Add Callbacks for logging and/or performing custom operations after every generation

[weixuanfu]

Similar to Keras's “callbacks”— we need add a class that can access the training state after every training iteration for better logging and/or performing custom operations after every generation during TPOT optimization. This class is related to a few features in current version (0.9.2) of TPOT (e.g. early stopping and verbosity in general).

Related issue #667

[rhiever]

Looking forward to this in TPOT. I think it will clean up the interface quite a bit.

[MarkusShepherd]

This would be a great feature, e.g., for MLflow to track metrics across generations. What would be required to implement this?

[agi1512]

Still looking forward for this feature to be implemented.

[kenimou]

This will be great to have. Would like to integrate with mlflow if possible so that we have more visibility into what has been done, etc.