Keras
Use the Keras callback to automatically save all the metrics and the loss values tracked in model.fit.
import wandbfrom wandb.keras import WandbCallbackwandb.init(config={"hyper": "parameter"})# Magicmodel.fit(X_train, y_train, validation_data=(X_test, y_test),callbacks=[WandbCallback()])
Try our integration out in a colab notebook, complete with video tutorial, or see our example projects for a complete script example.
Keras WandbCallback() class supports a number of options:
Keyword argument | Default | Description |
monitor | val_loss | The training metric used to measure performance for saving the best model. i.e. val_loss |
mode | auto | 'min', 'max', or 'auto': How to compare the training metric specified in |
save_weights_only | False | only save the weights instead of the entire model |
save_model | True | save the model if it's improved at each step |
log_weights | False | log the values of each layers parameters at each epoch |
log_gradients | False | log the gradients of each layers parametres at each epoch |
training_data | None | tuple (X,y) needed for calculating gradients |
data_type | None | the type of data we're saving, currently only "image" is supported |
labels | None | only used if data_type is specified, list of labels to convert numeric output to if you are building classifier. (supports binary classification) |
predictions | 36 | the number of predictions to make if data_type is specified. Max is 100. |
generator | None | if using data augmentation and data_type you can specify a generator to make predictions with. |
If you're setting use_multiprocessing=True and seeing the error Error('You must call wandb.init() before wandb.config.batch_size') then try this:
In the Sequence class init, add:
wandb.init(group='...')In your main program, make sure you're using
if __name__ == "__main__":and then put the rest of your script logic inside that.
We've created a few examples for you to see how the integration works:
Example on Github: Fashion MNIST example in a Python script
Run in Google Colab: A simple notebook example to get you started
Wandb Dashboard: View result on W&B
