Feat/use median in calculating recall cost, forget cost and learn cost

[L-M-Sherlock]

No description provided.

[Expertium]

Are zeros removed? As user1823 said, we should remove all values that are precisely equal to 0.

[L-M-Sherlock]

Are zeros removed? As user1823 said, we should remove all values that are precisely equal to 0.

Should these revlogs be removed in training?

[Expertium]

No, I meant that if review time=0 ms, it should be excluded from the median calculation. Though it might be a good idea to exclude them from everything.
@user1823 thoughts?

[user1823]

Though it might be a good idea to exclude them from everything.

I don't think that we should filter them during training the parameters unless someone gives us a specific scenario where they cause a problem.

[L-M-Sherlock]

Now zero values have been removed in calculating the cost.

[user1823]

I can't verify the accuracy of the code but broadly it looks good to me.

[user1823]

Wait a minute, I couldn't find anything that removes the reviews with time > 20 min.

[Expertium]

I couldn't find anything that removes the reviews with time > 20 min.

Yeah, those should be excluded.

[L-M-Sherlock]

Thanks for reviews!