Framework for rapid research and development of machine learning projects using PyTorch.
Project description
Framework for rapid research and development of machine learning projects using PyTorch.
Longer description coming soon!
Important features:
Reduced boilerplate and simple specifications of complex workflows
Easy and flexible customization of Learner by overriding methods
Built-in scoring algorithms like holdout and cross validation
Straightforward hyperparameter tuning with built-in tuning algorithms like random search and grid search
Basic usage: custom modules with luz.Module functionality. Simply define your model, inherit from luz.Module, and use model.train/model.test.
Next level: simplified training algorithms with transforms and handlers.
Next level: straightforward hyperparameter tuning with various scoring mechanisms.
Data preparation scheme.
Transforms: Functions for conditionining the training process which are invertible and whose inverse is applied to the resulting predictor.
Preprocessors: Functions for altering the prediction task which are not necessarily invertible.
Model.
Training scheme.
Overall learning algorithm.
Hyperparameter selection.
Unified development interface through Learner object. Simply inherit luz.Learner, define the model, loader, and param functions, and you’re good to go. Add a hyperparams function to enable tuning and make tuned parameters accessible in the model and param functions.
Getting Started
Prerequisites
Installing
To install, open a shell terminal and run:
`conda create -n luz -c conda-forge -c pytorch -c kijana luz`
Versioning
License
This project is licensed under the MIT License - see the LICENSE file for details.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.