Script 721: Profit Maximising Campaign Target (Trafficking)

Purpose

Python script that analyzes campaign-level forecasts to identify profit maximizing targets and generates a bulk sheet to update targets.

To Elaborate

This Python script solves the problem of determining profit maximizing targets for campaigns in a digital advertising platform. It analyzes campaign-level forecasts and applies business rules to identify the optimal targets for different bidding strategies (TargetCPA, TargetROAS, MaximizeConversions, MaximizeConversionValue). The script then generates a bulk sheet that can be used to update the targets in the advertising platform.

The key business rules and steps involved in the script are as follows:

Load the campaign data from a data source (inputDf).
Clean up the data by removing unnecessary columns and filtering based on specific criteria.
Calculate the optimal targets for each campaign based on the bidding strategy and other factors.
Apply damping to the ideal targets based on the current target and damping percentage.
Determine the expected and acceptable targets based on the optimal targets and a tolerance level.
Compare the optimal targets with the current targets to identify any changes.
Select the campaigns with changed targets and create a new dataframe (outputDf) with the relevant columns.
Rename the columns in the output dataframe to match the format required for the bulk sheet.
Output the final dataframe to a CSV file for further processing or analysis.

The script also includes some additional steps for local development and debugging purposes, such as loading data from a pickle file and printing debug information.

Note: The script uses certain user-changeable parameters, such as the tolerance level for target deviation and the target mode (Traffic or Preview). These parameters can be adjusted to customize the analysis and target generation process.

Vitals

Script ID : 721
Client ID / Customer ID: 1306927569 / 60270325
Action Type: Bulk Upload
Item Changed: Campaign
Output Columns: Account, Campaign, Publisher Target CPA, Publisher Target ROAS
Linked Datasource: FTP/Email Feed
Reference Datasource: None
Owner: Joe Southin (jsouthin@marinsoftware.com)
Created by Joe Southin on 2024-02-29 12:40
Last Updated by Grégory Pantaine on 2024-05-01 09:09

> See it in Action

Python Code

##
## name: Profit Maximising Campaign Target Trafficking
## description:
##  * Analyses campaign-level forecasts to identify profit maximising targets.  Generates bulk sheet to update targets
##  * Supports Google Smart Bidding targets only (TargetCPA, TargetROAS, MaximizeConversions, MaximizeConversionValue)
## 
## author: Joe Southin
## created: 2024-02-24
## updated: 2024-02-29  

##### Configurable Param #####
# Define tolerance of deviation from expected target
TARGET_TOLERANCE = 0.02
VAL_LIMIT_FOLDERS = [778828, 778910]
VAL_TARGET_MODE = 'Preview'
##############################


########### START - Local Mode Config ###########
# Step 1: Uncomment download_preview_input flag and run Preview successfully with the Datasources you want
download_preview_input=False
# Step 2: In MarinOne, go to Scripts -> Preview -> Logs, download 'dataSourceDict' pickle file, and update pickle_path below
pickle_path = '/Users/jsouthin/Downloads/datasource_dict_1709114229331.pkl'
# Step 3: Copy this script into local IDE with Python virtual env loaded with pandas and numpy.
# Step 4: Run locally with below code to init dataSourceDict

# determine if code is running on server or locally
def is_executing_on_server():
    try:
        # Attempt to access a known restricted builtin
        dict_items = dataSourceDict.items()
        return True
    except NameError:
        # NameError: dataSourceDict object is missing (indicating not on server)
        return False

local_dev = False

if is_executing_on_server():
    print("Code is executing on server. Skip init.")
elif len(pickle_path) > 3:
    print("Code is NOT executing on server. Doing init.")
    local_dev = True
    # load dataSourceDict via pickled file
    import pickle
    dataSourceDict = pickle.load(open(pickle_path, 'rb'))

    # print shape and first 5 rows for each entry in dataSourceDict
    for key, value in dataSourceDict.items():
        print(f"Shape of dataSourceDict[{key}]: {value.shape}")
        # print(f"First 5 rows of dataSourceDict[{key}]:\n{value.head(5)}")

    # set outputDf same as inputDf
    inputDf = dataSourceDict["1"]
    outputDf = inputDf.copy()

    # setup timezone
    import datetime
    # Chicago Timezone is GMT-5. Adjust as needed.
    CLIENT_TIMEZONE = datetime.timezone(datetime.timedelta(hours=0))

    # import pandas
    import pandas as pd
    import numpy as np

    # Printing out the version of Python, Pandas and Numpy
    # import sys
    # python_version = sys.version
    # pandas_version = pd.__version__
    # numpy_version = np.__version__

    # print(f"python version: {python_version}")
    # print(f"pandas version: {pandas_version}")
    # print(f"numpy version: {numpy_version}")

    # other imports
    import re
    import urllib

    # import Marin util functions
    from marin_scripts_utils import tableize, select_changed
else:
    print("Running locally but no pickle path defined. dataSourceDict not loaded.")
    exit(1)
########### END - Local Mode Setup ###########

today = datetime.datetime.now(CLIENT_TIMEZONE).date()

# primary data source and columns
inputDf = dataSourceDict["1"]
FEED_COL_ACCOUNT = 'Account'
FEED_COL_CAMPAIGN = 'Campaign'
FEED_COL_DAILY_BUDGET = 'daily_budget'
FEED_COL_BUDGET_TYPE = 'budget_type'
FEED_COL_CAMPAIGN_ID = 'campaign_id'
FEED_COL_CAMPAIGN_OVERRIDE = 'campaign_override'
FEED_COL_BIDDING_STRATEGY_OLD_TYPE = 'bidding_strategy_old_type'
FEED_COL_TARGET_CPA = 'target_cpa'
FEED_COL_TARGET_ROAS = 'target_roas'
FEED_COL_IS_PORTFOLIO = 'is_portfolio'
FEED_COL_IS_SHARED_BUDGET = 'is_shared_budget'
FEED_COL_FOLDER_ID = 'folder_id'
FEED_COL_START_DATE_TIME_ID = 'start_date_time_id'
FEED_COL_END_DATE_TIME_ID = 'end_date_time_id'
FEED_COL_OBJECTIVE_UNIT = 'objective_unit'
FEED_COL_CONSTRAINT_UNIT = 'constraint_unit'
FEED_COL_CONSTRAINT_VALUE = 'constraint_value'
FEED_COL_BID_AND_TARGET_DAMPING = 'bid_and_target_damping'
FEED_COL_BUDGET_DAMPING = 'budget_damping'
FEED_COL_BUDGET_CAP = 'budget_cap'
FEED_COL_FOLDER_BID_TRAFFICKING = 'folder_bid_trafficking'
FEED_COL_PUBLISHER_BUDGET_TRAFFICKING = 'publisher_budget_trafficking'
FEED_COL_PUBLISHER_BID_TRAFFICKING = 'publisher_bid_trafficking'
FEED_COL_BOOST_PERCENT = 'boost_percent'
FEED_COL_ABS_BOOST_PERCENT = 'abs_boost_percent'
FEED_COL_DIRECTION = 'direction'
FEED_COL_TARGET = 'target'
FEED_COL_DAILY_SPEND = 'daily_spend'
FEED_COL_DAILY_GROSS_PROFIT = 'daily_gross_profit'
FEED_COL_DAILY_PROFIT = 'daily_profit'
FEED_COL_DAYS = 'days'

# output columns and initial values
BULK_COL_ACCOUNT = 'Account'
BULK_COL_CAMPAIGN = 'Campaign'
BULK_COL_PUBLISHER_TARGET_CPA = 'Publisher Target CPA'
BULK_COL_PUBLISHER_TARGET_ROAS = 'Publisher Target ROAS'
outputDf[BULK_COL_PUBLISHER_TARGET_CPA] = "<<YOUR VALUE>>"
outputDf[BULK_COL_PUBLISHER_TARGET_ROAS] = "<<YOUR VALUE>>"

### User Code Starts Here

# intermediate columns
COL_DAMPENED_TARGET_CPA = FEED_COL_TARGET_CPA + '_dampened'
COL_DAMPENED_TARGET_ROAS = FEED_COL_TARGET_ROAS + '_dampened'
COL_EXPECTED = 'expected'
COL_ACCEPTABLE = 'acceptable'

# define Status values
VAL_STATUS_ACTIVE = 'Active'
VAL_STATUS_PAUSED = 'Paused'
VAL_BLANK = ''

# define ObjectiveUnit values
VAL_OBJECTIVE_UNIT_UNKNOWN = -1
VAL_OBJECTIVE_UNIT_CONVERSIONS = 1
VAL_OBJECTIVE_UNIT_REVENUE = 2
VAL_OBJECTIVE_UNIT_NONE = 3
 
VAL_CONSTRAINT_UNIT_UNKNOWN = -1
VAL_CONSTRAINT_UNIT_SPEND = 1
VAL_CONSTRAINT_UNIT_ROAS = 2
VAL_CONSTRAINT_UNIT_CPA = 3

print("inputDf shape", inputDf.shape)
print("inputDf info", inputDf.info())

## cleanup data
#inputDf.drop(columns=['Unnamed: 30'], inplace=True)

inputDf[FEED_COL_FOLDER_ID] = inputDf[FEED_COL_FOLDER_ID].astype(int)
inputDf[FEED_COL_CAMPAIGN_OVERRIDE] = inputDf[FEED_COL_CAMPAIGN_OVERRIDE].astype(int)

inputDf = inputDf[inputDf[FEED_COL_FOLDER_ID].isin(VAL_LIMIT_FOLDERS)]
inputDf = inputDf[inputDf[FEED_COL_CAMPAIGN_OVERRIDE] != 1]

if VAL_TARGET_MODE == 'Traffic':
    inputDf = inputDf[inputDf[FEED_COL_PUBLISHER_BID_TRAFFICKING]=='Traffic']


## force expected types
inputDf.loc[:,[FEED_COL_TARGET_CPA]] = pd.to_numeric(inputDf[FEED_COL_TARGET_CPA], errors='coerce')
inputDf.loc[:,[FEED_COL_TARGET_ROAS]] = pd.to_numeric(inputDf[FEED_COL_TARGET_ROAS], errors='coerce')


## Grab current target stored on zero boost; basis for change detection later
comparison_cols = [FEED_COL_ACCOUNT, FEED_COL_CAMPAIGN, FEED_COL_TARGET_CPA, FEED_COL_TARGET_ROAS]

original_df = inputDf.loc[inputDf[FEED_COL_BOOST_PERCENT] == 0, comparison_cols]

## maximize profit for picking out boost with highest daily profit
optimal_df = inputDf \
    .sort_values(by=[ \
                    FEED_COL_DAILY_PROFIT, \
                    FEED_COL_DAILY_GROSS_PROFIT, \
                    FEED_COL_ABS_BOOST_PERCENT, \
                    FEED_COL_TARGET
                ], \
                ascending=[ \
                    False, \
                    False, \
                    True, \
                    True \
                ]) \
    .groupby([FEED_COL_CAMPAIGN_ID, FEED_COL_FOLDER_ID]) \
    .first() \
    .reset_index()

## Apply damping to the ideal targets based on current target and damping percentage using numpy
optimal_df[COL_DAMPENED_TARGET_CPA] = np.where(optimal_df[FEED_COL_TARGET],
    np.where(
    optimal_df[FEED_COL_BIDDING_STRATEGY_OLD_TYPE].isin(['TargetCPA','MaximizeConversions']),
    np.clip(
        optimal_df[FEED_COL_TARGET],
        optimal_df[FEED_COL_TARGET_CPA] * (1 - optimal_df[FEED_COL_BID_AND_TARGET_DAMPING]/100),
        optimal_df[FEED_COL_TARGET_CPA] * (1 + optimal_df[FEED_COL_BID_AND_TARGET_DAMPING]/100)
        ).round(2),
    np.nan),optimal_df[FEED_COL_TARGET].round(2))


optimal_df[COL_DAMPENED_TARGET_ROAS] = np.where(optimal_df[FEED_COL_TARGET],
    np.where(
    optimal_df[FEED_COL_BIDDING_STRATEGY_OLD_TYPE].isin(['TargetROAS','MaximizeConversionValue']),
    np.clip(
        100 * optimal_df[FEED_COL_TARGET],
        optimal_df[FEED_COL_TARGET_ROAS] * (1 - optimal_df[FEED_COL_BID_AND_TARGET_DAMPING]/100),
        optimal_df[FEED_COL_TARGET_ROAS] * (1 + optimal_df[FEED_COL_BID_AND_TARGET_DAMPING]/100)
        ).round(0),
    np.nan),(100 * optimal_df[FEED_COL_TARGET]).round(0))



optimal_df[COL_EXPECTED] = np.where(optimal_df[FEED_COL_BIDDING_STRATEGY_OLD_TYPE].isin(['MaximizeConversions','TargetCPA']),
    round(optimal_df[FEED_COL_TARGET_CPA]*(100 + optimal_df[FEED_COL_BOOST_PERCENT])/100,2),
    round((1/100)*optimal_df[FEED_COL_TARGET_ROAS] * 100 / (100 + optimal_df[FEED_COL_BOOST_PERCENT]),2))
                                                                       
optimal_df[COL_ACCEPTABLE] = optimal_df[FEED_COL_TARGET].between(
    (1 - TARGET_TOLERANCE) * optimal_df[COL_EXPECTED],
    (1 + TARGET_TOLERANCE) * optimal_df[COL_EXPECTED]
)


results_cols = [FEED_COL_ACCOUNT, FEED_COL_CAMPAIGN, COL_DAMPENED_TARGET_CPA, COL_DAMPENED_TARGET_ROAS, COL_ACCEPTABLE, FEED_COL_BOOST_PERCENT]
results_df = optimal_df.loc[:, results_cols] \
                    .rename(columns = { \
                        COL_DAMPENED_TARGET_CPA: FEED_COL_TARGET_CPA, \
                        COL_DAMPENED_TARGET_ROAS: FEED_COL_TARGET_ROAS, \
                    }) \
                       .copy()


debugDf = optimal_df

# only include changed rows in bulk file
(outputDf, _) = select_changed(results_df, \
                                    original_df, \
                                    diff_cols = [ \
                                        FEED_COL_TARGET_CPA, \
                                        FEED_COL_TARGET_ROAS, \
                                    ], \
                                    select_cols = [ \
                                        FEED_COL_ACCOUNT, \
                                        FEED_COL_CAMPAIGN, \
                                        FEED_COL_TARGET_CPA, \
                                        FEED_COL_TARGET_ROAS, \
                                        COL_ACCEPTABLE, \
                                        FEED_COL_BOOST_PERCENT, \
                                    ], \
                                    merged_cols=[FEED_COL_ACCOUNT, FEED_COL_CAMPAIGN] \
                                    )

outputDf = outputDf[outputDf[COL_ACCEPTABLE]].drop(COL_ACCEPTABLE,axis=1)

outputDf = outputDf[outputDf[FEED_COL_BOOST_PERCENT]!=0].drop(FEED_COL_BOOST_PERCENT,axis=1)

# use Bulk column headers
outputDf = outputDf.rename(columns = { \
                    FEED_COL_ACCOUNT: BULK_COL_ACCOUNT, \
                    FEED_COL_CAMPAIGN: BULK_COL_CAMPAIGN, \
                    FEED_COL_TARGET_CPA: BULK_COL_PUBLISHER_TARGET_CPA, \
                    FEED_COL_TARGET_ROAS: BULK_COL_PUBLISHER_TARGET_ROAS, \
                })

print("outputDf shape", outputDf.shape)
print("outputDf", tableize(outputDf.tail(5)))

## local debug
if local_dev:
    output_filename = 'outputDf.csv'
    outputDf.to_csv(output_filename, index=False)
    print(f"Local Dev: Output written to: {output_filename}")

    debug_filename = 'debugDf.csv'
    debugDf.to_csv(debug_filename, index=False)
    print(f"Local Dev: Debug written to: {debug_filename}")

Post generated on 2024-05-15 07:44:05 GMT

29 Feb 2024

« Script 719: SBA Intraday Budget Cap Budget via Dimension Tags Script 723: Campaigns Anomaly Detection »

MarinOne Scripts Creator's Corner