Pytorch Light

Pytorch Light is a Python package that provides several tools about modelling and applications.

There are some features as follows:

• Model
• Data
• Log
• Visualization

requirements.txt is also provided.

Next feature is going to be achieved as soon as possible.

• Adversarial Learning

And some applications are coming:

• Computer Vision
• Natural Language
• Rereinforcement Learning

Note that these features are only validated on cpu.

Model

Application for model.

Features:

1. train
2. evaluate
3. batch loss and batch metric
4. model wrapper

Data

Application for data.

Features:

1. Numpy.ndarray to/from Torch.tensor
2. Numpy.ndarray to/from Torch.Dataset
3. Numpy.ndarray to/from Torch.DataLoader

Log

Application for log.

Features:

1. debug
2. info
3. warn
4. error
5. critical

Visualization

Application for visualization based on numpy.ndarray.

Require:

• In the notebook mode, %matplotlib inline should be implement first.

Features:

1. learning curves
2. batch curves
3. one or multi images