Download the notebook here! Interactive online version:

Metrics Approximation

This notebook demonstrates metrics-based impact approximation via the engine’s internal response function registry.

Workflow overview

1. User provides products.csv

2. User configures DATA.ENRICHMENT section

3. User calls measure_impact(config.yaml)

4. Engine handles everything internally (adapter, enrichment, transform, model)

Initial setup

from pathlib import Path

import pandas as pd
from impact_engine_measure import measure_impact, load_results
from impact_engine_measure.core.validation import load_config
from impact_engine_measure.core import apply_transform
from impact_engine_measure.models.factory import get_model_adapter
from online_retail_simulator import enrich, simulate
from online_retail_simulator.simulate import simulate_product_details

Step 1 — Product Catalog

In production, this would be your actual product catalog.

output_path = Path("output/demo_metrics_approximation")
output_path.mkdir(parents=True, exist_ok=True)

catalog_job = simulate("configs/demo_metrics_approximation_catalog.yaml", job_id="catalog")
products = catalog_job.load_df("products")

print(f"Generated {len(products)} products")
print(f"Products catalog: {catalog_job.get_store().full_path('products.csv')}")
products.head()

Generated 100 products
Products catalog: /home/runner/work/tools-impact-engine-measure/tools-impact-engine-measure/docs/source/methods/output/demo_metrics_approximation/catalog/products.csv

	product_identifier	category	price
0	BZXD798331	Clothing	48.71
1	B9XDF5TCLQ	Clothing	176.69
2	BB2RLWLEXN	Food & Beverage	37.22
3	BT5SD06XTC	Health & Beauty	16.78
4	BGWTXGK34V	Toys & Games	97.20

Step 2 — Engine configuration

Configure the engine with the following sections.

• ENRICHMENT — Quality boost parameters

• TRANSFORM — Prepare data for approximation

• MODEL — metrics_approximation with response function

config_path = "configs/demo_metrics_approximation.yaml"

Step 3 — Impact evaluation

A single call to measure_impact() handles everything.

• Engine creates CatalogSimulatorAdapter

• Adapter simulates metrics

• Adapter generates product_details

• Adapter applies enrichment (quality boost)

• Transform extracts quality_before/quality_after

• MetricsApproximationAdapter computes impact

job_info = measure_impact(config_path, str(output_path), job_id="results")
print(f"Job ID: {job_info.job_id}")

Job ID: results

Step 4 — Review results

result = load_results(job_info)

data = result.impact_results["data"]
model_params = data["model_params"]
estimates = data["impact_estimates"]
summary = data["model_summary"]

print("=" * 60)
print("METRICS-BASED IMPACT APPROXIMATION RESULTS")
print("=" * 60)

print(f"\nModel Type: {result.model_type}")
print(f"Response Function: {model_params['response_function']}")

print("\n--- Aggregate Impact Estimates ---")
print(f"Total Impact:        ${estimates['impact']:.2f}")
print(f"Number of Products:  {summary['n_products']}")

============================================================
METRICS-BASED IMPACT APPROXIMATION RESULTS
============================================================

Model Type: metrics_approximation
Response Function: linear

--- Aggregate Impact Estimates ---
Total Impact:        $3336.46
Number of Products:  100

# Per-product data from model artifacts
per_product_df = result.model_artifacts["product_level_impacts"]

print("\n--- Per-Product Breakdown (first 10) ---")
print("-" * 60)
print(f"{'Product':<20} {'Delta Quality':<15} {'Baseline':<12} {'Impact':<12}")
print("-" * 60)
for _, p in per_product_df.head(10).iterrows():
    print(f"{p['product_id']:<20} {p['delta_metric']:<15.4f} ${p['baseline_outcome']:<11.2f} ${p['impact']:<11.2f}")

print("\n" + "=" * 60)
print("Demo Complete!")
print("=" * 60)

--- Per-Product Breakdown (first 10) ---
------------------------------------------------------------
Product              Delta Quality   Baseline     Impact
------------------------------------------------------------
B0AD6L5JB0           0.2030          $672.66      $68.27
B132JTX7CT           0.1820          $39.51       $3.60
B17DXLZI73           0.1690          $2780.40     $234.94
B2H3WC734B           0.1600          $1954.28     $156.34
B30WWMDHGF           0.1830          $718.74      $65.76
B31SZV8ISX           0.1720          $1348.48     $115.97
B3OC36BZNE           0.2060          $0.00        $0.00
B4797UN9V2           0.1750          $0.00        $0.00
B4AP37USC0           0.1080          $112.50      $6.07
B4DDXDDV7G           0.1700          $1238.51     $105.27

============================================================
Demo Complete!
============================================================

Step 5 — Model validation

Compare the model’s estimate against the true causal effect computed from counterfactual vs factual data.

def calculate_true_effect(
    baseline_metrics: pd.DataFrame,
    enriched_metrics: pd.DataFrame,
) -> dict:
    """Calculate TRUE impact by comparing total revenue with vs without enrichment."""
    baseline_total = baseline_metrics["revenue"].sum()
    enriched_total = enriched_metrics["revenue"].sum()
    impact = enriched_total - baseline_total

    return {
        "baseline_total": float(baseline_total),
        "enriched_total": float(enriched_total),
        "impact": float(impact),
    }

baseline_metrics = catalog_job.load_df("metrics").rename(columns={"product_identifier": "product_id"})

# Generate product details (adds quality_score to products, required before enrichment)
pd_config = {"PRODUCT_DETAILS": {"FUNCTION": "simulate_product_details_mock"}}
store = catalog_job.get_store()
store.write_yaml("product_details_config.yaml", pd_config)
catalog_job = simulate_product_details(catalog_job, store.full_path("product_details_config.yaml"))

enrich("configs/demo_metrics_approximation_enrichment.yaml", catalog_job)
enriched_metrics = catalog_job.load_df("enriched").rename(columns={"product_identifier": "product_id"})

# Add quality_score (mirrors adapter._apply_enrichment logic)
parsed = load_config(config_path)
enrichment_start = pd.to_datetime(parsed["DATA"]["ENRICHMENT"]["PARAMS"]["enrichment_start"])

products_original = catalog_job.load_df("product_details_original")
products_enriched = catalog_job.load_df("product_details_enriched")
orig_quality = products_original.set_index("product_identifier")["quality_score"].to_dict()
enr_quality = products_enriched.set_index("product_identifier")["quality_score"].to_dict()

enriched_metrics["date"] = pd.to_datetime(enriched_metrics["date"])
enriched_metrics["quality_score"] = enriched_metrics.apply(
    lambda row: (
        orig_quality.get(row["product_id"], 0.5)
        if row["date"] < enrichment_start
        else enr_quality.get(row["product_id"], 0.5)
    ),
    axis=1,
)

print(f"Baseline records: {len(baseline_metrics)}")
print(f"Enriched records: {len(enriched_metrics)}")

Baseline records: 1400
Enriched records: 1400

true_effect = calculate_true_effect(baseline_metrics, enriched_metrics)

true_impact = true_effect["impact"]
model_impact = estimates["impact"]

if true_impact != 0:
    recovery_accuracy = (1 - abs(1 - model_impact / true_impact)) * 100
else:
    recovery_accuracy = 100 if model_impact == 0 else 0

print("=" * 60)
print("TRUTH RECOVERY VALIDATION")
print("=" * 60)
print(f"True impact:       ${true_impact:,.2f}")
print(f"Model estimate:    ${model_impact:,.2f}")
print(f"Recovery accuracy: {max(0, recovery_accuracy):.1f}%")
print("=" * 60)

============================================================
TRUTH RECOVERY VALIDATION
============================================================
True impact:       $126.48
Model estimate:    $3,336.46
Recovery accuracy: 0.0%
============================================================

Convergence analysis

How does the estimate converge to the true effect as sample size increases?

sample_sizes = [5, 10, 25, 50, 100]
estimates_list = []
truth_list = []

transform_config = parsed["DATA"]["TRANSFORM"]
measurement_config = parsed["MEASUREMENT"]
all_product_ids = enriched_metrics["product_id"].unique()

for n in sample_sizes:
    subset_ids = all_product_ids[:n]
    enriched_sub = enriched_metrics[enriched_metrics["product_id"].isin(subset_ids)]
    baseline_sub = baseline_metrics[baseline_metrics["product_id"].isin(subset_ids)]

    true = calculate_true_effect(baseline_sub, enriched_sub)
    truth_list.append(true["impact"])

    transformed = apply_transform(enriched_sub, transform_config)
    model = get_model_adapter("metrics_approximation")
    model.connect(measurement_config["PARAMS"])
    result = model.fit(data=transformed)
    estimates_list.append(result.data["impact_estimates"]["impact"])

print("Convergence analysis complete.")

Convergence analysis complete.

from notebook_support import plot_convergence

plot_convergence(
    sample_sizes,
    estimates_list,
    truth_list,
    xlabel="Number of Products",
    ylabel="Impact ($)",
    title="Metrics Approximation: Convergence of Estimate to True Effect",
)