Fix the Hyperband configuration formula

[AlexandreAbraham]

As highlighted in my blogpost [1], the formula used to computer the number of configurations for the successive halving step in hyperband deviates from the paper. In fact, there is an additional rounding operation on the s_max / (s+1) factor. Everything is explained in the blog post.

I also do not know why you are using an int() operator to round the value of n0 because it is originally a ceil operation. I suppose that this is why you need an additional max() in the computation of ns. The code is a bit intricated to me so I guess you can better tell than me. Do not hesitate to tell me if I can be of help or I am wrong.

[1] https://medium.com/data-from-the-trenches/a-slightly-better-budget-allocation-for-hyperband-bbd45af14481

[codecov-io]

Codecov Report

Merging #89 (0823035) into master (841db4b) will decrease coverage by 0.37%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master      #89      +/-   ##
==========================================
- Coverage   65.16%   64.78%   -0.38%     
==========================================
  Files          28       28              
  Lines        1846     1846              
==========================================
- Hits         1203     1196       -7     
- Misses        643      650       +7