How to use torch.flatten () and torch.nn.Flatten () in Pytorch

This article is a detailed introduction to "how to use torch.flatten() and torch.nn.Flatten() in Pytorch". The content is detailed, the steps are clear, and the details are properly handled. I hope this article "how to use torch.flatten() and torch.nn.Flatten() in Pytorch" can help you solve your doubts. Let's go deeper and learn new knowledge together with the ideas of the small editor.

torch.flatten(x) equals torch.flatten(x, 0) by default pulls the tensor into a one-dimensional vector, that is, flattening from the first dimension, torch.flatten(x, 1) means flattening from the second dimension.

import torchx=torch.randn(2,4,2)print(x) z=torch.flatten(x)print(z) w=torch.flatten(x,1)print(w) Output is: tensor([[-0.9814, 0. 8251], [ 0.8197, -1.0426], [-0.8185, -1.3367], [-0.6293, 0.6714]], [[-0.5973, -0.0944], [ 0.3720, 0.0672], [ 0.2681, 1.8025], [-0.0606, 0.4855]]]) tensor([-0.9814, 0.8251, 0.8197, -1.0426, -0.8185, -1.3367, -0.6293, 0.6714, -0.5973, -0.0944, 0.3720, 0.0672, 0.2681, 1.8025, -0.0606, 0.4855]) tensor([[-0.9814, 0.8251, 0.8197, -1.0426, -0.8185, -1.3367, -0.6293, 0.6714], [-0.5973, -0.0944, 0.3720, 0.0672, 0.2681, 1.8025, -0.0606, 0.4855]])

torch.flatten(x,0,1) represents flattening between the first and second dimensions

import torchx=torch.randn(2,4,2)print(x) w=torch.flatten(x,0,1) #Length of the first dimension is 2, length of the second dimension is 4, length after flattening is 2* 4 print (w.shape) print(w) Output is: tensor([[-0.5523, -0.1132], [-2.2659, -0.0316], [ 0.1372, -0.8486], [-0.3593, -0.2622]], [[-0.9130, 1.0038], [-0.3996, 0.4934], [ 1.7269, 0.8215], [ 0.1207, -0.9590]]]) torch.Size([8, 2]) tensor([[-0.5523, -0.1132], [-2.2659, -0.0316], [ 0.1372, -0.8486], [-0.3593, -0.2622], [-0.9130, 1.0038], [-0.3996, 0.4934], [ 1.7269, 0.8215], [ 0.1207, -0.9590]])

For torch.nn.Flatten(), because it is used in neural networks, the input is a batch of data, the first dimension is batch, and it is usually necessary to pull a data into one dimension instead of pulling a batch of data into one dimension. So torch.nn.Flatten() starts flattening from the second dimension by default.

import torch#random 32 channels 5*5 graph x=torch.randn(32,1,5,5) model=torch.nn.Sequential( #Input channel 1, output channel 6, 3*3 convolution kernel, step size 1, padding=1 torch.nn.Conv2d(1,6,3,1,1), torch.nn.Flatten())output=model(x)print (output.shape)#6 *(7-3+1)*(7-3+1) Output is: torch.Size([32, 150]) Read here, this article "How to use torch.flatten() and torch.nn.Flatten() in Pytorch" has been introduced, if you want to master the knowledge points of this article, you still need to practice it yourself to understand, if you want to know more about related articles, welcome to pay attention to the industry information channel.