# pytorch basic exercises, basic operations of tensor
import torch
# Several ways to generate tensor (the comment part is size)
"""
# The difference between
() is a Python class, and more explicitly, it is an alias for the default tensor type ().
([4,5]) will call the constructor of Tensor class __init__ to generate a single-precision floating point type tensor.
() is just a Python function, the function prototype is:
(data, dtype=None, device=None, requires_grad=False)
The data can be: list, tuple, array, scalar and other types.
() You can copy the data part in data (rather than refer directly).
Generate corresponding, according to the original data type.
"""
a = (4, 5) # [4,5]
b = (4, 5) # Random generation,[4,5]
aa = (2, 1, 2, 4, 2) # [2,1,2,4,2]
bb = ([-1, -2, 3]) # [3]
# mutual conversion between tensor and other types
# 1、numpy
c = () # tensor to numpy
d = torch.from_numpy(c) # numpy to tensor
# 2、list
c = () # tensor to list
# 3、CPU/GPU
# c = ()
# d = ()
#The mathematical operation of tensor
# Add up
c = a + b
c = (a, b)
d = (a, 20) # You can add scalars or tensors
# Find the absolute value of each element
c = torch.abs(bb)
# Mathematical Statistics
# Find the average value, the second parameter indicates which dimension; count from 0, from left to right; the final result is to remove the dimension at that position and the remaining dimensions
c = (a, 0) # Find the average by row, the returned size is the number of columns, original: [4,5], result: [5]
d = (a, 1) # Find the average by column, the returned size is the number of rows, original: [4,5], result: [4]
e = (aa, 1) # Original: [2,1,2,4,2], Results: [2,2,4,2]
# determine whether it is equal
c = (a, b) # For two tensors with the same size, return the tensor with the same size.
e = (a, 0) # The second parameter can be a tensor of the same size or a number
d = (a, b) # Return to True/False
# Common operations of neural network tensor
# Find all elements
c = ()
# Dimension compression, the second parameter represents the dimension, counting from left to right, starting from 0
d = (aa) # Remove the dimension with dimension 1, [2,2,4,2]
e = (aa, 2) # If the dimension of the specified dimension is 1, it will be removed, otherwise it will still be the original value [2,1,2,4,2]
# Dimension expansion, the second parameter represents the dimension, must be displayed, and the value cannot be greater than the original dimension
d = (a, 1) # Add one dimension to the specified position, [4,1,5]
e = (a, 2) # [4,5,1]
# tensor splicing stack
d = ((a, b)) # The input must be tuple/list. The stack splicing is to create a new dimension, and then splice it at that latitude. The default newly created dimension is dimension 0, [2,4,5]
e = ((a, b), 1) # [4,2,5]
f = ((a, b), -1) # [4,5,2]
g = ([a, b, b]) # [3,4,5]
# tensor splicing cat
d = ([a, b]) # The input must be tuple/list, cat is spliced on existing dimensions, and the default is dimension 0, [8,5]
e = ((a, b), 1) # [4,10]
# Tensor expansion
# 1. Only dimensions with dimension value 1 can be expanded, and the extended Tensor will not allocate new memory, but will create a new view and return it on the original basis;
# 2. Please keep the dimension value unchanged for dimensions that do not need to be expanded.
# d = (3, 1).expand(3, 4) # The value of the expanded dimension is the same, [3, 4]