Implementation of a POMDP node for classification

[Hororohoruru]

This is the PR for the implementation of POMDP as an alternative strategy to output predictions.

[mesca]

Since POMDP requires to learn from from the training dataset, wouldn't it be better to implement this algorithm as a sklearn estimator?

[Hororohoruru]

The POMDP does not learn from the training dataset, like the classification pipeline. Rather, it needs a confusion matrix generated from the trained ML pipeline (which means more total calibration). Even if it has a 'fit' step, the prediction part is not similar to a sklearn estimator, so I would not go in that direction at the moment. We can always revisit the implementation if we see it is more similar to an estimator down the way.