Improved handling of PyTorch module hooks

All your hooks in one place

TackleBox puts all registered hooks in one centralized location, and allows you to query them by function, module and id.

activate/deactivate on the fly

Register hooks with your PyTorch modules, then activate or deactivate them at any point during training/evaluation.

quick to integrate

Use python contexts to activate any collection of previously registered hooks using a single line of code.

easy to extend

With TackleBox you can analyze your model faster than ever. Visualize activations or prune features in the blink of an eye!

To add TackleBox to your deep learning project, simply install it from PyPi with

pip install tacklebox

Once installed, follow the videos below, or check out the github page for more complete documentation.