The Pytorch library has many inbuilt Math functions. These can be used to perform various specific Mathematical Operation. In this chapter of Pytorch Tutorial, you will learn about a handful of math functions in Pytorch.
torch.abs()
torch.abs(x) returns a tensor whose elements are the absolute values of the elements in the original tensor x. The abs() function does not make changes to the original tensor, it creates a new tensor to store the result and returns it. To make changes to the original tensor, you should call abs_() function.
abs() and abs_() can also be called directly on a tensor.
Example
x = torch.tensor([[-1, 4], [-2, 1]])
# Does not make changes to tensor x, stores the result in y
y = torch.abs(x)
print(y)
# Outputs- tensor([[1, 4], [2, 1]])
# abs() function is called directly on the tensor
print(x.abs())
# Outputs- tensor([[1, 4], [2, 1]])
# Changes the original tensor
x.abs_()
print(x)
# Outputs- tensor([[1, 4], [2, 1]])torch.round()
torch.round(x) returns a tensor with the elements of the original tensor x, rounded to the nearest integer. The round() function does not make changes to the original tensor, it creates a new tensor to store the result and returns it. To make changes to the original tensor, you should call round_() function.
round() and round_() can also be called directly on a tensor.
Example
x = torch.tensor([[-1.1, 4.7], [-2.9, 1.24]])
# Does not make changes to tensor x, stores the result in y
y = torch.round(x)
print(y)
# Outputs- tensor([[-1., 5.], [-3., 1.]])
# round() function is called directly on the tensor
print(x.round())
# Outputs- tensor([[-1., 5.], [-3., 1.]])
# Changes the original tensor
x.round_()
print(x)
# Outputs- tensor([[-1., 5.], [-3., 1.]])Note– The round() and round_() functions accepts tensors with floating point datatype only.
torch.ceil()
torch.ceil(x) returns a tensor with the elements of the original tensor x, rounded up to the nearest integer. The ceil() function does not make changes to the original tensor, it creates a new tensor to store the result and returns it. To make changes to the original tensor, you should call ceil_() function.
ceil() and ceil_() can also be called directly on a tensor.
Example
x = torch.tensor([[-1.1, 4.7], [-2.9, 1.24]])
# Does not make changes to tensor x, stores the result in y
y = torch.ceil(x)
print(y)
# Outputs- tensor([[-1., 5.], [-2., 2.]])
# ceil() function is called directly on the tensor
print(x.ceil())
# Outputs- tensor([[-1., 5.], [-2., 2.]])
# Changes the original tensor
x.ceil_()
print(x)
# Outputs- tensor([[-1., 5.], [-2., 2.]])Note– The ceil() and ceil_() functions accepts tensors with floating point datatype only.
torch.floor()
torch.floor(x) returns a tensor with the elements of the original tensor x, rounded down to the nearest integer. The floor() function does not make changes to the original tensor, it creates a new tensor to store the result and returns it. To make changes to the original tensor, you should call floor_() function.
floor() and floor_() can also be called directly on a tensor.
Example
x = torch.tensor([[-1.1, 4.7], [-2.9, 1.24]])
# Does not make changes to tensor x, stores the result in y
y = torch.floor(x)
print(y)
# Outputs- tensor([[-2., 4.], [-3., 1.]])
# floor() function is called directly on the tensor
print(x.floor())
# Outputs- tensor([[-2., 4.], [-3., 1.]])
# Changes the original tensor
x.floor_()
print(x)
# Outputs- tensor([[-2., 4.], [-3., 1.]])Note– The floor() and floor_() functions accepts tensors with floating point datatype only.
torch.sin()
torch.sin(x) returns a tensor whose elements are the sine of the elements(in radians) of the original tensor x. The sin() function does not make changes to the original tensor, it creates a new tensor to store the result and returns it. To make changes to the original tensor, you should call sin_() function.
sin() and sin_() can also be called directly on a tensor.
Example
x = torch.tensor([1.0, 2.0, 3.0])
# Does not make changes to tensor x, stores the result in y
y = torch.sin(x)
print(y)
# Outputs- tensor([0.8415, 0.9093, 0.1411])
# sin() function is called directly on the tensor
print(x.sin())
# Outputs- tensor([0.8415, 0.9093, 0.1411])
# Changes the original tensor
x.sin_()
print(x)
# Outputs- tensor([0.8415, 0.9093, 0.1411])Note– The sin() and sin_() functions take the input in radians.
Note– The sin() and sin_() functions accepts tensors with floating point datatype only.
torch.cos()
torch.cos(x) returns a tensor whose elements are the cosine of the elements(in radians) of the original tensor x. The cos() function does not make changes to the original tensor, it creates a new tensor to store the result and returns it. To make changes to the original tensor, you should call cos_() function.
cos() and cos_() can also be called directly on a tensor.
Example
x = torch.tensor([1.0, 2.0, 3.0])
# Does not make changes to tensor x, stores the result in y
y = torch.cos(x)
print(y)
# Outputs- tensor([0.5403, -0.4161, -0.9900])
# cos() function is called directly on the tensor
print(x.cos())
# Outputs- tensor([0.5403, -0.4161, -0.9900])
# Changes the original tensor
x.cos_()
print(x)
# Outputs- tensor([0.5403, -0.4161, -0.9900])Note– The cos() and cos_() functions take the input in radians.
Note– The cos() and cos_() functions accepts tensors with floating point datatype only.
torch.sqrt()
torch.sqrt(x) returns a tensor whose elements are the square root of the elements in the original tensor x. The sqrt() function does not make changes to the original tensor, it creates a new tensor to store the result and returns it. To make changes to the original tensor, you should call sqrt_() function.
sqrt() and sqrt_() can also be called directly on a tensor.
Example
x = torch.tensor([2., 3., 4., 5.])
# Does not make changes to tensor x, stores the result in y
y = torch.sqrt(x)
print(y)
# Outputs- tensor([1.4142, 1.7321, 2.0000, 2.2361])
# sqrt() function is called directly on the tensor
print(x.sqrt())
# Outputs- tensor([1.4142, 1.7321, 2.0000, 2.2361])
# Changes the original tensor
x.sqrt_()
print(x)
# Outputs- tensor([1.4142, 1.7321, 2.0000, 2.2361])Note– The sqrt() and sqrt_() function accept tensors with floating point datatype only.
torch.log()
torch.log(x) returns a tensor whose elements are the natural logarithm(to the base e) of the elements in the original tensor x. The log() function does not make changes to the original tensor, it creates a new tensor to store the result and returns it. To make changes to the original tensor, you should call log_() function.
log() and log_() can also be called directly on a tensor.
Example
x = torch.tensor([1.4, 2, 4.0])
# Does not make changes to tensor x, stores the result in y
y = torch.log(x)
print(y)
# Outputs- tensor([0.3365, 0.6931, 1.3863])
# log() function is called directly on the tensor
print(x.log())
# Outputs- tensor([0.5403, -0.4161, -0.9900])
# Changes the original tensor
x.log_()
print(x)
# Outputs- tensor([0.5403, -0.4161, -0.9900])Note– The log() and log_() function accept tensors with floating point datatype only.
Math Reference
For a complete reference of pytorch math functions, refer to the Pytorch Math Functions Reference.