Sandy4321

with control_campaigns as (
select
    'Control' as measurement_group
  , campaign_name as base_campaign
  , date
  , cost
  , conversions
  , conversion_value
from
  fivetran.adwords.google_campaign_performance

Sandy4321

def change_draft_target_cpa(original_campaign_tcpa, original_campaign_name, draft_campaign_id):
    # Create Portfolio for new campaign
    draft_target_cpa = round(original_campaign_tcpa * (1-0.30), 2)
    draft_portfolio_name = '{0}-Test_XYZ_Draft_Portfolio'.format(original_campaign_name)
    adwords_client = connect_to_adwords_api()
    bidding_strategy_service = adwords_client.GetService('BiddingStrategyService', version='v201809')
    portfolio_config = {
        'name': draft_portfolio_name,
        'biddingScheme': {
            'xsi_type': 'TargetCpaBiddingScheme',

Sandy4321

def create_draft():
    adwords_client = connect_to_adwords_api()
    draft_service = adwords_client.GetService('DraftService', version='v201809')
    draft = {
        'baseCampaignId': original_campaign_id,
        'draftName': {draft_name}
    }
    draft_operation = {'operator': 'ADD', 'operand': draft}
    draft = draft_service.mutate([draft_operation])['value'][0]
    draft_campaign_id = draft['draftCampaignId']

Sandy4321

def create_and_assign_budget():
    adwords_client = connect_to_adwords_api()
    
    # This bit here creates the budget
    
    budget_service = adwords_client.GetService('BudgetService', version='v201809')
    budget = {
        'name': {an appropriate name},  # Doesn't need to be unique. Budget Id is what matters for uniqueness
        'amount': {
            'microAmount': int(round({budget_amount} * 1000000))

Sandy4321

from googleads import adwords
from googleads import oauth2


def connect_to_adwords_api():
    customer_id = 'xxx-xxx-xxxx' # This is client customer ID you see in the google ads UI
    oauth2_client = oauth2.GoogleRefreshTokenClient(
        client_id={GOOGLE_CLIENT_ID},
        client_secret={GOOGLE_CLIENT_SECRET},
        refresh_token={GOOGLE_REFRESH_TOKEN}

Sandy4321

import pandas as pd
import numpy as np


raw_test_data = pd.read_csv('medium_ppc_inc_evaluate_dataset.csv')
# You can access the csv above in this link: https://drive.google.com/open?id=1IOuneJr-QFDYGsJPRjQ8ra-5C4VczeOW

raw_test_data['diff_conversions'] = raw_test_data['test_conversions'] - raw_test_data['ctl_conversions']

N = len(raw_test_data)

Sandy4321

import numpy as np
import pandas as pd

class Node:
    '''
    This class defines a node which creates a tree structure by recursively calling itself 
    whilst checking a number of ending parameters such as depth and min_leaf. It uses an exact greedy method
    to exhaustively scan every possible split point. Algorithm is based on Frieman's 2001 Gradient Boosting Machines
    
    Input

Sandy4321

import numpy as np
import pandas as pd
from math import e

class Node:
    '''
    This class defines a node which creates a tree structure by recursively calling itself 
    whilst checking a number of ending parameters such as depth and min_leaf. It uses an exact greedy method
    to exhaustively scan every possible split point. The gain metric of choice is conservation of varience.
    This is a Naive solution and does not comapre to Frieman's 2001 Gradient Boosting Machines

Sandy4321

import numpy as np
import pandas as pd
from math import e

class Node:
    
    '''
    A node object that is recursivly called within itslef to construct a regression tree. Based on Tianqi Chen's XGBoost 
    the internal gain used to find the optimal split value uses both the gradient and hessian. Also a weighted quantlie sketch 
    and optimal leaf values all follow Chen's description in "XGBoost: A Scalable Tree Boosting System" the only thing not 

Sandy4321

Data preparation

-- set mapred.max.split.size=128000000;
set hive.input.format=org.apache.hadoop.hive.ql.io.CombineHiveInputFormat;
set hive.tez.input.format=org.apache.hadoop.hive.ql.io.CombineHiveInputFormat;
set hive.mapjoin.smalltable.filesize=30000000;
-- set hive.optimize.s3.query=true;
set hive.exec.dynamic.partition.mode=nonstrict; 
set hive.optimize.sort.dynamic.partition=false;