渠道归因（一）传统渠道归因

# pip install Cython
# pip install ChannelAttribution

import pandas as pd
from ChannelAttribution import *

Looking for attribution at path level? Try markov_model_local_api() or ask for ChannelAttributionPro at info@channelattribution.io! Visit https://channelattribution.io for more information.
Version: 2.1.3

Data = pd.read_csv('./Markov/Data.csv',sep=";")
Data.head() 

# 传统启发式算法计算各渠道转化次数和转化收益
# first_touch：首次点击模型；last_touch：末尾点击模型；linear_touch：线性（平均）点击模型
heuristic_models(Data,"path","total_conversions", var_value='total_conversion_value')

# 自定义位置归因函数，设置首尾权重各50%
def posAttribution(df, path, conversions):
    '''
    df:数据集
    path：用户旅程路径
    conversions：成功转化次数
    '''
    df=df.copy()
    df['leng']=df[path].map(lambda x: len(x.split(">")))
    df['first_touch']=df[path].map(lambda x: [x.strip() for x in x.split(">")][0])
    df['last_touch']=df[path].map(lambda x: [x.strip() for x in x.split(">")][-1])

    df['first_touch_conversions']=df[['leng',conversions]].apply(
        lambda x: x[conversions]*1 if x['leng']==1 else x[conversions]*0.5, axis=1)
    df['last_touch_conversions']=df[['leng',conversions]].apply(
        lambda x: 0 if x['leng']==1 else x[conversions]*0.5, axis=1)

    df1=df.groupby(['first_touch'])['first_touch_conversions'].sum()
    df2=df.groupby(['last_touch'])['last_touch_conversions'].sum()
    df_f=pd.merge(df1, df2, left_index=True, right_index=True, how = 'outer').reset_index()
    df_f['first_last_touch_conversions']=df_f['first_touch_conversions']+df_f['last_touch_conversions']
    df_f.drop(['first_touch_conversions','last_touch_conversions'], axis=1, inplace=True)
    
    return df_f

# 位置归因（首尾归因）
posAttribution(Data,"path","total_conversions")

from sklearn.preprocessing import StandardScaler, MinMaxScaler 
from pandas.api.types import is_string_dtype, is_numeric_dtype
import warnings
import pandas as pd

# 读取数据 数据为自定义的
df=pd.read_excel('datas.xlsx')
df.head()

# 自定义时间衰减归因 衰减权重为半衰。例如1->2，则1的权重是66.6%；1->2->3，则1的权重是50%，2的权重是33.3%
def lapsAttribution(df, path, conversions, last_time_lapse):
    def scum(l):
        s=0
        for i in range(1, len(l)+1):
            s+=i
        return s
    
    def lw(l):
        lw=[]
        n=scum(l)
        for i in range(1, len(l)+1):
            lw.append(1/n*i)
        return lw

    df['last_time_sp']=df[last_time_lapse].map(lambda x: str(x).split('>'))
    df['path_sp']=df[path].map(lambda x: [x.strip() for x in x.split(">")])
    df['lts_w']=df['last_time_sp'].map(lambda x:lw(x))
    df['conversions_sp']=df[[conversions,'lts_w']].apply(
            lambda x: [i*x[conversions] for i in x['lts_w']], axis=1)
    
    laps_touch=[]
    conversions_all=[]
    for i in df[['path_sp', 'conversions_sp']].values:
        for m in zip(i[0],i[1]):
            laps_touch.append(m[0])
            conversions_all.append(m[1])

    df_temp=pd.DataFrame({'laps_touch':laps_touch, 'laps_touch_conversions':conversions_all})
    df_f=df_temp.groupby(['laps_touch'])['laps_touch_conversions'].sum().reset_index()
    
    return df_f

lapsAttribution(df, 'path', 'total_conversions', 'last_time_lapse')