datarobot-model-explainability

DataRobot Model Explainability Skill

This skill provides comprehensive guidance for understanding model decisions, analyzing prediction explanations, and interpreting model behavior using various explainability techniques.

Quick Start

Most common use case: Get prediction explanations for a specific prediction

1. Identify model/project: Get project_id + model_id (from a deployment’s champion model, or from training)
2. Compute explanations: Use dr.PredictionExplanations.create(project_id, model_id, dataset_id, ...)
3. Read explanation rows: Use dr.PredictionExplanations.get(project_id, prediction_explanations_id) and iterate get_rows()

Example: "Explain why the model predicted 0.85 for this customer record"

When to use this skill

Use this skill when you need to:

• Understand why a model made a specific prediction
• Analyze feature contributions to predictions (SHAP values)
• Generate prediction explanations for stakeholders
• Understand model behavior and decision-making
• Generate model diagnostics (ROC curves, lift charts, confusion matrices)
• Analyze partial dependence and feature interactions
• Meet regulatory compliance requirements for model explainability

Key capabilities

1. Prediction Explanations

• Get SHAP (SHapley Additive exPlanations) values for individual predictions
• Understand feature contributions to each prediction
• Generate human-readable explanations
• Export explanations for reporting

2. Feature Impact Analysis

• Analyze how individual features impact predictions
• Compare feature importance across different predictions
• Understand feature interactions
• Identify key drivers for model decisions

3. Model Diagnostics

• Generate ROC curves for classification models
• Create lift charts and gain charts
• Generate confusion matrices
• Analyze model calibration
• Create partial dependence plots
• Generate ICE (Individual Conditional Expectation) plots

4. Global Model Understanding

• Understand overall model behavior
• Analyze feature importance at model level
• Compare explainability across models
• Generate model interpretability reports

Workflow examples

Example 1: Explain a specific prediction

User request: "Why did the model predict 0.85 probability for customer ID 12345?"

Agent workflow:

1. Get the prediction record for customer 12345
2. Retrieve prediction explanations (SHAP values) for this prediction
3. Sort features by contribution (positive/negative impact)
4. Explain which features increased the prediction and which decreased it
5. Provide human-readable summary of the explanation

Example 2: Generate model diagnostics

User request: "Generate ROC curve and confusion matrix for model xyz123"

Agent workflow:

1. Get model information and validation data
2. Generate ROC curve with AUC score
3. Generate confusion matrix with precision/recall metrics
4. Create lift chart showing model performance
5. Compile diagnostics into a report

Using DataRobot SDK

This skill guides you to use the DataRobot Python SDK directly. Install the SDK if needed:

pip install datarobot

Key SDK Operations

Use these DataRobot SDK methods for model explainability:

Prediction Explanations (SDK v3.10.0):

• dr.PredictionExplanations.create(project_id, model_id, dataset_id, ...) - Compute prediction explanations on a dataset
• dr.PredictionExplanations.get(project_id, prediction_explanations_id) - Fetch explanations and read rows
• prediction_explanations.get_rows() - Iterate explanation rows

Model Diagnostics:

• model.get_roc_curve(source='validation') - Get ROC curve with AUC
• model.get_confusion_matrix() - Get confusion matrix
• model.get_lift_chart() - Get lift chart data
• model.get_feature_impact() - Get feature importance

Feature Analysis:

• model.get_partial_dependence(feature_name) - Get partial dependence data

See the Common Patterns section below for complete examples.

Best practices

1. Use SHAP for local explanations: SHAP values provide the best local (per-prediction) explanations
2. Combine with global insights: Use feature importance for global understanding, SHAP for local
3. Explain in context: Always explain predictions in the context of the business problem
4. Visualize explanations: Use charts and graphs to make explanations more understandable
5. Document explanations: Save explanations for compliance and auditing purposes
6. Compare across models: Use explainability to compare different models

Common patterns

Pattern 1: Get prediction explanations

import datarobot as dr
import os

# Initialize client
client = dr.Client(
    token=os.getenv("DATAROBOT_API_TOKEN"),
    endpoint=os.getenv("DATAROBOT_ENDPOINT")
)

# Example: compute explanations for a dataset (batch)
project_id = "project_id_here"
model_id = "model_id_here"
dataset_id = "dataset_id_here"  # dataset to explain

pe = dr.PredictionExplanations.create(
    project_id=project_id,
    model_id=model_id,
    dataset_id=dataset_id,
    max_explanations=5,
)

pe2 = dr.PredictionExplanations.get(project_id, pe.id)
for row in pe2.get_rows():
    # row contains per-row explanation info
    print(row)

Pattern 2: Generate model diagnostics

import datarobot as dr

# Get model
model = dr.Model.get("xyz123")

# Get ROC curve
roc_curve = model.get_roc_curve(source='validation')
print(f"AUC Score: {roc_curve.auc:.3f}")

# Get confusion matrix (for classification models)
try:
    cm = model.get_confusion_matrix(source='validation')
    print(f"Precision: {cm.precision:.3f}")
    print(f"Recall: {cm.recall:.3f}")
    print(f"F1 Score: {cm.f1_score:.3f}")
except:
    print("Confusion matrix not available for this model type")

# Get lift chart
lift_chart = model.get_lift_chart(source='validation')
print(f"Lift at 10%: {lift_chart.lift[10]:.3f}")

Understanding SHAP Values

SHAP (SHapley Additive exPlanations) values explain the output of any machine learning model:

• Positive SHAP value: Feature increases the prediction
• Negative SHAP value: Feature decreases the prediction
• Magnitude: Larger absolute values indicate stronger impact
• Sum: SHAP values sum to the difference between prediction and baseline

Example interpretation:

• Feature "age" has SHAP value +0.15 → Age increases prediction by 0.15
• Feature "income" has SHAP value -0.08 → Income decreases prediction by 0.08

Model Diagnostics Explained

ROC Curve

• Purpose: Evaluate classification model performance
• AUC Score: Area under curve (0-1), higher is better
• Use: Compare models, select optimal threshold

Confusion Matrix

• Purpose: Show classification accuracy breakdown
• Metrics: Precision, Recall, F1-Score
• Use: Understand model errors and biases

Lift Chart

• Purpose: Show model performance vs. baseline
• Use: Evaluate model value, select top predictions

Partial Dependence

• Purpose: Show how a feature affects predictions on average
• Use: Understand feature relationships and interactions

Error handling

Common errors and solutions:

• Prediction not found: Ensure prediction_id is valid and accessible
• Explanations unavailable: Some model types don't support explanations
• Feature not found: Verify feature name matches model features
• Insufficient data: Need validation data for some diagnostics

SDK Setup

Install DataRobot SDK

pip install datarobot

Initialize Client

import datarobot as dr
import os

client = dr.Client(
    token=os.getenv("DATAROBOT_API_TOKEN"),
    endpoint=os.getenv("DATAROBOT_ENDPOINT", "https://app.datarobot.com")
)

Resources

• DataRobot Python SDK Documentation
• DataRobot Model Insights Documentation
• Prediction Explanations Guide
• SHAP Documentation