Skip to main content

PyTorch utilities for ML, specifically speech

Project description

Build Status Documentation Status License

pydrobert-pytorch

PyTorch utilities for Machine Learning. This is an eclectic mix of utilities that I've used in my various projects, but have been tailored to be as generic as possible.

This is student-driven code, so don't expect a stable API. I'll try to use semantic versioning, but the best way to keep functionality stable is by pinning the version in the requirements or by forking.

Overview

Functionality is split by submodule. They include

  • pydrobert.torch.estimators: Implements a number of popular gradient estimators in ML literature. Useful for RL tasks, or anything that needs discrete samples.
  • pydrobert.torch.training: Utilities that should be useful to most model training loops, even the most esoteric. TrainingStateController can be used to persist model and optimizer states across runs, and manage non-determinism.
  • pydrobert.torch.data: Primarily serves as a means to manipulate speech data. It contains subclasses of torch.utils.data.DataLoader for both random and sequential access of speech data, as well as examples of how to use them. pydrobert.torch.data also contains functions for transducing back and forth between tensors and transcriptions. In particular, this package comes with command line hooks for converting to and from NIST sclite file formats. Feature data and senone alignments from Kaldi can be converted to this format using the command line hooks from pydrobert-kaldi.

Documentation

Installation

pydrobert-pytorch is available through both Conda and PyPI.

conda install -c sdrobert pydrobert-pytorch
pip install pydrobert-pytorch

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

pydrobert-pytorch-0.0.2.tar.gz (40.0 kB view hashes)

Uploaded Source

Built Distribution

pydrobert_pytorch-0.0.2-py2.py3-none-any.whl (41.2 kB view hashes)

Uploaded Python 2 Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page