Skip to main content

Hidden Touch Attribution model

Import libraries​

from ChannelAttributionPro import *

Set the password​

password="mypassword"

Load data​

Data = pd.read_csv("https://app.channelattribution.net/data/Data_ata.csv")

print(Data.to_string())

Define list of channels you are interested in performing attribution​

channels=["facebook_impressions","tv_impressions","google_adv_clicks",
          "direct_searches","youtube_impressions","linkedin_impressions"]

Define the global conversion rate for the use case or set it to None if not known​

glob_conv_rate=None

Define the conversion rates, one for each channel if known, or set it to None otherwise​

L_conv_rate=None

Define dictionary of weights (between 0 and 1), one for each channel (suggestion: use weight=1 for clicks)​

L_weight=dict()
L_weight["facebook_impressions"]=0.01
L_weight["tv_impressions"]=0.01
L_weight["google_adv_clicks"]=1
L_weight["direct_searches"]=1
L_weight["youtube_impressions"]=0.01
L_weight["linkedin_impressions"]=0.01

Define the how many units of time are long the customer journeys on average for the use case considered​

mean_cj_length=1

Define the adstock rates (between 0 and 1), one for each channels​

L_ads_rate=dict()
L_ads_rate["facebook_impressions"]=0
L_ads_rate["tv_impressions"]=0.40
L_ads_rate["google_adv_clicks"]=0
L_ads_rate["direct_searches"]=0
L_ads_rate["youtube_impressions"]=0
L_ads_rate["linkedin_impressions"]=0

Run the model​

we set nsim=1 for speed reasons but at least nsim=10 is suggested​

res=hta_model(Data, var_time="week", var_conv="conversions", channels=channels, 
              var_value=None, glob_conv_rate=glob_conv_rate, 
              L_conv_rate=L_conv_rate, L_weight=L_weight, mean_cj_length=mean_cj_length, 
              L_ads_rate=L_ads_rate, verbose=True, nsim=1, password=password)

print(res.round(2).to_string())

Plot results​

import matplotlib.pyplot as plt 
fig,ax = plt.subplots(figsize=(15,6))
#res.set_index('week',inplace=True)
res.index=range(1,len(res)+1)
del res['week']
res.plot(kind='bar',ax=ax,stacked=True)
plt.ylabel('conversions')
plt.show()