wandb-callbacks
provides some additional Callbacks for Weights & Biases.
Callbacks currently implemented:
ActivationCallback
- visualizes the activations of a layer
- activations are computed for a sample of each class
DeadReluCallback
- logs the number of dead relus in each layer
- prints warning if the percentage is higher than a threshold
GradCAMCallback
- Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization
- produces a coarse localization map highlighting the important regions in the image for predicting the class of the image
pip install wandb-callbacks
git clone https://github.com/FabianGroeger96/wandb-callbacks
Can be found in notebooks/sample_implementation.ipynb
.
Open to ideas and for helpers to develop the package further.