Added new way to calculate the probability an alternative is the winner

[caser]

I added a new way to calculate the probability that a given alternative is the best. This is important for experiments with multiple alternatives, as the current z-score testing only compares the alternatives to the control and not to each other. The new feature simulates 10,000 draws from a probability distribution given each alternative's past performance, and then outputs the probability that it is the winner, given the rest of the alternatives.

[andrew]

This looks really interesting, I doubt I'll have time to properly review it until later in the week though.

One thing I did notice whilst having a quick browse was a file called changes.rtf, is that supposed to be in there, if so what's it for?

[caser]

Okay, no worries on the time.

And good catch - that was a scratch file I was using for notes as I worked. I just pushed a new commit that no longer includes the file.

Cheers.

[caser]

Hey - saw this hasn't been merged yet - would you like me to help walk you through the math & reasoning behind what I did?

[andrew]

@caser sorry for the delay, I'd like to get this merged in an released as 0.8.0, there are some merge conflicts at the moment, could you resolve those and then I can merge it in.

[caser]

No worries - will look through it this week and get back to you.

[andrew]

We've move over here: #266