Take a look at keras_tuner_cifar.py example script, which demonstrates the integration of ClearML in a code that uses Keras Tuner.
The script does the following:
- Creates a
Hyperbandobject, which uses Keras Tuner's
Hyperbandtuner. It finds the best hyperparameters to train a network on a CIFAR10 dataset.
- When the
Hyperbandobject is created, instantiates a
ClearMLTunerLoggerobject and assigns it to the
ClearMLTunerLoggerclass provides the required binding for ClearML automatic logging.
When the script runs, it logs:
- Tabular summary of hyperparameters tested and their metrics by trial ID
- Scalar plot showing metrics for all runs
- Summary plot
- Output model with configuration and snapshot location.
ClearML logs the scalars from training each network. They appear in the project's page in the ClearML web UI, under RESULTS > SCALARS.
ClearML automatically logs the parameters of each experiment run in the hyperparameter search. They appear in tabular form in RESULTS > PLOTS.
ClearML automatically stores the output model. It appears in ARTIFACTS > Output Model.
Model details, such as snap locations, appear in the MODELS tab.
The model configuration is stored with the model.
ClearML automatically logs the TensorFlow Definitions, which appear in RESULTS > CONFIGURATION > HYPER PARAMETERS.
The Task configuration appears in RESULTS > CONFIGURATION > General.