Learn practical skills, build real-world projects, and advance your career
Updated 4 years ago
bold text# Torch functions
Overview
An short introduction about PyTorch and about the chosen functions.
- Torch.nn.functional.conv2d
- Torch.functional.relu and sigmoid
- Tensor views
- Tensror sparse
- distribution Bernoulli
# Import torch and other required modules
import torch
import torch.nn as nn
import torch.nn.functional as F
Function 1 - torch.nn.ConvNd
Add some explanations
# Example 1
weights = torch.randn(8,4,3,3)
inputs = torch.randn( 1,4,5,5)
conv2 = nn.functional.conv2d(inputs, weights, padding=1)
print(conv2)
tensor([[[[ 3.8680e+00, 5.0820e+00, -1.0063e+01, -8.8937e-01, 2.9871e+00],
[ 4.9803e+00, 1.2524e+00, -3.5393e-01, -5.7376e-02, 6.9019e+00],
[-5.3350e+00, -9.8769e+00, -3.6961e+00, -3.9565e+00, -4.4567e+00],
[-6.1955e+00, 2.1419e+00, 2.0403e-01, -2.7475e+00, 1.1002e+01],
[ 9.4631e+00, 8.7178e-01, 2.1688e+00, 7.1002e+00, 1.4491e+00]],
[[ 2.5075e+00, -1.3761e+00, 5.8364e+00, -6.6801e+00, -1.5880e+00],
[-2.2470e+00, 1.1989e+01, -3.5544e+00, -6.0607e+00, 4.7767e+00],
[ 8.3233e+00, -4.9677e+00, -4.8019e+00, 5.1346e+00, 1.2297e+00],
[-5.5617e+00, -5.0186e+00, 6.7778e-01, -1.2989e+01, -1.4212e+00],
[-3.0105e+00, -3.6638e+00, 5.0936e+00, -3.5987e+00, -1.7590e+00]],
[[ 1.3004e+00, 3.8636e+00, 1.3745e+00, -4.9612e+00, -1.8940e+00],
[ 2.1954e-02, 1.5935e+00, 8.0963e+00, 5.9663e+00, 1.7166e+00],
[ 4.0973e+00, 2.3024e+00, 1.8693e+00, -1.5356e-01, 4.0245e+00],
[-1.9318e+00, -2.2527e-01, 2.9208e+00, -2.0158e+00, -1.8689e+00],
[-6.4032e-01, -1.0677e+01, 4.3496e+00, -7.2807e-01, 1.0686e+00]],
[[-9.1417e-03, -6.8313e+00, 8.9343e+00, -4.0506e+00, -1.3380e+00],
[-1.2367e+01, 3.8457e+00, 5.6170e+00, -1.0560e+01, 5.7460e+00],
[ 3.8650e+00, 3.5976e+00, -2.8892e+00, 1.2291e+00, -1.8745e+00],
[ 7.3313e+00, -5.0042e+00, 1.3423e+00, 7.3137e-01, -6.0769e+00],
[-2.5561e+00, 1.6833e+00, 6.1848e+00, -2.8550e+00, 3.3420e+00]],
[[ 3.3553e+00, -3.4772e+00, -6.8116e-01, -8.9013e-01, -2.8952e+00],
[ 6.5434e-01, -1.3575e+00, 8.0224e+00, -1.2259e+00, -1.6162e+00],
[ 3.4559e+00, 7.5026e+00, -1.7468e+00, 6.3646e+00, -8.8414e-01],
[ 6.0009e+00, 3.7210e+00, -5.2953e+00, -4.7632e+00, -2.7142e+00],
[ 3.4380e-01, -8.2212e-01, 4.7768e+00, -5.3364e+00, -6.1744e+00]],
[[ 3.6878e-01, -1.3324e-01, -2.7351e+00, -1.0917e+00, -4.7788e+00],
[-2.2061e+00, 6.6370e-01, -9.1234e-01, -3.1340e+00, -3.5064e+00],
[-4.6640e+00, 3.9090e+00, -9.8812e+00, -2.6794e+00, -7.2603e-02],
[ 4.6101e-01, 2.0525e+00, 3.3524e-01, -4.6056e+00, -2.4658e+00],
[ 5.7663e-01, 3.5680e-01, -1.5147e+00, 4.4018e+00, -9.7672e-01]],
[[-1.7987e+00, 6.5154e-01, -1.5246e+00, 3.6759e+00, 4.4590e+00],
[ 2.1427e-01, 2.2629e-01, -2.5939e+00, -2.4905e+00, 7.1225e-01],
[ 2.6632e+00, -1.9831e+00, 3.7383e+00, -1.4604e+00, -1.1919e+00],
[ 2.7843e+00, -5.6450e+00, -7.5471e+00, 2.8669e-02, -2.2600e+00],
[ 9.1320e-01, 1.1682e+00, -3.0183e+00, -4.0003e+00, 6.1508e+00]],
[[ 5.3196e+00, -4.0816e+00, -7.5325e-01, -2.1066e+00, -4.3520e+00],
[-2.3228e+00, -2.5519e+00, 3.2655e+00, -2.2304e+00, 9.4970e-01],
[ 3.7479e+00, -2.7900e+00, 2.4141e+00, -2.6348e-02, -1.7779e+00],
[-6.8566e+00, -5.0837e+00, -1.0266e+01, -2.6781e+00, -4.6948e+00],
[ 5.6717e-01, -3.0513e+00, -3.4173e+00, -1.8979e+00, 4.9804e+00]]]])
Explanation about example:create convolutional NN matrix of [1, 8, 5, 5] dim for a kernel given input and filter list values