IntegratedML Custom Models - EAP Frequently Asked Questions

Program Status: Early Access Program (EAP) Target GA Release: IRIS 2026.1 Last Updated: 2025-01-12

Purpose

This document answers frequently asked questions about participating in the IntegratedML Custom Models Early Access Program.

For technical questions about using Custom Models, see the main documentation:

user_guide.md - How to use Custom Models
TROUBLESHOOTING.md - Common technical issues
EAP_KNOWN_ISSUES.md - Known limitations and bugs

Program Overview
Getting Started
Technical Questions
Feedback & Support
Timeline & Roadmap
Production Use
After EAP

Program Overview

What is the EAP?

Q: What is an Early Access Program?

A: An Early Access Program (EAP) gives select users exclusive access to new features before public launch. You get to try Custom Models early, provide feedback, and help shape the final product.

Benefits:

✅ Early access to cutting-edge ML capabilities
✅ Direct influence on product design
✅ Direct communication with product team
✅ Credit in GA release (if desired)

Responsibilities:

⏰ ~5-10 hours time commitment
📊 Provide structured feedback
🤝 Respect confidentiality until GA

Q: Why was I selected for the EAP?

A: You were selected because you have relevant ML/data science experience with IRIS or expressed interest in advanced IntegratedML capabilities. We're looking for participants who can:

Evaluate Custom Models for real-world use cases
Provide technical feedback on usability
Test documentation clarity
Identify gaps or missing features

Q: How many participants are in the EAP?

A: 5 participants in this initial EAP cohort. This small group allows for:

Personalized support and attention
Direct communication with product team
Manageable feedback volume for rapid iteration
Higher quality feedback per participant

What is IntegratedML Custom Models?

Q: What problem does Custom Models solve?

A: Custom Models solves the "last mile" problem for advanced ML use cases:

Problem: IntegratedML AutoML is great for quick models without code, but what if you need:

Custom preprocessing or feature engineering?
Domain-specific algorithms?
Third-party libraries (Prophet, LightGBM, XGBoost)?
Full control over model training?

Solution: Custom Models lets you write custom Python models while keeping IntegratedML's core benefits:

✅ Same SQL interface (CREATE MODEL, PREDICT())
✅ In-database execution (no data movement)
✅ Real-time predictions on live data
✅ Works alongside AutoML (use the right tool for each job)

Q: How does Custom Models relate to AutoML?

A: They're complementary, not competitive:

Use AutoML when:

You need quick models without writing code
You don't have ML expertise
Standard algorithms work for your use case
You want automatic feature engineering

Use Custom Models when:

You need custom preprocessing logic
You want to use specific third-party libraries
You have domain-specific algorithms
You need full control over model training

Both work together in the same IRIS instance. Choose the right tool for each use case.

Q: Is Custom Models replacing AutoML?

A: No! Custom Models is an addition to IntegratedML, not a replacement for AutoML.

AutoML remains the recommended choice for quick models without code
Custom Models extends IntegratedML for advanced use cases
You can use both in the same IRIS instance
Same SQL commands work with both providers

Getting Started

Installation

Q: What do I need to install?

A: Prerequisites:

IRIS 2025.2 or later
Python 3.8+ (3.11+ recommended)
Docker (recommended) or local IRIS installation
Git
5GB+ free disk space

See INSTALLATION.md for complete installation guide.

Q: How long does installation take?

A: Target: <30 minutes for most users

Typical timeline:

Docker setup: 15-20 minutes (recommended)
Local installation: 20-30 minutes
Running first demo: 5-10 minutes

If installation takes longer than 30 minutes, please contact support - we want to understand why!

Q: Which platform should I use?

A: Recommended priority:

macOS (primary support, most tested)
Linux (Ubuntu 22.04+, secondary support)
Windows (WSL2 or Docker, limited testing)

Best experience: Use Docker on any platform for most consistent results.

Q: Do I need a licensed IRIS installation?

A: No - IRIS Community Edition works fine for the EAP.

However, if you have a licensed installation and want to evaluate Custom Models in your production environment, that's encouraged (for evaluation only, not production deployment during EAP).

First Steps

Q: Where should I start?

A: Recommended path:

Read EAP_GUIDE.md - Understand the program (15 min)
Follow INSTALLATION.md - Get set up (20-30 min)
Run Credit Risk demo - See it work (15 min)
```
make demo-credit
```
Review QUICK_GUIDE_CUSTOM_MODELS.md - 5-minute overview
Complete entry survey - Tell us about your background (5 min)

Total time to first success: ~1 hour

Q: Which demo should I try first?

A: Start with Credit Risk - it's the simplest and shows core concepts:

Demo order (easiest to most complex):

Credit Risk (classification, feature engineering)
DNA Similarity (custom algorithms, pattern matching)
Fraud Detection (ensemble, multiple sub-models)
Sales Forecasting (hybrid Prophet + LightGBM, most complex)

Run demos with: make demo-credit, make demo-fraud, etc.

Evaluation

Q: What should I evaluate during the EAP?

A: Three levels of evaluation:

Level 1 - Minimum (~2 hours):

Run 2-3 demos successfully
Review documentation for clarity
Report any installation/usage issues

Level 2 - Recommended (~5-6 hours):

All of Level 1, plus:
Identify a real use case in your organization
Assess whether Custom Models could solve it
Review production deployment considerations

Level 3 - Advanced (~10+ hours):

All of Level 2, plus:
Create a simple custom model for your domain
Test with realistic data volumes
Evaluate operational requirements (monitoring, security, performance)

Choose the level that fits your available time.

Q: Should I create a custom model for my use case?

A: Optional but encouraged!

If you have time (extra 2-3 hours):

Pick a simple ML use case from your domain
Try creating a custom model following demo patterns
Document what worked well and what was confusing

If time is limited:

Just evaluate whether Custom Models could work for your use cases
Provide feedback on documentation and existing demos
That's still very valuable feedback!

Technical Questions

Using Custom Models

Q: Can I use any Python ML library?

A: Almost! As long as the library follows the scikit-learn interface (has fit() and predict() methods), you can wrap it.

Confirmed working:

✅ scikit-learn
✅ LightGBM
✅ XGBoost
✅ Prophet
✅ CatBoost
✅ Any sklearn-compatible library

Not directly supported:

❌ TensorFlow/Keras (not sklearn-compatible, but can be wrapped)
❌ PyTorch (not sklearn-compatible, but can be wrapped)
⚠️ Pure timeseries libraries (need wrapper pattern, see sales demo)

See demos for integration examples.

Q: How do I debug my custom model?

A: Recommended workflow:

Develop outside IRIS first:

# Test your model with pytest
pytest demos/my_demo/tests/

Deploy to IRIS:

# Copy model file to IRIS
docker cp my_model.py iml-custom-models-iris:/path/to/models/

Test in SQL:

CREATE MODEL MyModel PREDICTING (target) FROM MyTable
USING '{"model_name": "MyModel", ...}'

TRAIN MODEL MyModel

Check IRIS logs:
```
docker logs iml-custom-models-iris
```

See TROUBLESHOOTING.md for common debugging scenarios.

Q: Why do I need to restart IRIS after changing my model?

A: This is a current limitation of the model loading architecture:

Why: Python modules are loaded once and cached. Changes to .py files aren't picked up until IRIS restarts.

Workaround:

Develop and test models outside IRIS first
Only deploy to IRIS when model is working
Restart IRIS terminal or container after deploying:
```
docker restart iml-custom-models-iris
```

Future: Hot-reload is under investigation for post-GA release.

See: EAP_KNOWN_ISSUES.md#bug-006

Q: Can I use my existing Python ML code?

A: Mostly yes, with some adaptation:

If your code uses sklearn interface:

# Your existing code
class MyModel:
    def fit(self, X, y): ...
    def predict(self, X): ...

Wrap it for IntegratedML:

from shared.models.classification import ClassificationModel

class MyCustomModel(ClassificationModel):
    def __init__(self, **kwargs):
        super().__init__(**kwargs)
        self.my_model = MyExistingModel()

    def fit(self, X, y):
        X_processed = self._engineer_features(X)
        return self.my_model.fit(X_processed, y)

    def predict(self, X):
        X_processed = self._engineer_features(X)
        return self.my_model.predict(X_processed)

See demo applications for complete patterns.

SQL Integration

Q: What SQL commands work with Custom Models?

A: Same commands as AutoML:

-- Create model definition
CREATE MODEL MyModel
PREDICTING (target_column)
FROM MyTable
USING '{"model_name": "MyCustomModel", ...}'

-- Train model
TRAIN MODEL MyModel

-- Validate model
VALIDATE MODEL MyModel FROM ValidationTable

-- Make predictions
SELECT id,
       PREDICT(MyModel) as prediction,
       PROBABILITY(MyModel FOR 1) as confidence
FROM NewData

-- Drop model
DROP MODEL MyModel

See user_guide.md for complete SQL reference.

Q: How do I pass parameters to my custom model?

A: Use the JSON USING clause:

TRAIN MODEL MyModel
USING '{
    "model_name": "MyCustomModel",
    "path_to_classifiers": "/opt/iris/mgr/python/custom_models/classifiers",
    "user_params": {
        "enable_feature_1": 1,
        "threshold": 0.75,
        "custom_setting": "value"
    }
}'

In your model:

def __init__(self, **kwargs):
    super().__init__(**kwargs)
    # Access user_params
    params = kwargs.get('user_params', {})
    self.enable_feature_1 = params.get('enable_feature_1', 0)
    self.threshold = params.get('threshold', 0.5)

See api_reference.md for parameter handling details.

Feedback & Support

Providing Feedback

Q: How do I provide feedback?

A: Three channels:

Survey (structured, recommended):
- Entry survey: Background and initial impressions
- Exit survey: Overall evaluation and priorities
- Survey links provided by Data Platforms Product Team
Email (anytime):
- thomas.dyar@intersystems.com
- Use for questions, bug reports, suggestions
GitHub Issues (optional, if enabled):
- Bug reports
- Feature requests
- Documentation gaps

See EAP_GUIDE.md#how-to-provide-feedback for details.

Q: What makes good feedback?

A: Specific and actionable:

Great feedback:

✅ "Installation failed on Windows 11 with error: [specific error]. I was following step 3 of INSTALLATION.md."
✅ "I couldn't figure out how to pass parameters to my model. Documentation says 'use user_params' but doesn't show an example."
✅ "For my use case (customer churn prediction), I need to load pre-trained embeddings. Is this possible?"

Less helpful feedback:

❌ "Installation didn't work"
❌ "Documentation is confusing"
❌ "Needs more features"

Include:

What you were trying to do
What went wrong (specific error messages)
Your environment (OS, Python version, etc.)
Suggestions for improvement

Q: Will my feedback be implemented?

A: It depends:

Critical bugs: Fixed before GA (guaranteed) High-impact improvements: Strong consideration for GA Feature requests: Evaluated for GA or post-GA roadmap Nice-to-haves: Tracked for future releases

We review all feedback and will acknowledge receipt within 1-2 business days.

Your feedback directly influences priorities. The more specific and detailed, the more likely we can act on it!

Getting Support

Q: What if I get stuck?

A: Support process:

Check documentation:
- TROUBLESHOOTING.md - Common issues
- EAP_KNOWN_ISSUES.md - Known limitations
- EAP_FAQ.md - This document
Search repository: Check if others encountered same issue
Contact support:
- Email: thomas.dyar@intersystems.com
- Response time: 1-2 business days

Include in support request:

What you were trying to do
Steps to reproduce
Error messages (full text)
Environment details (OS, Python version, IRIS version)
Screenshots if applicable

Q: What support response time should I expect?

A: Target response time: 1-2 business days

What we provide:

✅ Help with installation issues
✅ Clarify documentation
✅ Troubleshoot bugs
✅ Guidance on model development

What we don't provide:

❌ 24/7 support (EAP is for evaluation, not production)
❌ Custom model development (but we can provide guidance)
❌ Guaranteed fixes within specific timeframe

Critical issues will be prioritized and may get faster response.

Q: Can I report the same issue multiple times?

A: Please don't! Check first:

Is it in EAP_KNOWN_ISSUES.md?
Did you already report it via email?
Is it in GitHub Issues (if enabled)?

If yes to any: We're already tracking it, no need to report again.

If no: Please report! We want to know about all issues.

Timeline & Roadmap

EAP Duration

Q: How long will the EAP run?

A: Approximately 6-8 weeks from EAP launch to GA preparation.

Timeline:

Week 1-2: Onboarding, installation, initial exploration
Week 3-6: Use case evaluation, custom model development
Week 7-8: Final feedback, GA preparation

Exact timeline will be communicated via email.

Q: When is the GA release?

A: Target: IRIS 2026.1 (Q2 2026)

Milestones:

EAP Launch: January 2025
Feature Freeze: ~2 months before GA
GA Release: Q2 2026

Exact GA date will be announced by InterSystems.

Q: What happens between EAP and GA?

A: Iteration and improvement:

Bug fixes: Fix critical bugs from EAP
Documentation: Fill gaps identified during EAP
Performance: Optimize based on EAP benchmarks
Features: Refine based on EAP feedback
Testing: Full QA testing before GA
Docs integration: Coordinate with docs.intersystems.com

See EAP_ROADMAP.md for complete roadmap.

Q: Can I influence the roadmap?

A: Absolutely!

Your input shapes:

Bug fix priorities
Documentation improvements
Feature priorities for GA
Post-GA roadmap direction

How to influence:

Complete surveys with detailed priorities
Share your use cases and requirements
Report what's working well (not just bugs!)
Participate in optional feedback calls

The more specific your feedback, the more actionable it is.

Production Use

Using in Production

Q: Can I use Custom Models in production during EAP?

A: No - evaluation only during EAP.

Reasons:

EAP is pre-release software
API may change based on feedback
Bugs may exist
No SLA or production support

After GA (2026.1):

✅ Production use supported
✅ Stable API with backward compatibility
✅ Full documentation
✅ Standard support channels

However: You can evaluate production readiness during EAP:

Test with realistic data volumes
Review security considerations
Assess operational requirements
Plan production deployment

Q: What should I consider for production deployment?

A: Key considerations (evaluate during EAP):

Security:

Model validation and testing
Access control
Data privacy

Performance:

Prediction latency requirements
Throughput requirements
Hardware sizing

Operations:

Monitoring and alerting
Model update process
Backup and recovery
Troubleshooting procedures

See deployment.md for current guidance. Production documentation will be expanded based on EAP feedback.

Q: Will my EAP models work in GA?

A: Goal: Yes, with minimal changes.

We aim for:

API compatibility between EAP and GA
Clear migration guide for any breaking changes
Deprecation warnings if API changes are needed

However: EAP is pre-release, so some changes may be necessary based on feedback.

We will:

Communicate any breaking changes clearly
Provide migration instructions
Support you through any required updates

After EAP

GA Release

Q: What happens when GA is released?

A: You get:

Stable release:
- Production-ready Custom Models
- Stable API with backward compatibility guarantee
- Full documentation on docs.intersystems.com
Acknowledgment (if desired):
- Credit as EAP participant
- Optional testimonial/case study opportunity
- Recognition in GA announcement materials
Continued access:
- Full access to GA release
- Standard InterSystems support channels
- Community forums and resources

Q: Will I be acknowledged for EAP participation?

A: Yes, if you want to be!

We'd like to:

Acknowledge EAP participants in GA release notes
Share use cases/testimonials (with permission)
Recognize contributors who provided exceptional feedback

Your choice:

Opt-in or opt-out via exit survey
Choose level of recognition (name only, testimonial, case study, etc.)
Review any materials before publication

No pressure - anonymous participation is fine too!

Q: Can I continue providing feedback after GA?

A: Yes, please do!

Post-GA feedback channels:

Standard InterSystems support
WRC (Worldwide Response Center)
Community forums
Feature requests via standard channels
GitHub Issues (if enabled)

Your EAP relationship gives you a direct line to the product team for complex feedback.

Future Enhancements

Q: What features are coming after GA?

A: Under consideration (based on EAP feedback):

Timeseries model native support
Model versioning and rollback
Hot reload for model changes
Model monitoring and drift detection
Model marketplace/template library

See EAP_ROADMAP.md#post-ga-future-considerations for full list.

Your EAP feedback will help prioritize these!

Q: Can I contribute to Custom Models after GA?

A: We're exploring community contribution options:

Potential areas:

Example models for specific industries
Model templates and best practices
Documentation improvements
Tutorials and blog posts

Details TBD - we'll share more as we approach GA.

Still Have Questions?

Not answered here?

📧 Email: thomas.dyar@intersystems.com
📚 Documentation: Check docs/ directory
🔍 Search: GitHub repository for similar questions

We'll update this FAQ as new questions come up during the EAP!

Thank you for participating in the EAP!

— The InterSystems Data Platforms Product Team

Document Version: 1.0 Last Updated: 2025-01-12 Next Update: Based on EAP questions

Latest Version: https://github.com/intersystems-community/integratedml-custom-models

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

IntegratedML Custom Models - EAP Frequently Asked Questions

Purpose

Table of Contents

Program Overview

What is the EAP?

What is IntegratedML Custom Models?

Getting Started

Installation

First Steps

Evaluation

Technical Questions

Using Custom Models

SQL Integration

Feedback & Support

Providing Feedback

Getting Support

Timeline & Roadmap

EAP Duration

Production Use

Using in Production

After EAP

GA Release

Future Enhancements

Still Have Questions?

FilesExpand file tree

EAP_FAQ.md

Latest commit

History

EAP_FAQ.md

File metadata and controls

IntegratedML Custom Models - EAP Frequently Asked Questions

Purpose

Table of Contents

Program Overview

What is the EAP?

What is IntegratedML Custom Models?

Getting Started

Installation

First Steps

Evaluation

Technical Questions

Using Custom Models

SQL Integration

Feedback & Support

Providing Feedback

Getting Support

Timeline & Roadmap

EAP Duration

Production Use

Using in Production

After EAP

GA Release

Future Enhancements

Still Have Questions?