Rework Non i.i.d. data notebook

[ArturoAmorQ]

The plot showcasing the use of a ShuffleSplit strategy Non i.i.d. data notebook changed in a recent version of the pandas plotting utility:

In previous versions:

Now:

We could use the opportunity to rework this whole notebook, as mentioned in #784 (review):

• Not use groups with TimeSeriesSplit (it's currently rising a UserWarning)
• Use a more realistic dataset (optional)
• Give interpretation to resulting R2 and MSE scores:
  • results are above or below chance level?
  • they are over optimistic when not evaluated properly
  • predictions are not realistic, when using the current dataset, a simple DecisionTreeRegressor can foresee a sudden drop in quotes
• Mention the actual good practices for modeling, e.g. aligning the test size of TimeSeriesSplit with the forecasting task
• In general give more focus to the use of TimeSeriesSplit

[glemaitre]

I wanted to report this issue because I just saw the plot. A hot fix for the plot is to make sure that the data point are ordered by increasing date.