Version: Latest (3.20.2)

mmm_attribution

Python

D_variables maps each raw input variable to a final channel and measure. It must contain only the structural columns variable, channel, and measure.

D_measures is optional. When provided, the recommended user-facing structure is one row per measure with measure and aggregation_weight:

measure	aggregation_weight
`direct_searches`	`0.45`
`clicks`	`0.45`
`impressions`	`0.10`

aggregation_weight controls how strongly each measure contributes when variable-level attributions are aggregated into final channel-level attribution. The internal signal scaling is estimated automatically from the data.

import pandas as pd
from ChannelAttributionPro import mmm_attribution

token = "yourtoken"

Data = pd.read_csv("https://app.channelattribution.io/data/data_mmm_3.csv")

D_variables = pd.DataFrame({
    "variable": [
        "direct_searches",
        "facebook_impressions",
        "facebook_clicks",
        "google_impressions",
        "google_clicks",
        "tv"
    ],
    "channel": [
        "direct",
        "facebook_ads",
        "facebook_ads",
        "google_ads",
        "google_ads",
        "tv"
    ],
    "measure": [
        "direct_searches",
        "impressions",
        "clicks",
        "impressions",
        "clicks",
        "impressions"
    ],
})

D_measures = pd.DataFrame({
    "measure": ["direct_searches", "clicks", "impressions"],
    "aggregation_weight": [0.45, 0.45, 0.10],
})

target = "conversions"

res = mmm_attribution(
    Data=Data,
    D_variables=D_variables,
    D_measures=D_measures,
    target=target,
    model="linear",  # "linear", "reward", or "copula"
    max_p=12,
    seed=1234567,
    verbose=1,
    password=token,
)

print(res.head())

Using a different internal model:

res_reward = mmm_attribution(
    Data=Data,
    D_variables=D_variables,
    D_measures=D_measures,
    target=target,
    model="reward",
    max_p=12,
    seed=1234567,
    password=token,
)

res_copula = mmm_attribution(
    Data=Data,
    D_variables=D_variables,
    D_measures=D_measures,
    target=target,
    model="copula",
    max_p=12,
    nsim=1000,
    seed=1234567,
    password=token,
)

Python​

Python