chainer.function_hooks.PrintHook

class chainer.function_hooks.PrintHook(sep='', end='n', file=<_io.TextIOWrapper name='<stdout>' mode='w' encoding='UTF-8'>, flush=True)

Function hook that prints debug information.

This function hook outputs the debug information of input arguments of forward and backward methods involved in the hooked functions at preprocessing time (that is, just before each method is called).

Unlike simple “debug print” technique, where users insert print functions at every function to be inspected, we can show the information of all functions involved with single with statement.

Further, this hook enables us to show the information of backward methods without inserting print functions into Chainer’s library code.

Variables:

• sep – Separator of print function.
• end – Character to be added at the end of print function.
• file – Output file_like object that that redirect to.
• flush – If True, this hook forcibly flushes the text stream at the end of preprocessing.

Example

The basic usage is to use it with with statement.

>>> from chainer import function_hooks
>>> l = L.Linear(10, 10)
>>> x = chainer.Variable(np.zeros((1, 10), 'f'))
>>> with chainer.function_hooks.PrintHook():
...     y = l(x)
...     z = F.sum(y)
...     z.backward() 

In this example, PrintHook shows the debug information of forward propagation of LinearFunction (which is implicitly called by l) and Sum (called by F.sum) and backward propagation of z and y.

Methods

__enter__()

__exit__(*_)

added(function=None)

Callback function invoked when a function hook is added

Parameters:function (FunctionNode) – Function object to which the function hook is added.

backward_postprocess(function, in_data, out_grad)

Callback function invoked after backward propagation.

Parameters:

• function (FunctionNode) – Function object to which the function hook is registered.
• in_data (tuple of numpy.ndarray or tuple of cupy.ndarray) – Input of forward propagation.
• out_grad (tuple of numpy.ndarray or tuple of cupy.ndarray) – Gradient data of backward propagation.

backward_preprocess(function, in_data, out_grad)

deleted(function=None)

Callback function invoked when a function hook is deleted

Parameters:function (FunctionNode) – Function object to which the function hook is deleted.

forward_postprocess(function, in_data)

Callback function invoked after forward propagation.

Parameters:

• function (FunctionNode) – Function object to which the function hook is registered.
• in_data (tuple of numpy.ndarray or tuple of cupy.ndarray) – Input data of forward propagation.

forward_preprocess(function, in_data)