Pytorch Light is a Python package that provides several tools about modelling and applications.
- Installing from pypi:
pip install pytorchlight
There are some features as follows:
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.
Application for model.
Features:
- train
- evaluate
- batch loss and batch metric
- model wrapper
Application for data.
Features:
- Numpy.ndarray to/from Torch.tensor
- Numpy.ndarray to/from Torch.Dataset
- Numpy.ndarray to/from Torch.DataLoader
Application for log.
Features:
- debug
- info
- warn
- error
- critical
Application for visualization based on numpy.ndarray.
Require:
- In the notebook mode,
%matplotlib inline
should be implement first.
Features:
- learning curves
- batch curves
- one or multi images