Implement hyper-parameter saving when lunching new training loop.

[marued]

Adding code to TensorFlow to support MLFlow. Look at what TensorBoard already provide

[sgaut023]

@YanCote To save the experiments, we need to do:

import mlflow mlflow.set_experiment(‘NOM DE L’EXPERIENCE’)

to start an experiment

with mlflow.start_run():

log tes params

    mlflow.log_param(‘pca_n’,pca_n_components)
   mlflow.log_metric(‘accuracy’,metric)