Implement metrics

[rswc]

We agreed to use the following:

• Gmean(Recall)
• Macro F1
• Attack Detection Rate
• False Alarm Rate
• (possibly others as well)

We'd prefer them to be evaluated in rolling-window mode. Based on a quick glance at River's documentation, this will most likely mean having to write that ourselves.

Can we implement that via a wrapper class over the metrics offered by River?

[rswc]

Good news: the rolling-window functionality is available in river under utils.Rolling

See also:

• metrics.GeometricMean
• metrics.MacroF1
• if by "Attack Detection Rate" we mean $\frac{TP}{P}$ (where $P = TP+FN$ is the number of actually positive examples), recall is obviously also available — we just need to ensure the class is "cast" into the binary attack/benign
• if by "False Alarm Rate" we mean $\frac{FP}{PP}$ (where $PP$ is "predicted positive"), that would be $1 - precision$?