Hyperparameter variability

[AlexOlza]

For hyperparameter tuning of the ensemble models, we define a hyperparameter space and train 50 models at random within that space. Those 50 models (i.e. the subspace) are evaluated with cross-validation, and we select the one with the highest accuracy.

The chosen subspace depends on the random seed. Therefore, the final model does too.

TASK: Vary the random seed to explore different subspaces, obtaining a set of "final models". Analyze the distribution of the metrics of interest.

Expected code:

• [x] A new experiment file, that receives the seed via sys args - Check out argparse module
• [x] Modify the algorithms and the main programs to use this seed
• [x] Modify the main programs to accept the model name via argparse. The name will include the random seed used.
• [x] Analysis tools: plots and tables of the distribution of metrics, and maybe histograms of the selected hyperparameters

[AlexOlza]

We need two different seeds for undersampling and hyperparameter tuning!!!