API Reference

• Index

Package

pytorch_pfn_extras

Training Loop

Trainer

engine.create_trainer(models, optimizers, ...)	Creates a trainer object.
engine.create_evaluator(models, *[, ...])	Creates an evaluator object.
handler.BaseLogic([options])
handler.Logic([model_name, options])	A set of methods that defines the training logic.
handler.BaseHandler(logic, options, *args, ...)	Base class of Handler.
handler.Handler(logic, entry_runtime, options)	A set of callback functions to perform device-specific operations.
runtime.BaseRuntime(device_spec, options)	A base class for collections of device-specific callback functions.
runtime.PyTorchRuntime(device_spec, options)	A collections of callback functions for the devices that PyTorch supports by default.

Extensions Manager

training.ExtensionsManager(models, ...[, ...])	Manages the extensions and the current status.
training.IgniteExtensionsManager(engine, ...)	Manages extensions and the current status in Ignite training loop.

Extensions

training.extension.make_extension([trigger, ...])	Decorator to make given function into an extension.
training.extension.Extension()	Base class of extensions.
training.extension.ExtensionEntry(extension, *)	Extension and options.

training.extensions.BestValue(key, compare)	Extension traces the best value of a specific key in the observation.
training.extensions.Evaluator(self, ...[, ...])	An extension to evaluate models on a validation set.
training.extensions.LogReport([keys, ...])	An extension to output the accumulated results to a log file.
training.extensions.MaxValue(key[, trigger])	Extension traces the maximum value of a specific key in the observation.
training.extensions.MicroAverage(...[, trigger])	Calculates micro-average ratio.
training.extensions.MinValue(key[, trigger])	Extension traces the maximum value of a specific key in the observation.
training.extensions.observe_lr(optimizer[, ...])	Returns an extension to record the learning rate.
training.extensions.observe_value(...)	Returns an extension to continuously record a value.
training.extensions.ParameterStatistics(links)	An extension to report parameter statistics.
training.extensions.PlotReport(y_keys[, ...])	An extension to output plots.
training.extensions.PrintReport([entries, ...])	An extension to print the accumulated results.
training.extensions.ProgressBar([...])	An extension to print a progress bar and recent training status.
training.extensions.ProfileReport([...])	Writes the profile results to a file.
training.extensions.snapshot([savefun, ...])	Returns a trainer extension to take snapshots of the trainer.
training.extensions.Slack(channel[, msg, ...])	An extension to communicate with Slack.
training.extensions.SlackWebhook(url[, msg, ...])	An extension to communicate with Slack using Incoming Webhook.
training.extensions.VariableStatisticsPlot(targets)	An extension to plot statistics for Tensors.

Triggers

training.triggers.EarlyStoppingTrigger(self)	Trigger for Early Stopping
training.triggers.IntervalTrigger(period, unit)	Trigger based on a fixed interval.
training.triggers.ManualScheduleTrigger(...)	Trigger invoked at specified point(s) of iterations or epochs.
training.triggers.BestValueTrigger(key, compare)	Trigger invoked when specific value becomes best.
training.triggers.MaxValueTrigger(key[, trigger])	Trigger invoked when specific value becomes maximum.
training.triggers.MinValueTrigger(key[, trigger])	Trigger invoked when specific value becomes minimum.
training.triggers.OnceTrigger([call_on_resume])	Trigger based on the starting point of the iteration.
training.triggers.TimeTrigger(period)	Trigger based on a fixed time interval.

Reporting

reporting.Reporter()	Object to which observed values are reported.
reporting.report(values[, observer])	Reports observed values with the current reporter object.
reporting.report_scope(observation)	Returns a report scope with the current reporter.

Logging

logging.get_logger(name)

Returns a child logger to be used by applications.

Profiler

profiler.TimeSummary.report(tag[, use_cuda])	Context manager to automatically report execution times.
profiler.clear_tracer()	Resets the status of the global tracer.
profiler.enable_global_trace(enable)	Enable or disable tracing for all the threads.
profiler.enable_thread_trace(enable)	Enable or disable tracing for the current thread.
profiler.get_tracer([tracer_cls])	Gets the current global tracer.
profiler.ChromeTracer([max_event_count, ...])	Tracer object that outputs a timeline in Chrome format.
profiler.TraceableDataset(dataset, tag[, tracer])	Utility class to trace a Dataset inside the DataLoader worker threads.

Distributed Training

nn.parallel.DistributedDataParallel(module)	Module for distributed data parallelism
distributed.initialize_ompi_environment(*[, ...])	Initialize torch.distributed environments with values taken from OpenMPI.

Check Pointing

utils.checkpoint

Lazy Modules

nn.Ensure(*[, shape, dtype, broadcastable, ...])	Module to check the shape of a tensor.
nn.ensure(tensor[, shape, dtype, ...])	Checks the shape and type of a tensor.
nn.LazyLinear(in_features, *args, **kwargs)	Linear module with lazy weight initialization.
nn.LazyConv1d(in_channels, *args, **kwargs)	Conv1d module with lazy weight initialization.
nn.LazyConv2d(in_channels, *args, **kwargs)	Conv2d module with lazy weight initialization.
nn.LazyConv3d(in_channels, *args, **kwargs)	Conv3d module with lazy weight initialization.
nn.LazyBatchNorm1d(num_features, *args, **kwargs)	BatchNorm1d module with lazy weight initialization.
nn.LazyBatchNorm2d(num_features, *args, **kwargs)	BatchNorm2d module with lazy weight initialization.
nn.LazyBatchNorm3d(num_features, *args, **kwargs)	BatchNorm3d module with lazy weight initialization.

ONNX

Export

onnx.export(model, args, f[, return_output, ...])	Export model into ONNX Graph.
onnx.export_testcase(model, args, out_dir, *)	Export model and I/O tensors of the model in protobuf format.

Annotation

onnx.annotate(**attrs)	Annotation parameters to the target function.
onnx.apply_annotation(fn, *args, **attrs)	Annotation applier to the target function
onnx.scoped_anchor(**attrs)	Add anchor node to the scoped modules
onnx.export(model, args, f[, return_output, ...])	Export model into ONNX Graph.
onnx.export_testcase(model, args, out_dir, *)	Export model and I/O tensors of the model in protobuf format.

Datasets

dataset.SharedDataset(sm_size[, cache_type])	Dataset that caches the load samples in shared memory
dataset.TabularDataset(*args, **kwds)	An abstract class that represents tabular dataset.
dataset.ItemNotFoundException

Config

config.Config(config[, types])

config_types.optuna_types(trial)
config_types.load_path_with_optuna_types(...)

NumPy/CuPy Compatibility

from_ndarray(ndarray)	Creates a torch.Tensor from a numpy.ndarray or cupy.ndarray.
as_ndarray(tensor)	Creates a numpy.ndarray or cupy.ndarray from torch.Tensor.
get_xp(obj)	Returns a module of ndarray implementation (numpy or cupy) for the given obj.
as_numpy_dtype(torch_dtype)	Returns NumPy dtype for the given PyTorch dtype.
from_numpy_dtype(numpy_dtype)	Returns PyTorch dtype for the given NumPy dtype.

cuda.stream(stream)	Context-manager that selects a given stream.
cuda.use_torch_mempool_in_cupy()	Use the PyTorch memory pool in CuPy.
cuda.use_default_mempool_in_cupy()	Use the default memory pool in CuPy.