Model Evaluation

Skill: Model Evaluation, Visualisation & Interpretation

1. Description & Rationale

This skill establishes a standardized framework for evaluating and visualizing machine learning models. It covers simple metrics (like accuracy) for balanced datasets, and advanced metrics (like AUC-PR and Confusion Matrices) for complex or imbalanced datasets (e.g., fraud detection, rare event prediction).

2. Environment Prerequisites

• Libraries: matplotlib, seaborn, pandas, numpy, scikit-learn (for local visualization).
• Engines: PySpark MLlib or BigQuery ML.

3. Agent Execution Guidelines (System Prompts)

When writing evaluation code:

1. Match Complexity:
   • For simple, balanced classification: Use Accuracy (standard).
   • For imbalanced classification (e.g., predicting a rare action): Enforce Area Under the Precision-Recall Curve (AUC-PR) and Confusion Matrix.
2. Local Visualization: If plotting is required, copy a small sample of predictions to local memory using .toPandas() (Spark) or .to_pandas() (BigFrames) only for the visualization step.

4. Opinionated Code Patterns

A. Simple Evaluation (Accuracy)

Spark MLlib

from pyspark.ml.evaluation import MulticlassClassificationEvaluator

# Evaluate accuracy
predictions = pipeline_model.transform(test)
evaluator = MulticlassClassificationEvaluator(metricName="accuracy")
accuracy = evaluator.evaluate(predictions)
print(f"Model Accuracy: {accuracy:.2%}")

BigFrames ML

# Score the model (returns a DataFrame with metrics, including accuracy)
metrics = model.score(X_test, y_test)
print("Evaluation Metrics:")
print(metrics)

B. Advanced Evaluation (AUC-PR & Confusion Matrix for Imbalanced Data)

Spark MLlib AUC-PR

from pyspark.ml.evaluation import BinaryClassificationEvaluator

# Initialize evaluator for AUC-PR
evaluator = BinaryClassificationEvaluator(
    labelCol="label",
    rawPredictionCol="probability",
    metricName="areaUnderPR"
)
auc_pr = evaluator.evaluate(predictions)
print(f"AUC-PR: {auc_pr:.4f}")

Local Confusion Matrix Visualization (Spark or BigFrames)

import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.metrics import confusion_matrix
import pandas as pd

# Convert a small sample of predictions to Pandas for plotting
# (Ensure 'predictions' is the evaluated DataFrame)
y_true = predictions.select("label").toPandas()
y_pred = predictions.select("prediction").toPandas()

cm = confusion_matrix(y_true, y_pred)
cm_df = pd.DataFrame(cm, index=['Actual Negative', 'Actual Positive'],
                         columns=['Predicted Negative', 'Predicted Positive'])

plt.figure(figsize=(6, 5))
sns.heatmap(cm_df, annot=True, fmt='d', cmap='Blues', cbar=False)
plt.title('Confusion Matrix')
plt.ylabel('Actual')
plt.xlabel('Predicted')
plt.show()

5. Verification Checklist

• Metric Selection: Verify if the dataset is imbalanced. If yes, ensure AUC-PR is used instead of just accuracy.
• Safe Local Conversion: Confirm that .toPandas() is only called on the final prediction summary/evaluation set, never on the main training dataset.