The thing

Before anything else, this post is just my rewriting for memory and comprehension of the following amazing blogpost: http://blog.davidkaleko.com/feature-engineering-cyclical-features.html. All credits go to this blogpost and his creator.

Usually, we handle feature like month and hour with dummy variable (11/12 for months and 23/24 for hours). It works, but it is far from perfect because 23h will be considered far away from 00h when in reality those two hours should be as close as 00h and 01h. But lately, I found (on the internet, not by myself) a really good way to handle this type of cyclical features to take into account the cyclical property.

The solution is to map those cyclicals variable in a circle, computing the x and y component of the 2-dimension circle using cos and sin trigonometrics functions. It is very obvious for hours, because it will be similar to what can be found on an old-fashion clock.

In code

Library & constant definition
    
    import numpy as np

    # For hours
    df['hr_sin'] = np.sin(df.hr*(2.*np.pi/24))
    df['hr_cos'] = np.cos(df.hr*(2.*np.pi/24))

    # For months
    df['mnth_sin'] = np.sin((df.mnth-1)*(2.*np.pi/12))
    df['mnth_cos'] = np.cos((df.mnth-1)*(2.*np.pi/12))

Note that in the previous code, we took month minus 1 to map values from 0 to 11 for convenience.