giagrad.nn.Module

class giagrad.nn.Module(*args, **kwargs)

Base class for all neural network modules.

Your models should also subclass this class.

Modules can also contain other Modules, allowing to nest them in a tree structure. You can assign the submodules as regular attributes:

import giagrad.nn as nn

class Model(nn.Module):
    def __init__(self):
        super().__init__()
        self.l1 = nn.Linear(28*28, 100)
        self.l2 = nn.Linear(100, 10)

    def __call__(self, x):
        x = self.l1(x).relu()
        return self.l2(x).relu()

Submodules assigned in this way will be registered thanks to Model constructor in __odict__ variable in the same order as they are defined.

>>> m = Model()
>>> m.__odict__
OrderedDict([('l1', Layer(in=784, out=100, bias=True)),
             ('l2', Layer(in=100, out=10, bias=True))])

Note

For the example above, an __init__() call to the parent class must be made before assignment on the child if you want to have it in __odict__.

Variables:

training (bool, default: True) – Represents whether this module is in training or evaluation mode.

Methods

add_module	Adds a child module to the current module.
apply	Applies fn recursively to every submodule as well as self.
eval	Sets all submodules in evaluation mode including self.
parameters	Returns an iterator over all submodules and self parameters.
train	Sets all submodules in training mode including self.
zero_grad	Sets gradients of all tensors returned by parameters to zero.