FAQ & Concepts

Common questions and important concepts about Rassket, its capabilities, and best practices.

What Makes Rassket Different?

Domain-Aware Intelligence

Unlike generic AutoML tools, Rassket understands domain context. When you select Energy or Research, it applies specialized feature engineering and provides domain-appropriate explanations.

Decision Intelligence Focus

Rassket doesn't just build models—it provides insights and explanations that help you make decisions. The platform explains what matters, why it matters, and how to use predictions.

Production-Ready Outputs

Every model comes with export-ready packages, comprehensive reports, and deployment documentation. No additional work needed to move from prototype to production.

Accessibility

Rassket makes ML accessible to non-technical users while providing transparency and control for technical users. You don't need deep ML expertise to get value.

When to Use Sub-Domains

Use Sub-Domains When:

  • Your problem closely matches a sub-domain description
  • You want the most refined feature engineering
  • You need domain-specific explanations
  • Your use case is clearly defined

Skip Sub-Domains When:

  • Your problem spans multiple sub-domains
  • You're unsure which sub-domain fits
  • You want to start quickly
  • Your use case is exploratory
Best Practice: Start with the base domain if unsure. You can always retrain with a sub-domain later. The base domain still provides significant benefits over generic AutoML.

How Accurate Models Are Evaluated

Evaluation Process

  1. Data Splitting: Data is split into train/validation/test sets
  2. Cross-Validation: Models are evaluated using k-fold cross-validation
  3. Test Set Evaluation: Final metrics are computed on held-out test set
  4. Multiple Metrics: Comprehensive metrics are calculated, not just one

Metrics Used

  • Regression: R², RMSE, MAE, MSE
  • Classification: Accuracy, Precision, Recall, F1, ROC-AUC
  • Additional: Cross-validation scores, confidence intervals

Model Comparison

When multiple models are trained:

  • All models are evaluated on the same test set
  • Metrics are compared side-by-side
  • Best model is selected based on primary metric
  • Statistical significance is considered

Reliability Assessment

Rassket provides:

  • Model reliability scores
  • Potential issues identification
  • Overfitting detection
  • Confidence intervals

Data Privacy and Handling

Data Storage

  • Uploaded files are stored securely
  • Data is processed according to your domain selection
  • Models are stored with unique identifiers
  • Raw data is not retained in exported packages

Data Processing

  • Data is processed server-side
  • Preprocessing pipelines are preserved
  • Feature engineering is reproducible
  • No data is shared with third parties

Export Security

  • Model packages contain no raw data
  • Only model files and preprocessing pipelines are exported
  • Reports contain aggregated metrics only
  • No sensitive data in exports
Privacy: Rassket processes your data to train models but prioritizes privacy. Exported packages and reports contain no raw data—only model artifacts and aggregated insights.

Common Questions

How long does training take?

Training time depends on:

  • Dataset Size: Small datasets (<10K rows) take minutes; large datasets (>100K rows) can take 30+ minutes
  • Number of Models: Training multiple models multiplies time
  • Model Complexity: Complex models take longer
  • Hyperparameter Tuning: More optimization trials increase time

Most training completes within 10-30 minutes. Keep the browser tab open during training.

What if my data doesn't fit Energy or Research?

You can still use Rassket! Select the closest domain, or use the base domain. Rassket will build accurate models regardless. Domain selection enhances feature engineering and explanations, but models work for any tabular data.

Can I use Rassket without selecting a domain?

Domain selection is required for preprocessing, but you can choose the base domain (Energy or Research) without selecting a sub-domain. This still provides domain-aware benefits.

What file formats are supported?

Currently, Rassket supports CSV files. Ensure your CSV:

  • Has headers in the first row
  • Uses commas as delimiters
  • Is UTF-8 encoded
  • Is under 50MB

Can I train multiple models?

Yes! During analysis, you can select multiple recommended models. Rassket will train all selected models and compare their performance, helping you find the best one.

How do I deploy models to production?

Export the model package (ZIP file). It contains:

  • Trained model file
  • Preprocessing pipeline
  • Inference code
  • Documentation
  • Requirements file

Follow the included documentation to deploy in your environment.

What if training fails?

Common causes:

  • Missing Target: Ensure target column is selected
  • Insufficient Data: Very small datasets may fail
  • Data Quality: Check for data issues
  • Timeout: Large datasets may timeout—try with a sample

Check error messages for specific issues. Most problems are data-related and can be fixed.

Can I retrain models with different settings?

Yes! Upload your data again and select different options:

  • Different domain or sub-domain
  • Different models
  • Different problem type

Compare results to find the best configuration.

Key Concepts

AutoML

Automated Machine Learning—the process of automating ML pipeline steps including data preprocessing, feature engineering, model selection, and hyperparameter tuning.

Domain Awareness

Understanding domain context (Energy vs. Research) to apply specialized feature engineering and provide domain-appropriate explanations.

Decision Intelligence

The combination of ML predictions with explanations and insights that help users make informed decisions based on model outputs.

Feature Engineering

The process of creating new features from raw data to improve model performance. Rassket automates this with domain-specific enhancements.

Hyperparameter Tuning

Optimizing model hyperparameters (settings that control model behavior) to improve performance. Rassket uses Optuna for automated optimization.

SHAP Values

SHapley Additive exPlanations—a unified measure of feature importance that explains individual predictions and overall model behavior.

Best Practices

Data Preparation

  • Ensure CSV has headers
  • Clean data before upload (remove obvious errors)
  • Include relevant features
  • Ensure target column is present (for training)

Domain Selection

  • Choose domain that best matches your data
  • Select sub-domain if your use case matches closely
  • Don't worry if unsure—base domain still helps

Model Training

  • Start with automatic model selection
  • Train multiple models for comparison if needed
  • Review metrics and diagnostics
  • Export models you want to use

Evaluation

  • Review multiple metrics, not just one
  • Check feature importance
  • Read AI insights for context
  • Review visualizations for issues

Next Steps

Still have questions?