module training.data_loader

Short summary

module onnxcustom.training.data_loader

Manipulate data for training.

source on GitHub

Classes

class	truncated documentation
OrtDataLoader	Draws consecutive random observations from a dataset by batch. It iterates over the datasets by drawing batch_size …

Properties

property	truncated documentation
data_np	Returns a tuple of the datasets in numpy.
data_ort	Returns a tuple of the datasets in onnxruntime C_OrtValue.

Methods

method	truncated documentation
__getstate__	Removes any non pickable attribute.
__init__
__len__	Returns the number of observations.
__repr__	usual
__setstate__	Restores any non pickable attribute.
_next_iter
iter_bind	Iterates over the datasets by drawing batch_size consecutive observations. Modifies a bind structure.
iter_numpy	Iterates over the datasets by drawing batch_size consecutive observations. This iterator is slow as it …
iter_ortvalue	Iterates over the datasets by drawing batch_size consecutive observations. This iterator is slow as it …

Documentation

Manipulate data for training.

source on GitHub

class onnxcustom.training.data_loader.OrtDataLoader(X, y, sample_weight=None, batch_size=20, device='cpu', random_iter=True)

Bases: object

Draws consecutive random observations from a dataset by batch. It iterates over the datasets by drawing batch_size consecutive observations.

Parameters:

• X – features

• y – labels

• sample_weight – weight or None

• batch_size – batch size (consecutive observations)

• device – C_OrtDevice or a string such as ‘cpu’

• random_iter – random iteration

See example Train a scikit-learn neural network with onnxruntime-training on GPU.

source on GitHub

__getstate__()

Removes any non pickable attribute.

__init__(X, y, sample_weight=None, batch_size=20, device='cpu', random_iter=True)

__len__()

Returns the number of observations.

__repr__()

usual

__setstate__(state)

Restores any non pickable attribute.

_next_iter(previous)

property data_np

Returns a tuple of the datasets in numpy.

property data_ort

Returns a tuple of the datasets in onnxruntime C_OrtValue.

iter_bind(bind, names)

Iterates over the datasets by drawing batch_size consecutive observations. Modifies a bind structure.

source on GitHub

iter_numpy()

Iterates over the datasets by drawing batch_size consecutive observations. This iterator is slow as it copies the data of every batch. The function yields C_OrtValue.

source on GitHub

iter_ortvalue()

Iterates over the datasets by drawing batch_size consecutive observations. This iterator is slow as it copies the data of every batch. The function yields C_OrtValue.

source on GitHub