Study different partitioning of events

[orelgueta]

Following the idea of dividing our dataset to 20% good PSF, 60% average PSF and 20% bad PSF, I calculated the confusion matrix for that configuration (regression is awesome). Results are below.
I expected to see better performance, but there is still quite a bit of confusion between the good and the average (less so with the bad). Perhaps we need to look at other configuration and also improve our regression performance further.

[TarekHC]

Oh, wow. This is very interesting! Now the intermediate event type has very good performance!

Although note now the event statistics are "broken", so the color scheme of this plot is not properly showing the "performance".

I would do what I told you in the past. Sum the events withing each "row" and divide all values in the row by that number. That way we would show % of events falling into the right/wrong box per event type (and it would not matter the statistics in each row).

[orelgueta]

Funny, I just sent you an e-mail asking if this is what you meant because I noticed the same issue =)