fannypack
A loose collection of tools for training PyTorch models.
Contents include helpers for:
- Experiment management, Tensorboard logging, and checkpointing (Python + CLI)
- Reading and manipulating arrays and tensors stored in containers: converting between types, moving across (Torch) devices, slicing across shared dimensions
- Freezing and unfreezing portions of PyTorch modules
- Reading and writing time series data/trajectory files via hdf5
- and a lot more!
See documentation for full overview of functionality.
Installation
Standard installation:
pip install fannypack
Install from source:
git clone https://github.com/brentyi/fannypack.git
cd fannypack && pip install -e .
Development
Tests can be run with pytest, and documentation can be built by running
make github in the docsource/ directory.
Tooling: black and isort for formatting, flake8 for linting, and mypy for static type-checking.
Until numpy 1.20.0 is released,
type-checking also requires that NumPy stubs are installed manually:
pip install https://github.com/numpy/numpy-stubs/tarball/master