torchlib.module.misc package

Submodules

torchlib.module.misc.transform module

class torchlib.module.misc.transform.Standardization(mean=None, std=None, axis=None, unbiased=False, extra=False)

Bases: torch.nn.modules.module.Module

\[\bar{X} = \frac{X-\mu}{\sigma} \]

Parameters

• X (tensor) – data to be normalized,

• mean (list or None, optional) – mean value (the default is None, which means auto computed)

• std (list or None, optional) – Standard deviation (the default is None, which means auto computed)

• axis (list or int, optional) – Specify the axis for computing mean and standard deviation (the default is None, which means all elements)

• unbiased (bool, optional) – If unbiased is False, then the standard-deviation will be calculated via the biased estimator. Otherwise, Bessel’s correction will be used.

• extra (bool, optional) – If True, also return the mean and std (the default is False, which means just return the standardized data)

Examples

import torchlib as tl
tl.setseed(seed=2020, target='torch')
x = th.randn(5, 2, 4, 3)

f = Standardization(axis=(2, 3), unbiased=False, extra=True)
y, meanv, stdv = f(x)
print(y[0], y.shape)

g = th.nn.InstanceNorm2d(2)

z = g(x)
print(z[0], z.shape)

f = Standardization(axis=(0, 2, 3), unbiased=False, extra=True)
y, meanv, stdv = f(x)
print(y[0], y.shape)

g = th.nn.BatchNorm2d(2)

z = g(x)
print(z[0], z.shape)

The results are:

tensor([[[ 0.2761, -0.1161, -1.3316],
         [ 0.4918,  0.5450, -0.7350],
         [ 1.5699, -1.8567,  1.7366],
         [-0.1463, -0.1318, -0.3019]],

        [[-1.0576,  0.5794, -0.6489],
         [-0.3410, -1.6589,  0.2531],
         [ 1.2150,  0.7262,  0.3333],
         [-1.1270, -0.2132,  1.9397]]]) torch.Size([5, 2, 4, 3])
tensor([[[ 0.2761, -0.1161, -1.3316],
         [ 0.4918,  0.5450, -0.7350],
         [ 1.5699, -1.8567,  1.7366],
         [-0.1463, -0.1318, -0.3019]],

        [[-1.0576,  0.5794, -0.6489],
         [-0.3410, -1.6588,  0.2531],
         [ 1.2150,  0.7262,  0.3333],
         [-1.1270, -0.2132,  1.9397]]]) torch.Size([5, 2, 4, 3])
tensor([[[ 0.0498, -0.2576, -1.2101],
         [ 0.2188,  0.2605, -0.7426],
         [ 1.0637, -1.6216,  1.1943],
         [-0.2812, -0.2698, -0.4032]],

        [[-1.1965,  0.0760, -0.8788],
         [-0.6395, -1.6639, -0.1776],
         [ 0.5701,  0.1901, -0.1153],
         [-1.2505, -0.5402,  1.1335]]]) torch.Size([5, 2, 4, 3])
tensor([[[ 0.0498, -0.2576, -1.2101],
         [ 0.2188,  0.2605, -0.7426],
         [ 1.0637, -1.6216,  1.1943],
         [-0.2812, -0.2698, -0.4032]],

        [[-1.1965,  0.0760, -0.8788],
         [-0.6395, -1.6639, -0.1776],
         [ 0.5701,  0.1901, -0.1153],
         [-1.2505, -0.5401,  1.1335]]], grad_fn=<SelectBackward>) torch.Size([5, 2, 4, 3])

forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

Module contents