Determine time stamp conversion for linear regression

[friedrichknuth]

The method of time stamp conversion to ordinal or decimal date slightly changes the final results

e.g.

X_train = np.array([utils.date_time_to_decyear(i) for i in dem_dates]) -> array([..., 2000.54440242, ...])

vs
X_train = np.array(matplotlib.dates.date2num(dem_dates)) -> array([..., 11156.25128592, ...])

Using matplotlib.dates.date2num requires a conversion back to m/yr from m/day, e.g. slope *= 365.25.

Need to determine which option is best, and why, or if other options for time stamp conversion prior to linear regression are preferred.

[friedrichknuth]

Realted to #52