Inspired by Andrej Karpathy's implementation of 1D tensors, I implemented n-dimensional tensors. Since I am not sure whether a PR (pull request) is needed, I did not submit a PR to his repository. To maintain the integrity of a single file, I did not reuse some code; instead, I copied it directly.

tensor1d.c readme

currently support

• create tensor
  • create empty tensor using shape
  • create tensor using np.ndarray
• tensor getitem
  • get a float item
• tensor nelement
• tensor.item()
• get tensor meta data
  • ndim
  • shape

currently testing

• slice tensor
• slice tensor keepdim

TODOs:

• feature
  • create tensor using list, tuple
  • get tensor item as tensor
  • slice tensor
  • slice tensor keepdim
  • tensor_setitem()
  • tensor_slice_setitem()
• make tests better
• make the code more readable and cleaner
• write a post to explain my implementation and help greeners approach this comfortable

Good related resources:

• PyTorch internals
• Numpy paper

MIT