Compute and report quality metrics for models in covlmc about mixvlmc HOT 4 CLOSED

Two commits provide the first steps toward implementing this feature:

• commit 6f3c319 adds a way to report occurrences of contexts in a "local" way, that is to count only occurrences that do not corresponds to occurrences of a longer context. Those occurrences are "predictive" ones: the corresponding context will be use as reference to compute conditional probabilities of the following state.
• commit 6c88070 adds a metrics generic function (implemented for vlmc) that reports global predictive metrics. Accuracy is reported for convenience, but the main important values are the confusion matrix (which can then be used to compute other metrics such as precision, recall, etc.) and the AUC.

from mixvlmc.

Commits abe5ec9 and 8300dd7 provide additional steps toward the implementation this feature. They implement predictive metrics calculation during model tuning in the covlmc construction. Predictive metrics (mainly the confusion matrix and the AUC) are stored with the model. They can be extracted using contexts.covlmc.

from mixvlmc.

Commits e4a9a2c and 307c662 unify metrics calculation and implement global metrics for covlmc (via metrics.covlmc).

from mixvlmc.

Commit e2d56a0 adds the final part of this feature by providing local metrics for vlmc via contexts.vlmc.

from mixvlmc.