probability normalization in exp4.p

[ianlini]

https://github.com/ntucllab/striatum/blob/master/striatum/bandit/exp4p.py#L102
This normalization is not in the paper. Is this correct?

[ianlini]

I think this only transform query_vector to be np.ndarray?

[taweihuang]

sum(query_vector) should be 1 form the theoretical aspect, but in python there may be some numerical errors so that the sum(query_vector) will be slightly larger or smaller than 1. So I further add this line to make sure that the sum of query_vector is 1.

[ianlini]

It transform query_vector to ndarray because every values in query_vector is np.float64, so that division make query_vector to be ndarray.
It's quite unexpected.

[ianlini]

and why is it called query_vector?
I think action_probability or something like this is better.