Error in "Handling datetime features with the DatetimeEncoder" Example + Minor Corrections

[semyonbok]

The example juxtaposes the performance of three regression models that use different configurations of DatetimeEncoder. These lines highlight that the one with spline feature pre-processing is better:

skrub/examples/03_datetime_encoder.py

Lines 221 to 227 in 00a6d70

	print(f"Base transformer - Mean RMSE : {-score_base.mean():.2f}")
	print(f"Transformer with weekday - Mean RMSE : {-score_weekday.mean():.2f}")
	print(f"Transformer with periodic features - Mean RMSE : {-score_periodic.mean():.2f}")
	###############################################################################
	# As expected for linear models, introducing the periodic features improved
	# the RMSE by a noticeable amount.

Please notice that spline periodic encoding actually yields significantly worse RMSE. Its performance on the first two splits is preposterous, as visualized below. The DataViz also shows RMSE on entire test set, as well as on its last 96 entries: the spline pipeline keeps under-performing.

Also, a couple of minor corrections should be done:

• just "don't want to" is meant, and resolution is not explicitly set to "hour" in the preceding code example:
  

  skrub/examples/03_datetime_encoder.py

  Lines 89 to 92 in 00a6d70

  	# During the instantiation of the |DatetimeEncoder|, we specify that we want
  	# don't want to extract features with a resolution finer than hours. This is
  	# because we don't want to extract minutes, seconds and lower units, as they
  	# are unimportant here.

• slight improvements to prediction plotting can be made: