[MRG] EHN new metrics

[glemaitre]

Reference Issue

Address issue #152

What does this implement/fix? Explain your changes.

There some metrics which are not implemented in scikit-learn but that are usually used for the evaluation on imbalanced set:

• sensitivity + specificity
• geometric mean
• generalized index balanced accuracy
• F measure <- should be already in sklearn

There already some implementation there which could be put together to fit the scikit-learn API

There is a need for:

• a classification report for imbalanced data

Any other comments?

[coveralls]

Coverage decreased (-0.2%) to 98.66% when pulling 2085360 on glemaitre:metrics into db45249 on scikit-learn-contrib:master.

[coveralls]

Coverage decreased (-0.03%) to 98.785% when pulling 0eddca6 on glemaitre:metrics into db45249 on scikit-learn-contrib:master.

[coveralls]

Coverage decreased (-0.08%) to 98.734% when pulling ac3d0de on glemaitre:metrics into db45249 on scikit-learn-contrib:master.

[coveralls]

Coverage decreased (-0.03%) to 98.786% when pulling e51afd3 on glemaitre:metrics into db45249 on scikit-learn-contrib:master.

[coveralls]

Coverage decreased (-0.5%) to 98.304% when pulling 965c5a1 on glemaitre:metrics into db45249 on scikit-learn-contrib:master.

[coveralls]

Coverage decreased (-0.1%) to 98.688% when pulling c8cb6d6 on glemaitre:metrics into db45249 on scikit-learn-contrib:master.

[glemaitre]

@chkoar I implemented the IBA metric which can take any scoring function as input.
I draft the method but it could be nice to have a second point-of-view on that one,

[coveralls]

Coverage decreased (-0.4%) to 98.407% when pulling 4e3f296 on glemaitre:metrics into db45249 on scikit-learn-contrib:master.

[glemaitre]

The problem is that all the scoring method do not have the same arguments

[glemaitre]

and I need some specific argument for the sensitivity and specificity.
Somehow this is how they build the scorer

[coveralls]

Coverage decreased (-0.1%) to 98.933% when pulling 40c3955 on glemaitre:metrics into b45a3e4 on scikit-learn-contrib:master.

[coveralls]

Coverage decreased (-0.1%) to 98.933% when pulling adcd811 on glemaitre:metrics into 305025b on scikit-learn-contrib:master.

[coveralls]

Coverage decreased (-0.1%) to 98.933% when pulling adcd811 on glemaitre:metrics into 305025b on scikit-learn-contrib:master.

[coveralls]

Coverage decreased (-0.1%) to 98.933% when pulling 87fa58c on glemaitre:metrics into fc3f69e on scikit-learn-contrib:master.

[coveralls]

Coverage decreased (-0.1%) to 98.944% when pulling f6d051a on glemaitre:metrics into 75addb8 on scikit-learn-contrib:master.

[coveralls]

Coverage decreased (-0.1%) to 98.944% when pulling 49a154f on glemaitre:metrics into 75addb8 on scikit-learn-contrib:master.

[coveralls]

Coverage decreased (-0.1%) to 98.944% when pulling 28ab278 on glemaitre:metrics into 75addb8 on scikit-learn-contrib:master.