Check that combine_surprise_normalised isn't cheatig by using the observed rather than calculated surprises.

[Kazjon]

While looking at something else I happened to notice that line 81 of eval_personalised is pulling from reader.cooccurrence, not from the estimated cooc matrix. I have to think about this a bit more, but wouldn't that mean it's cheating?

I'll take another look at this when I get a chance.