MCPcopy Create free account
hub / github.com/PaddlePaddle/Paddle / train_from_dataset

Method train_from_dataset

python/paddle/base/executor.py:3011–3091  ·  view source on GitHub ↗

Train from a pre-defined Dataset. Dataset is defined in paddle.base.dataset. Given a program, either a program or compiled program, train_from_dataset will consume all data samples in dataset. Input scope can be given by users. By default, scope is global_scope(). Th

(
        self,
        program: Program | CompiledProgram | None = None,
        dataset: DatasetBase | _FleetDatasetBase | None = None,
        scope: core._Scope | None = None,
        thread: int = 0,
        debug: bool = False,
        fetch_list: list[Tensor] | None = None,
        fetch_info: list[str] | None = None,
        print_period: int = 100,
        fetch_handler: FetchHandler | None = None,
    )

Source from the content-addressed store, hash-verified

3009 return trainer_instance
3010
3011 def train_from_dataset(
3012 self,
3013 program: Program | CompiledProgram | None = None,
3014 dataset: DatasetBase | _FleetDatasetBase | None = None,
3015 scope: core._Scope | None = None,
3016 thread: int = 0,
3017 debug: bool = False,
3018 fetch_list: list[Tensor] | None = None,
3019 fetch_info: list[str] | None = None,
3020 print_period: int = 100,
3021 fetch_handler: FetchHandler | None = None,
3022 ) -> None:
3023 """
3024 Train from a pre-defined Dataset. Dataset is defined in paddle.base.dataset.
3025 Given a program, either a program or compiled program, train_from_dataset will
3026 consume all data samples in dataset. Input scope can be given by users. By default,
3027 scope is global_scope(). The total number of thread run in training is `thread`.
3028 Thread number used in training will be minimum value of threadnum in Dataset and
3029 the value of thread in this interface. Debug can be set so that executor will display
3030 Run-Time for all operators and the throughputs of current training task.
3031
3032 Note: train_from_dataset will destroy all resources created within executor for each run.
3033
3034 Args:
3035 program(Program|CompiledProgram): the program that needs to be run,
3036 if not provided, then default_main_program (not compiled) will be used.
3037 dataset(paddle.base.Dataset): dataset created outside this function,
3038 a user should provide a well-defined dataset before calling this function.
3039 Please check the document of Dataset if needed.
3040 scope(Scope): the scope used to run this program, you can switch it to different scope
3041 for each run. default is global_scope
3042 thread(int): number of thread a user wants to run in this function. Default is 0, which
3043 means using thread num of dataset
3044 debug(bool): whether a user wants to run train_from_dataset
3045 fetch_list(Tensor List): fetch Tensor list, each variable will be printed
3046 during training
3047 fetch_info(String List): print information for each Tensor, its length should be equal
3048 to fetch_list
3049 print_period(int): the number of mini-batches for each print, default is 100
3050 fetch_handler(FetchHandler): a user define class for fetch output.
3051
3052 Returns:
3053 None
3054
3055 Examples:
3056
3057 .. code-block:: pycon
3058
3059 >>> # doctest: +SKIP("This does not supported in PIR mode")
3060
3061 >>> import paddle
3062
3063 >>> paddle.enable_static()
3064 >>> place = paddle.CPUPlace() # you can set place = paddle.CUDAPlace(0) to use gpu
3065 >>> exe = paddle.static.Executor(place)
3066 >>> x = paddle.static.data(name="x", shape=[None, 10, 10], dtype="int64")
3067 >>> y = paddle.static.data(name="y", shape=[None, 1], dtype="int64", lod_level=1)
3068 >>> dataset = paddle.base.DatasetFactory().create_dataset()

Callers 10

fl_ps_trainFunction · 0.95
do_dataset_trainingMethod · 0.95
do_dataset_trainingMethod · 0.95
callback_trainMethod · 0.45
dataset_train_loopMethod · 0.45
do_dataset_trainingMethod · 0.45

Calls 1

_run_from_datasetMethod · 0.95

Tested by 2