Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/kamo-naoyuki/pytorch_complex
A temporal module for PyTorch-ComplexTensor
https://github.com/kamo-naoyuki/pytorch_complex
complex-numbers dnn python python3 pytorch
Last synced: about 1 month ago
JSON representation
A temporal module for PyTorch-ComplexTensor
- Host: GitHub
- URL: https://github.com/kamo-naoyuki/pytorch_complex
- Owner: kamo-naoyuki
- Created: 2018-12-27T12:20:22.000Z (about 6 years ago)
- Default Branch: master
- Last Pushed: 2024-06-28T07:10:19.000Z (7 months ago)
- Last Synced: 2024-12-20T11:07:01.487Z (about 1 month ago)
- Topics: complex-numbers, dnn, python, python3, pytorch
- Language: Python
- Homepage:
- Size: 113 KB
- Stars: 45
- Watchers: 5
- Forks: 18
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
- awesome-speech-enhancement - [Code
README
# pytorch_complex
[](https://badge.fury.io/py/torch-complex)
[](https://pypi.org/project/torch-complex/)
[](https://pepy.tech/project/torch-complex)
[](https://travis-ci.org/kamo-naoyuki/pytorch_complex)
[](https://codecov.io/gh/kamo-naoyuki/pytorch_complex)
A temporal python class for PyTorch-ComplexTensor
## What is this?
A Python class to perform as `ComplexTensor` in PyTorch: Nothing except for the following,
```python
class ComplexTensor:
def __init__(self, ...):
self.real = torch.Tensor(...)
self.imag = torch.Tensor(...)
```
### Why?
PyTorch is great DNN Python library, except that it doesn't support `ComplexTensor` in Python level.
https://github.com/pytorch/pytorch/issues/755
I'm looking forward to the completion, but I need `ComplexTensor` for now.
I created this cheap module for the temporal replacement of it. Thus, I'll throw away this project as soon as `ComplexTensor` is completely supported!
## Requirements
```
Python>=3.6
PyTorch>=1.0
```
## Install
```
pip install torch_complex
```
## How to use
### Basic mathematical operation
```python
import numpy as np
from torch_complex.tensor import ComplexTensor
real = np.random.randn(3, 10, 10)
imag = np.random.randn(3, 10, 10)
x = ComplexTensor(real, imag)
x.numpy()
x + x
x * x
x - x
x / x
x ** 1.5
x @ x # Batch-matmul
x.conj()
x.inverse() # Batch-inverse
```
All are implemented with combinations of computation of `RealTensor` in python level, thus the speed is not good enough.
### Functional
```python
import torch_complex.functional as F
F.cat([x, x])
F.stack([x, x])
F.matmul(x, x) # Same as x @ x
F.einsum('bij,bjk,bkl->bil', [x, x, x])
```
### For DNN
Almost all methods that `torch.Tensor` has are implemented.
```python
x.cuda()
x.cpu()
(x + x).sum().backward()
```