5 cool PyTorch Functions that are beginner-friendly

PyTorch is a Python library with a wide variety of functions and operations, mostly used for deep learning. Data is stored in tensors. A tensor is a vector, number, matrix, or an n-dimensional array. Let’s see a few simple functions which can come in handy for you.

  • torch.randn
  • torch.cat
  • torch.reshape
  • torch.unbind
  • torch.take

torch.randn

This generates a tensor filled with random numbers from a normal distribution with mean 0 and variance 1. The shape of the resulting tensor is determined by argument size.

size can be a sequence of integers defining the shape of the tensor or it can be a collection like a tuple

requires_grad specifies if autograd should record operations on the tensor

you cannot give two tuples as argument as that doesn’t define a valid shape

torch.cat

This function is used to concatenate given sequence of tensors. To do thus, all tensors should either have the same shape or be empty

This works since all tensors are of same shape

concatenates according to the given dimension

breaks since all tensors are not of the same shape

torch.reshape

This function is useful when you want to change the shape of a given tensor if it is possible

returns a tensor of shape (2,6) from a tensor of shape (3,4)

This fails because we cannot obtain a tensor of shape (4,4) from a tensor of shape (3,4)

torch.unbind

This is used to remove a dimension of a tensor

By default, the dimension is set to 0

torch.take

This function is useful to extract the elements at specific indices from a tensor

simple use case of the function

You cannot access elements out of range

Conclusion

This wraps up the 5 useful torch.Tensor functions that are simple to use.