Outputs & Artifacts

Rassket generates comprehensive outputs and artifacts that are ready for production use, decision-making, and sharing with stakeholders. This page details what you get and how to use it.

Models

Trained Model Files

Each trained model includes:

• Model Binary: Serialized model file (pickle format)
• Model Metadata: Problem type, feature names, target column, domain
• Preprocessing Pipeline: Feature engineering and transformation steps
• Model Configuration: Hyperparameters and training configuration

Model Information

• Model ID: Unique identifier for your model
• Model Type: Algorithm used (XGBoost, LightGBM, etc.)
• Problem Type: Regression or Classification
• Training Date: When the model was trained
• Performance Metrics: Evaluation metrics on test set

Model Comparison

When multiple models are trained, you receive:

• Comparison Table: Side-by-side performance metrics
• Best Model Identification: Top performer highlighted
• Individual Model Details: Full information for each model

Metrics

Regression Metrics

• R² (R-squared): Proportion of variance explained (0-1, higher is better)
• RMSE: Root Mean Squared Error (lower is better)
• MAE: Mean Absolute Error (lower is better)
• MSE: Mean Squared Error (lower is better)

Classification Metrics

• Accuracy: Overall correctness (0-1, higher is better)
• Precision: True positives / (True positives + False positives)
• Recall: True positives / (True positives + False negatives)
• F1 Score: Harmonic mean of precision and recall
• ROC-AUC: Area under ROC curve (for binary classification)

Comprehensive Metrics

Beyond basic metrics, Rassket provides:

• Train/Validation/Test Metrics: Performance on each split
• Cross-Validation Scores: Robust performance estimates
• Confidence Intervals: Uncertainty estimates
• Statistical Tests: Significance testing (when applicable)

Forecasts

Prediction Interface

The Results page includes a prediction interface where you can:

• Input feature values manually
• Get instant predictions
• See prediction confidence (for classification)

Batch Predictions

Model packages include code for making batch predictions on new data:

• Load model and preprocessing pipeline
• Process new data through pipeline
• Generate predictions
• Export results

Forecast Quality

Forecasts include:

• Point Predictions: Single predicted value
• Confidence Intervals: Uncertainty ranges (when available)
• Feature Contributions: SHAP values showing feature impact

Explanations

Feature Importance

• SHAP Values: Unified measure of feature importance
• Global Importance: Overall feature contributions
• Local Importance: Feature contributions for individual predictions
• Visualizations: Interactive charts showing feature importance

AI-Powered Insights

• Performance Summary: Plain-language explanation of model performance
• Feature Insights: Which features matter most and why
• Model Reliability: Assessment of model confidence and limitations
• Real-World Implications: How to use predictions in decision-making

Domain-Aware Explanations

When a domain is selected, explanations include:

• Domain-specific terminology
• Context-appropriate interpretations
• Industry-relevant insights
• Actionable recommendations

Reports

PDF Executive Reports

Exportable PDF reports include:

• Executive Summary: High-level overview of results
• Model Performance: Comprehensive metrics and visualizations
• Feature Analysis: Feature importance and insights
• Visualizations: Key charts and diagnostics
• Domain Insights: Domain-aware analysis (if domain selected)
• Recommendations: Actionable next steps

Model Packages (ZIP)

Exportable model packages include:

• Trained Model: Serialized model file
• Preprocessing Pipeline: Feature engineering code
• Inference Script: Code for making predictions
• Requirements: Python dependencies list
• Documentation: Model documentation and usage guide
• Metadata: Model configuration and training details

Visualizations

Default Visualizations

• Feature Importance: Bar chart of feature contributions
• Confusion Matrix: Classification accuracy breakdown
• ROC Curve: Classification performance visualization
• Prediction Distribution: Histogram of predicted values
• Metrics Charts: Visual representation of evaluation metrics
• Correlation Heatmap: Feature correlation matrix

Advanced Diagnostics

• Residual Diagnostics: Q-Q plots, residuals vs fitted (regression)
• R² Indicators: Goodness-of-fit visualizations (regression)
• Calibration Curve: Prediction probability calibration (classification)
• Outcomes Summary: Pie chart of prediction outcomes

Visualization Format

• Interactive Charts: Plotly-based interactive visualizations
• Exportable: Can be exported as images
• Embedded in Reports: Included in PDF reports

How Outputs Are Used

In Production

• Model Deployment: Use model packages to deploy models to production
• API Integration: Integrate models into APIs and services
• Batch Processing: Run batch predictions on new data
• Real-Time Predictions: Make predictions in real-time applications

In Decision-Making

• Executive Reports: Share PDF reports with decision-makers
• Insights: Use AI-powered insights to inform decisions
• Feature Analysis: Understand what drives predictions
• Forecasts: Use predictions to plan and optimize

In Communication

• Stakeholder Presentations: Use PDF reports for presentations
• Documentation: Include reports in project documentation
• Collaboration: Share visualizations and insights with teams
• Reporting: Include metrics and forecasts in regular reports

Export Options

Export Model Package

Click "Export Model Package" to download a ZIP file containing:

• Everything needed to deploy and use the model
• Preprocessing pipeline
• Inference code
• Documentation

Export PDF Report

Click "Export PDF Report" to generate a comprehensive PDF including:

• Executive summary
• All visualizations
• Metrics and analysis
• Domain-aware insights

Data Privacy and Handling

Data Storage

• Uploaded files are stored securely
• Models are stored with unique identifiers
• Data is processed according to your domain selection

Export Security

• Model packages contain no raw data
• Only model files and preprocessing pipelines are exported
• Reports contain aggregated metrics and visualizations

Next Steps

Now that you understand outputs and artifacts:

• See the Dashboard Walkthrough to see outputs in context
• Explore Use Cases to see how outputs are used in practice
• Check the FAQ for common questions about outputs