Fix csv issue in the points_to_csv method

[mattiaciollaro]

The current header argument in the np.savetxt call inside of the points_to_csv method is not compatible with the format expected from the initialize_df method.

A call to initialize_df when the argument is a pandas DataFrame constructed by reading the output of points_to_csv with pandas.read_csv will fail because of a wrong format of the header.

In particular, the issues here are the following:

1. the header of the csv file is generated with a leading pound character ('#') that comes from the default comments='#' of np.savetxt

2. the column names are prefixed with an unnecessary space in by ', '.join(self.space.keys + ['target']) inside of points_to_csv.

This pull request fixes the above two issues. I tested these changes on a notebook that I am happy to share upon request.

At the cost of making pandas an additional dependency, it may be easier to use pandas in methods such as points_to_csv and initialize_df.

[codecov-io]

Codecov Report

Merging #80 into master will not change coverage.
The diff coverage is 0%.

@@           Coverage Diff           @@
##           master      #80   +/-   ##
=======================================
  Coverage   82.75%   82.75%           
=======================================
  Files           4        4           
  Lines         290      290           
  Branches       34       34           
=======================================
  Hits          240      240           
  Misses         45       45           
  Partials        5        5

Impacted Files	Coverage Δ
bayes_opt/bayesian_optimization.py	66.08% <0%> (ø)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 44bd13c...ddf762d. Read the comment docs.

[fmfn]

Thanks for the PR. Indeed, pandas is not used in points_to_csv as to avoid the unnecessary dependence, and I would like to keep it this way.