This (inefficient) function accepts a python datetime and shows how to transform repeating features (day of the year, weekday, hour) on the numberline into continuous features, (ie days 1 and 365 are topologically close to one another at the top of a circle)

date_to_feature.py

from datetime import datetime as dt
import calendar
import numpy as np

def num_to_circle(z):
    #this functions accepts a number between 0 and 1 and projects it onto x, y coordinates on the unit circle
    x = np.cos(2*np.pi*z)
    y = np.sin(2*np.pi*z)
    return((x, y))
  

def transform_date_to_feature(y):
    #this function accepts a datetime object and converts the following to x, y coordinates on the unit circle:
    #minute of the day
    #hour of the day
    #day of the week
    #day of the year
    #month of the year

    date_ints = np.array([y.minute, y.hour+1, y.weekday()+1, int(y.strftime('%j'))+1, y.month ])
    #normalize these intergers to between 0,1, probably some off by one errors here depend on 
    if calendar.isleap(y.year):
        normed_ints = np.array([1/60.0, 1/24.0, 1/7.0, 1/366.0, 1/12.0])*date_ints
    else:
        normed_ints = np.array([1/60.0, 1/24.0, 1/7.0, 1/365.0, 1/12.0])*date_ints

    output = []
    for i in normed_ints:
        output.append(num_to_circle(i))

    return(output)
      
x = dt.now()
print(transform_date_to_feature(x))