mirror of
https://github.com/zebrajr/pytorch.git
synced 2025-12-07 12:21:27 +01:00
32b3841553
* Improvize documentation 1. Add formula for erf, erfinv 2. Make exp, expm1 similar to log, log1p 3. Symbol change in ge, le, ne, isnan * Fix minor nit in the docstring * More doc improvements 1. Added some formulae 2. Complete scanning till "Other Operations" in Tensor docs * Add more changes 1. Modify all torch.Tensor wherever required * Fix Conv docs 1. Fix minor nits in the references for LAPACK routines * Improve Pooling docs 1. Fix lint error * Improve docs for RNN, Normalization and Padding 1. Fix flake8 error for pooling * Final fixes for torch.nn.* docs. 1. Improve Loss Function documentation 2. Improve Vision Layers documentation * Fix lint error * Improve docstrings in torch.nn.init * Fix lint error * Fix minor error in torch.nn.init.sparse * Fix Activation and Utils Docs 1. Fix Math Errors 2. Add explicit clean to Makefile in docs to prevent running graph generation script while cleaning 3. Fix utils docs * Make PYCMD a Makefile argument, clear up prints in the build_activation_images.py * Fix batch norm doc error
95 lines
2.5 KiB
Python
95 lines
2.5 KiB
Python
"""
|
|
This script will generate input-out plots for all of the activation
|
|
functions. These are for use in the documentation, and potentially in
|
|
online tutorials.
|
|
"""
|
|
|
|
import os.path
|
|
import torch.nn.modules.activation
|
|
import torch.autograd
|
|
import matplotlib
|
|
|
|
matplotlib.use('Agg')
|
|
|
|
import pylab
|
|
|
|
|
|
# Create a directory for the images, if it doesn't exist
|
|
DOCS_PATH = os.path.realpath(os.path.join(__file__, "../../.."))
|
|
ACTIVATION_IMAGE_PATH = os.path.join(
|
|
DOCS_PATH,
|
|
"source/_static/img/activation/"
|
|
)
|
|
|
|
print(ACTIVATION_IMAGE_PATH)
|
|
|
|
if not os.path.exists(ACTIVATION_IMAGE_PATH):
|
|
os.mkdir(ACTIVATION_IMAGE_PATH)
|
|
|
|
# In a refactor, these ought to go into their own module or entry
|
|
# points so we can generate this list programmaticly
|
|
functions = [
|
|
'ELU',
|
|
'Hardshrink',
|
|
'Hardtanh',
|
|
'LeakyReLU', # Perhaps we should add text explaining slight slope?
|
|
'LogSigmoid',
|
|
'PReLU',
|
|
'ReLU',
|
|
'ReLU6',
|
|
'RReLU',
|
|
'SELU',
|
|
'Sigmoid',
|
|
'Softplus',
|
|
'Softshrink',
|
|
'Softsign',
|
|
'Tanh',
|
|
'Tanhshrink'
|
|
# 'Threshold' Omit, pending cleanup. See PR5457
|
|
]
|
|
|
|
|
|
def plot_function(function, **args):
|
|
"""
|
|
Plot a function on the current plot. The additional arguments may
|
|
be used to specify color, alpha, etc.
|
|
"""
|
|
xrange = torch.arange(-7.0, 7.0, 0.01) # We need to go beyond 6 for ReLU6
|
|
pylab.plot(
|
|
xrange.numpy(),
|
|
function(torch.autograd.Variable(xrange)).data.numpy(),
|
|
**args
|
|
)
|
|
|
|
|
|
# Step through all the functions
|
|
for function_name in functions:
|
|
plot_path = os.path.join(ACTIVATION_IMAGE_PATH, function_name + ".png")
|
|
if not os.path.exists(plot_path):
|
|
function = torch.nn.modules.activation.__dict__[function_name]()
|
|
|
|
# Start a new plot
|
|
pylab.clf()
|
|
|
|
# Add an overlay on the background of faint traces of all the other
|
|
# functions. This is nice when flipping through images
|
|
for background_function in functions:
|
|
plot_function(
|
|
torch.nn.modules.activation.__dict__[background_function](),
|
|
alpha=0.03, color='k'
|
|
)
|
|
|
|
# Plot the current function
|
|
plot_function(function)
|
|
|
|
# The titles are a little redundant, given context?
|
|
pylab.title(function_name + " activation function")
|
|
pylab.xlabel("Input")
|
|
pylab.ylabel("Output")
|
|
pylab.xlim([-7, 7])
|
|
pylab.ylim([-7, 7])
|
|
|
|
# And save it
|
|
pylab.savefig(plot_path)
|
|
print('Saved activation image for {} at {}'.format(function, plot_path))
|