meraldo-aliz

params_mean = pm.summary(trace_trunc, var_names = ["r", "alpha", "a", "b"])["mean"]
params_mean

meraldo-aliz

from pymc3.math import log, exp, where
import pymc3 as pm
import numpy as np


# We use the "calibration" portion of the dataset to train the model
N = rfm_cal_holdout.shape[0] # number of customers
x = rfm_cal_holdout['frequency_cal'].values # repeat purchase frequency
t_x = rfm_cal_holdout['recency_cal'].values # recency
T = rfm_cal_holdout['T_cal'].values # time since first purchase (T)

meraldo-aliz

from sklearn.metrics import mean_squared_error

RMSE = mean_squared_error(y_true = rfm_cal_holdout["n_transactions_holdout_real"],
                          y_pred = rfm_cal_holdout["n_transactions_holdout_pred"],
                          squared = False)

RMSE # = 1.3536793286521

meraldo-aliz

# the real number of transactions in the observation period, which equals frequency_holdout + 1
rfm_cal_holdout["n_transactions_holdout_real"]  = rfm_cal_holdout["frequency_holdout"] + 1

# the predicted number of transactions in the next 26 weeks (length of the observation period)
rfm_cal_holdout["n_transactions_holdout_pred"] = bgf.predict(t=26, 
                                                    frequency=rfm_cal_holdout['frequency_cal'], 
                                                    recency=rfm_cal_holdout['recency_cal'], 
                                                    T=rfm_cal_holdout['T_cal'])

# comparison of the real and predicted transactions

meraldo-aliz

VALUE_10_PRED_THRESHOLD = 10

# filtering for high-value customers
rfm_cal_holdout.loc[rfm_cal_holdout["value_10_pred"]>VALUE_10_PRED_THRESHOLD, ["value_10_pred"]]

meraldo-aliz

import seaborn as sns 

fig, ax = plt.subplots(figsize = (12, 7))
ax = sns.histplot(rfm_cal_holdout["value_10_pred"], 
                  kde=False, 
                  binwidth = 2)
ax.set_title(f'Customer value histogram')
ax.set_xlabel(r'Customer value estimate for 10 periods ($)')
ax.set_ylabel(r'Number of customers in each bin')
ax.set_xlim(-2,20)

meraldo-aliz

rfm_cal_holdout["value_10_pred"].describe()

meraldo-aliz

# the predicted number of transactions in the next 10 weeks
rfm_cal_holdout["n_transactions_10_pred"] = bgf.predict(t=10, 
                                                        frequency=rfm_cal_holdout['frequency_cal'], 
                                                        recency=rfm_cal_holdout['recency_cal'], 
                                                        T=rfm_cal_holdout['T_cal'])

# the probability of being alive
rfm_cal_holdout["alive_prob"] = bgf.conditional_probability_alive(frequency=rfm_cal_holdout['frequency_cal'], 

meraldo-aliz

from lifetimes.plotting import plot_probability_alive_matrix

_ = plot_probability_alive_matrix(bgf)

meraldo-aliz

from lifetimes.plotting import plot_frequency_recency_matrix

_ = plot_frequency_recency_matrix(bgf)