Texar

Texar is an open source toolkit for machine learning, with a focus on natural language processing and text generation. It comes from ASYML, which describes its approach as machine learning as machine assembly, and it is offered in two editions, Texar TF for TensorFlow and Texar PyTorch with mostly the same interfaces. It is aimed at researchers and engineers who want to compose their own models and algorithms from modular parts. Texar provides modules for data processing, model architectures, loss functions, training and inference algorithms and evaluation. The PyTorch edition adds ready to use data APIs and high level training interfaces that remove boilerplate code. The same organization publishes related open source projects: Forte, a toolkit for building complex NLP pipelines that can wrap models from spaCy, StanfordNLP and AllenNLP, and Stave, a toolkit for building annotation and visualization front ends for NLP projects. Installation is through pip, conda or source, and each project has GitHub repositories, documentation, examples and tutorials. The toolkits are free to use, and no commercial plans or prices are published.