Washover

This notebook shows how to use the washover module. In particular, it shows how to apply a ConstantWashover object with a 30min time delta.

In [1]:

## Import Libraries

import numpy as np
import pandas as pd
from datetime import datetime, timedelta

from cluster_experiments import ConstantWashover

## Import Libraries import numpy as np import pandas as pd from datetime import datetime, timedelta from cluster_experiments import ConstantWashover

In [2]:

## Generate a dummy dataset of 4 time periods of 2 hours with 10 orders each

np.random.seed(42)

num_rows = 5

def random_timestamp(start_time, end_time):
    time_delta = end_time - start_time
    random_seconds = np.random.randint(0, time_delta.total_seconds())
    return start_time + timedelta(seconds=random_seconds)

def generate_data(start_time, end_time, treatment):
    data = {
        'order_id': np.random.randint(10**9, 10**10, size=num_rows),
        'city_code': 'VAL',
        'activation_time_local': [random_timestamp(start_time, end_time) for _ in range(num_rows)],
        'bin_start_time_local': start_time,
        'treatment': treatment
    }
    return pd.DataFrame(data)

start_times = [datetime(2024, 1, 22, 9, 0), datetime(2024, 1, 22, 11, 0),
               datetime(2024, 1, 22, 13, 0), datetime(2024, 1, 22, 15, 0)]

treatments = ['control', 'variation', 'variation', 'control']

dataframes = [generate_data(start, start + timedelta(hours=2), treatment) for start, treatment in zip(start_times, treatments)]

df = pd.concat(dataframes).sort_values(by='activation_time_local').reset_index(drop=True)

df.head(20)

## Generate a dummy dataset of 4 time periods of 2 hours with 10 orders each np.random.seed(42) num_rows = 5 def random_timestamp(start_time, end_time): time_delta = end_time - start_time random_seconds = np.random.randint(0, time_delta.total_seconds()) return start_time + timedelta(seconds=random_seconds) def generate_data(start_time, end_time, treatment): data = { 'order_id': np.random.randint(10**9, 10**10, size=num_rows), 'city_code': 'VAL', 'activation_time_local': [random_timestamp(start_time, end_time) for _ in range(num_rows)], 'bin_start_time_local': start_time, 'treatment': treatment } return pd.DataFrame(data) start_times = [datetime(2024, 1, 22, 9, 0), datetime(2024, 1, 22, 11, 0), datetime(2024, 1, 22, 13, 0), datetime(2024, 1, 22, 15, 0)] treatments = ['control', 'variation', 'variation', 'control'] dataframes = [generate_data(start, start + timedelta(hours=2), treatment) for start, treatment in zip(start_times, treatments)] df = pd.concat(dataframes).sort_values(by='activation_time_local').reset_index(drop=True) df.head(20)

Out[2]:

	order_id	city_code	activation_time_local	bin_start_time_local	treatment
0	8395928407	VAL	2024-01-22 09:19:44	2024-01-22 09:00:00	control
1	5298312065	VAL	2024-01-22 10:15:55	2024-01-22 09:00:00	control
2	3563451924	VAL	2024-01-22 10:24:11	2024-01-22 09:00:00	control
3	1787846414	VAL	2024-01-22 10:28:31	2024-01-22 09:00:00	control
4	5537253172	VAL	2024-01-22 10:47:00	2024-01-22 09:00:00	control
5	2855189739	VAL	2024-01-22 11:21:07	2024-01-22 11:00:00	variation
6	8667272366	VAL	2024-01-22 11:25:28	2024-01-22 11:00:00	variation
7	7548779029	VAL	2024-01-22 11:31:39	2024-01-22 11:00:00	variation
8	6152559666	VAL	2024-01-22 11:53:22	2024-01-22 11:00:00	variation
9	2250819632	VAL	2024-01-22 12:56:15	2024-01-22 11:00:00	variation
10	8767007473	VAL	2024-01-22 13:22:43	2024-01-22 13:00:00	variation
11	3609385266	VAL	2024-01-22 13:34:01	2024-01-22 13:00:00	variation
12	9370399619	VAL	2024-01-22 13:43:32	2024-01-22 13:00:00	variation
13	1279394470	VAL	2024-01-22 13:47:04	2024-01-22 13:00:00	variation
14	5147358011	VAL	2024-01-22 14:57:21	2024-01-22 13:00:00	variation
15	7643070057	VAL	2024-01-22 15:02:41	2024-01-22 15:00:00	control
16	5164334270	VAL	2024-01-22 16:11:37	2024-01-22 15:00:00	control
17	7528642437	VAL	2024-01-22 16:34:35	2024-01-22 15:00:00	control
18	2111451555	VAL	2024-01-22 16:49:05	2024-01-22 15:00:00	control
19	8140478823	VAL	2024-01-22 16:54:33	2024-01-22 15:00:00	control

In [3]:

## Define washover with 30 min duration
washover = ConstantWashover(washover_time_delta=timedelta(minutes=30))

## Apply washover to the dataframe, the orders with activation time within the first 30 minutes after every change in the treatment column, clustering by city and 2h time bin, will be dropped
df_analysis_washover = washover.washover(
    df=df,
    truncated_time_col='bin_start_time_local',
    treatment_col='treatment',
    cluster_cols=['city_code','bin_start_time_local'],
    original_time_col='activation_time_local',
)

## Define washover with 30 min duration washover = ConstantWashover(washover_time_delta=timedelta(minutes=30)) ## Apply washover to the dataframe, the orders with activation time within the first 30 minutes after every change in the treatment column, clustering by city and 2h time bin, will be dropped df_analysis_washover = washover.washover( df=df, truncated_time_col='bin_start_time_local', treatment_col='treatment', cluster_cols=['city_code','bin_start_time_local'], original_time_col='activation_time_local', )

In [4]:

## Show the rows that have been dropped
anti_joined_df = df.merge(df_analysis_washover['order_id'], how='left', indicator=True, on='order_id')
anti_joined_df = anti_joined_df[anti_joined_df['_merge'] == 'left_only'].drop(columns=['_merge'])

anti_joined_df.head(10)

## Show the rows that have been dropped anti_joined_df = df.merge(df_analysis_washover['order_id'], how='left', indicator=True, on='order_id') anti_joined_df = anti_joined_df[anti_joined_df['_merge'] == 'left_only'].drop(columns=['_merge']) anti_joined_df.head(10)

Out[4]:

	order_id	city_code	activation_time_local	bin_start_time_local	treatment
0	8395928407	VAL	2024-01-22 09:19:44	2024-01-22 09:00:00	control
5	2855189739	VAL	2024-01-22 11:21:07	2024-01-22 11:00:00	variation
6	8667272366	VAL	2024-01-22 11:25:28	2024-01-22 11:00:00	variation
15	7643070057	VAL	2024-01-22 15:02:41	2024-01-22 15:00:00	control

In [5]:

## Check DF shapes
print('df:', df.shape)
print('df_analysis_washover:', df_analysis_washover.shape)

## Check DF shapes print('df:', df.shape) print('df_analysis_washover:', df_analysis_washover.shape)

df: (20, 5)
df_analysis_washover: (16, 5)