improves (#53) speed of train/selection/validation split

[JanBenisek]

Solves #53 by improving speed of the train/selection/validation split

• I removed the dependancy of sklearn train_test_split because it involved too many dataframes splitting and merging
• as a consequence, I dropped the option of stratify_split
  • I hope this is temporary, but this gives us a quick fix and hopefully later we can reintroduce it
• I kept the tests, they are fine (only small modification was made)
• I updated the documentation

I tested the solution's efficiency on the earnings dataset (the one we used at Data Science meetup Leuven)

• current solution:

%timeit train_selection_validation_split(data,
              target_column_name=target_column_name,
              train_prop=0.6,
              selection_prop=0.2,
              validation_prop=0.2)
# 7.29 ms ± 821 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

• proposed solution:

%timeit train_selection_validation_split(data=data, 
            train_prop=0.6, 
            selection_prop=0.2, 
            validation_prop=0.2)

# 109 µs ± 15.7 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)

And also memory consumption (using memory_profiler):

• current solution: memory usage 111.1 MiB
• proposed solution: 77.4 MiB

[sborms]

Looks good. Nice speed and memory consumption improvements! Also the code simplification is always a plus.