }, "lr": { "default": 0.01, "title": "Lr", "type": "number" }, "optimizer": { "default": "pFedMac", "title": "Optimizer", "type": "string" }, "scheduler": { "anyOf": [ { "additionalProperties": true, "type": "object" }, { "type": "null" } ], "default": null, "title": "Scheduler" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "latency": { "default": 0.0, "title": "Latency", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "lam": { "default": 15.0, "title": "Lam", "type": "number" }, "vr": { "default": false, "title": "Vr", "type": "boolean" } }, "required": [ "batch_size", "num_epochs" ], "title": "pFedMacClientConfig", "type": "object" } .. _config-pfedme: pFedMe ------ **Server config:** ``pFedMeServerConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``num_iters`` - int - * * - ``num_clients`` - Optional[int] - None * - ``clients_sample_ratio`` - float - * * - ``log_dir`` - Optional[str] - None * - ``txt_logger`` - bool - True * - ``json_logger`` - bool - True * - ``eval_every`` - int - 1 * - ``visiable_gpus`` - Optional[list[int]] - None * - ``extra_observes`` - list[str] - (factory) * - ``seed`` - int - 0 * - ``tag`` - Optional[str] - None * - ``verbose`` - int - 1 * - ``gpu_proportion`` - float - 0.2 * - ``algorithm`` - str - "" * - ``beta`` - float - 1.0 .. dropdown:: ``pFedMeServerConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Server config for the pFedMe algorithm.\n\nParameters\n----------\nnum_iters : int\n The number of (outer) iterations.\nnum_clients : int\n The number of clients.\nclients_sample_ratio : float\n The ratio of clients to sample for each iteration.\nbeta : float, default 1.0\n The beta (inertia) parameter for model aggregation.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``log_dir`` : str or pathlib.Path, optional\n The log directory.\n If not specified, will use the default log directory.\n If not absolute, will be relative to the default log directory.\n - ``txt_logger`` : bool, default True\n Whether to use txt logger.\n - ``json_logger`` : bool, default True\n Whether to use json logger.\n - ``eval_every`` : int, default 1\n The number of iterations to evaluate the model.\n - ``visiable_gpus`` : Sequence[int], optional\n Visable GPU IDs for allocating devices for clients.\n Defaults to use all GPUs if available.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training.\n - ``seed`` : int, default 0\n The random seed.\n - ``tag`` : str, optional\n The tag of the experiment.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``gpu_proportion`` : float, default 0.2\n The proportion of clients to use GPU.\n Used to similate the system heterogeneity of the clients.\n Not used in the current version, reserved for future use.", "properties": { "num_iters": { "title": "Num Iters", "type": "integer" }, "num_clients": { "anyOf": [ { "type": "integer" }, { "type": "null" } ], "default": null, "title": "Num Clients" }, "clients_sample_ratio": { "title": "Clients Sample Ratio", "type": "number" }, "log_dir": { "anyOf": [ { "type": "string" }, { "format": "path", "type": "string" }, { "type": "null" } ], "default": null, "title": "Log Dir" }, "txt_logger": { "default": true, "title": "Txt Logger", "type": "boolean" }, "json_logger": { "default": true, "title": "Json Logger", "type": "boolean" }, "eval_every": { "default": 1, "title": "Eval Every", "type": "integer" }, "visiable_gpus": { "anyOf": [ { "items": { "type": "integer" }, "type": "array" }, { "type": "null" } ], "default": null, "title": "Visiable Gpus" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "seed": { "default": 0, "title": "Seed", "type": "integer" }, "tag": { "anyOf": [ { "type": "string" }, { "type": "null" } ], "default": null, "title": "Tag" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "gpu_proportion": { "default": 0.2, "title": "Gpu Proportion", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "beta": { "default": 1.0, "title": "Beta", "type": "number" } }, "required": [ "num_iters", "clients_sample_ratio" ], "title": "pFedMeServerConfig", "type": "object" } **Client config:** ``pFedMeClientConfig`` .. list-table:: :header-rows: 1 :widths: 25 30 45 * - Field - Type - Default * - ``batch_size`` - int - * * - ``num_epochs`` - int - * * - ``lr`` - float - 0.005 * - ``optimizer`` - str - "pFedMe" * - ``scheduler`` - Optional[dict] - None * - ``extra_observes`` - list[str] - (factory) * - ``verbose`` - int - 1 * - ``latency`` - float - 0.0 * - ``algorithm`` - str - "" * - ``num_steps`` - int - 30 * - ``lamda`` - float - 15.0 * - ``eta`` - float - 0.001 * - ``mu`` - float - 0.001 .. dropdown:: ``pFedMeClientConfig`` JSON schema :animate: fade-in-slide-down .. code-block:: json { "additionalProperties": true, "description": "Client config for the pFedMe algorithm.\n\nParameters\n----------\nbatch_size : int\n The batch size.\nnum_epochs : int\n The number of epochs.\nlr : float, default 5e-3\n The learning rate for personalized model training.\nnum_steps : int, default 30\n The number of steps for each epoch.\nlamda : float, default 15.0\n The lambda parameter for pFedMe,\n i.e. the coefficient of the proximal term.\neta : float, default 1e-3\n The eta (learning rate) parameter for pFedMe.\nmu : float, default 1e-3\n The mu (momentum) parameter for pFedMe.\n**kwargs : dict, optional\n Additional keyword arguments:\n\n - ``scheduler`` : dict, optional\n The scheduler config.\n None for no scheduler, using constant learning rate.\n - ``extra_observes`` : List[str], optional\n Extra attributes to observe during training,\n which would be recorded in evaluated metrics,\n sent to the server, and written to the log file.\n - ``verbose`` : int, default 1\n The verbosity level.\n - ``latency`` : float, default 0.0\n The latency of the client.\n Not used in the current version, reserved for future use.\n\n\nReferences\n----------\n.. [1] https://github.com/CharlieDinh/pFedMe/blob/master/FLAlgorithms/users/userpFedMe.py\n\n.. note::\n\n 1. `lr` is the `personal_learning_rate` in the original implementation\n 2. `eta` is the `learning_rate` in the original implementation\n 3. `mu` is the momentum factor in the original implemented optimzer", "properties": { "batch_size": { "title": "Batch Size", "type": "integer" }, "num_epochs": { "title": "Num Epochs", "type": "integer" }, "lr": { "default": 0.005, "title": "Lr", "type": "number" }, "optimizer": { "default": "pFedMe", "title": "Optimizer", "type": "string" }, "scheduler": { "anyOf": [ { "additionalProperties": true, "type": "object" }, { "type": "null" } ], "default": null, "title": "Scheduler" }, "extra_observes": { "items": { "type": "string" }, "title": "Extra Observes", "type": "array" }, "verbose": { "default": 1, "title": "Verbose", "type": "integer" }, "latency": { "default": 0.0, "title": "Latency", "type": "number" }, "algorithm": { "default": "", "title": "Algorithm", "type": "string" }, "num_steps": { "default": 30, "title": "Num Steps", "type": "integer" }, "lamda": { "default": 15.0, "title": "Lamda", "type": "number" }, "eta": { "default": 0.001, "title": "Eta", "type": "number" }, "mu": { "default": 0.001, "title": "Mu", "type": "number" } }, "required": [ "batch_size", "num_epochs" ], "title": "pFedMeClientConfig", "type": "object" }