Makov Chain trong MKT đa nền tảng.py

import pandas as pd
import numpy as np
from collections import defaultdict
#------------- IMPORT DATA -----------------------------
df = pd.read_csv(r'C:\HỌC TẬP\PYTHON LEARNING\STATISTIC\DATA RESEARCH\attribution data.csv')
df = df.sort_values(['cookie', 'time'],
                    ascending=[False, True])
df['visit_order'] = df.groupby('cookie').cumcount() + 1

# gộp data để xác định thành phần chuyển đổi cuối cùng
df_paths = df.groupby('cookie')['channel'].aggregate(
    lambda x: x.unique().tolist()).reset_index()
    
df_last_interaction = df.drop_duplicates('cookie', keep='last')[['cookie', 'conversion']]
df_paths = pd.merge(df_paths, df_last_interaction, how='left', on='cookie')

df_paths['path'] = np.where(
    df_paths['conversion'] == 0,
    ['Start'] + df_paths['channel'] + ['Null'],
    ['Start'] + df_paths['channel'] + ['Conversion'])

df_paths = df_paths[['cookie', 'path']]

# import makov chain
list_of_paths = df_paths['path']
total_conversions = sum(path.count('Conversion') for path in df_paths['path'].tolist())
base_conversion_rate = total_conversions / len(list_of_paths)

def transition_states(list_of_paths):
    list_of_unique_channels = set(x for element in list_of_paths for x in element)
    transition_states = {x + '>' + y: 0 for x in list_of_unique_channels for y in list_of_unique_channels}

    for possible_state in list_of_unique_channels:
        if possible_state not in ['Conversion', 'Null']:
            for user_path in list_of_paths:
                if possible_state in user_path:
                    indices = [i for i, s in enumerate(user_path) if possible_state in s]
                    for col in indices:
                        transition_states[user_path[col] + '>' + user_path[col + 1]] += 1

    return transition_states
trans_states = transition_states(list_of_paths)

def transition_prob(trans_dict):
    list_of_unique_channels = set(x for element in list_of_paths for x in element)
    trans_prob = defaultdict(dict)
    for state in list_of_unique_channels:
        if state not in ['Conversion', 'Null']:
            counter = 0
            index = [i for i, s in enumerate(trans_dict) if state + '>' in s]
            for col in index:
                if trans_dict[list(trans_dict)[col]] > 0:
                    counter += trans_dict[list(trans_dict)[col]]
            for col in index:
                if trans_dict[list(trans_dict)[col]] > 0:
                    state_prob = float((trans_dict[list(trans_dict)[col]])) / float(counter)
                    trans_prob[list(trans_dict)[col]] = state_prob
    return trans_prob
trans_prob = transition_prob(trans_states)
def transition_matrix(list_of_paths, transition_probabilities):
    trans_matrix = pd.DataFrame()
    list_of_unique_channels = set(x for element in list_of_paths for x in element)

    for channel in list_of_unique_channels:
        trans_matrix[channel] = 0.00
        trans_matrix.loc[channel] = 0.00
        trans_matrix.loc[channel][channel] = 1.0 if channel in ['Conversion', 'Null'] else 0.0

    for key, value in transition_probabilities.items():
        origin, destination = key.split('>')
        trans_matrix.at[origin, destination] = value

    return trans_matrix


trans_matrix = transition_matrix(list_of_paths, trans_prob)