Pandas rolling stats for time series non-unique data

rolling_stats.py

import pandas as pd

def nunique_rolling_time_series(data_series, step_freqency, window_size, output_name=''):
    """
    Calculate a rolling statistic of nunique of a time series. The input series has a DateTime index.
    """

    data_series = data_series.sort_index()

    min_date = data_series.index.min()
    max_date = data_series.index.max()
    date_steps = pd.date_range(min_date, max_date, freq=step_freqency, normalize=True)
    date_steps.freq = pd.tseries.offsets.Day() # Change the offset so when +1 it does it in days

    output = pd.Series(index=date_steps, name=output_name)

    for i, e in enumerate(date_steps):
        # Skip untill got a full window width
        if i < window_size:
            continue

        window_start = date_steps[i-window_size]
        window_end = e

        data_window = data_series.loc[(window_start+1):window_end] # +1 As date slice is inclusive of both dates
        n_unique = data_window.nunique()

        output[i] = n_unique
        
    return output