app.py

# Libraries
import pandas as pd
import datetime as dt
from lifetimes import BetaGeoFitter
from lifetimes import GammaGammaFitter
from sklearn.preprocessing import MinMaxScaler

def outlier_thresholds(dataframe, variable):
    quartile1 = dataframe[variable].quantile(0.01)
    quartile3 = dataframe[variable].quantile(0.99)
    interquantile_range = quartile3 - quartile1
    up_limit = quartile3 + 1.5 * interquantile_range
    low_limit = quartile1 - 1.5 * interquantile_range
    return low_limit, up_limit

def replace_with_thresholds(dataframe, variable):
    low_limit, up_limit = outlier_thresholds(dataframe, variable)
    dataframe.loc[(dataframe[variable] < low_limit), variable] = round(low_limit,0)
    dataframe.loc[(dataframe[variable] > up_limit), variable] = round(up_limit,0)


def create_cltv_df(dataframe):

    # Data Preparation
    columns = ["order_num_total_ever_online", "order_num_total_ever_offline", "customer_value_total_ever_offline","customer_value_total_ever_online"]
    for col in columns:
        replace_with_thresholds(dataframe, col)

    dataframe["order_num_total"] = dataframe["order_num_total_ever_online"] + dataframe["order_num_total_ever_offline"]
    dataframe["customer_value_total"] = dataframe["customer_value_total_ever_offline"] + dataframe["customer_value_total_ever_online"]
    dataframe = dataframe[~(dataframe["customer_value_total"] == 0) | (dataframe["order_num_total"] == 0)]
    date_columns = dataframe.columns[dataframe.columns.str.contains("date")]
    dataframe[date_columns] = dataframe[date_columns].apply(pd.to_datetime)

    # Creating CLTV data structure
    dataframe["last_order_date"].max()  # 2021-05-30
    analysis_date = dt.datetime(2021, 6, 1)
    cltv_df = pd.DataFrame()
    cltv_df["customer_id"] = dataframe["master_id"]
    cltv_df["recency_cltv_weekly"] = ((dataframe["last_order_date"] - dataframe["first_order_date"]).astype('timedelta64[D]')) / 7
    cltv_df["T_weekly"] = ((analysis_date - dataframe["first_order_date"]).astype('timedelta64[D]')) / 7
    cltv_df["frequency"] = dataframe["order_num_total"]
    cltv_df["monetary_cltv_avg"] = dataframe["customer_value_total"] / dataframe["order_num_total"]
    cltv_df = cltv_df[(cltv_df['frequency'] > 1)]

    # BG-NBD Model
    bgf = BetaGeoFitter(penalizer_coef=0.001)
    bgf.fit(cltv_df['frequency'],
            cltv_df['recency_cltv_weekly'],
            cltv_df['T_weekly'])
    cltv_df["exp_sales_3_month"] = bgf.predict(4 * 3,
                                               cltv_df['frequency'],
                                               cltv_df['recency_cltv_weekly'],
                                               cltv_df['T_weekly'])
    cltv_df["exp_sales_6_month"] = bgf.predict(4 * 6,
                                               cltv_df['frequency'],
                                               cltv_df['recency_cltv_weekly'],
                                               cltv_df['T_weekly'])

    # # Gamma-Gamma Model
    ggf = GammaGammaFitter(penalizer_coef=0.01)
    ggf.fit(cltv_df['frequency'], cltv_df['monetary_cltv_avg'])
    cltv_df["exp_average_value"] = ggf.conditional_expected_average_profit(cltv_df['frequency'],
                                                                           cltv_df['monetary_cltv_avg'])

    # CLTV Prediction
    cltv = ggf.customer_lifetime_value(bgf,
                                       cltv_df['frequency'],
                                       cltv_df['recency_cltv_weekly'],
                                       cltv_df['T_weekly'],
                                       cltv_df['monetary_cltv_avg'],
                                       time=6,
                                       freq="W",
                                       discount_rate=0.01)
    cltv_df["cltv"] = cltv

    # CLTV segmentation
    cltv_df["cltv_segment"] = pd.qcut(cltv_df["cltv"], 4, labels=["D", "C", "B", "A"])

    return cltv_df


if __name__ == "__main__":

    df = pd.read_csv('data/flo_data_20k.csv')
    cltv_df = create_cltv_df(df)
    print(cltv_df)
    cltv_df.to_csv('output/cltv_segments.csv', index=False)