PyPyTorch is a simple deep learning framework implemented by Python3
Contents
Introduction
PyPyTorch is very similar to PyTorch, most of APIs are the same as PyTorch, at this stage, this framework enables us to understand how dynamic neural network works. If you master PyTorch, you can master PyPyTorch in a short time. There are some important and core modules in PyPyTorch. By the way, you can call its nickname PPT for short
Module | Description |
---|---|
pypytorch | The entry of PyPyTorch framework, once you import pypytorch, you can work everything with PyPyTorch. |
data | Privides dataset, dataloader and transforms. |
functions | A operator library for Tensor object in PyPyTorch. |
nn | A high level neural network library in PyPyTorch which built on functions module, you can build a neural network model very fast with the help of nn. |
optim | There are some optimizers, including SGD, Adam and so on. |
Download the mnist.zip dataset at link BaiduNetDisk, then unzip it to examples/data/, if examples/data/ doesn't exist, you need create directory example/data/ by yourself. The full path of mnist dataset is examples/data/mnist/
Besides, you can change configuration at examples/scripts/mnist/config.py
Installation
PyPI
-
pip install pypytorch
- As for Windows user, I recommend you to use Miniconda.
From Source
git clone https://github.com/dark-ai/pypytorch.git pypython
cd pypytorch
pip3 install -r requirements.txt
make install && make clean
Docker
TODO
- SGD Optimizer
- Adam Optimizer(Adam implemented by me sucks)
- Sequential
- ReLU
- BatchNorm
- DeConv2d(Waiting to test)
- Upsample