Skip to content

Roadmap

What's coming in future versions of KladML.

Current Version: v0.5.x

✅ Available Now

  • ONNX Export - Export models to ONNX format for edge deployment
  • TorchScript Export - Self-contained deployment artifacts
  • Multiple Evaluators - Anomaly, Classification, Regression evaluation
  • Plot Types - Histogram, CDF, Log-Log, Loss Curves
  • Run Tracking - Automatic run ID generation and tracking
  • Early Stopping - Configurable patience and minimum delta
  • Registry System - Architecture and component registration
  • Terminal UI (TUI) - Interactive terminal interface (kladml ui)
  • CLI Commands - train, evaluate, data, project, family, experiment

v0.6.x - Core CLI Enhancements

Complete the core command-line workflow:

  • [ ] Enhanced Training - kladml train --config <path> with better progress tracking
  • [ ] Run Evaluation - kladml evaluate --run <run_id> --evaluator <name>
  • [ ] Model Export - kladml export --run <run_id> --format <onnx|torchscript>
  • [ ] Run Comparison - kladml compare --runs run_001,run_002
  • [ ] Component Listing - kladml list <runs|datasets|architectures>
  • [ ] Better Error Messages - Helpful suggestions when something goes wrong
  • [ ] Progress Indicators - Real-time progress for long operations

v0.7.x - Preprocessing Pipelines

Composable data preprocessing:

  • [ ] YAML Pipeline Definition - Define preprocessing steps in config
  • [ ] Component Chaining - Link preprocessors with schema validation
  • [ ] CLI Command - kladml data process --pipeline <config>
  • [ ] Auto-split - Train/val/test splitting with reproducibility

v0.8.x - TUI Improvements

Enhancements to the existing Terminal UI:

  • [ ] Live Training Monitor - Real-time loss curves in terminal
  • [ ] Inline Plots - View evaluation plots directly in TUI
  • [ ] One-click Actions - Export, evaluate, compare from UI
  • [ ] Improved Navigation - Better keyboard shortcuts

v0.9.x - Smart Config Generator

Intelligent configuration assistant:

  • [ ] Data Analysis - Auto-detect data type (tabular, timeseries, image, text)
  • [ ] Task Suggestion - Recommend task based on data characteristics
  • [ ] Config Templates - Generate ready-to-use configuration files
  • [ ] CLI Command: 
    kladml config generate --data my_data.csv
# Output: configs/my_data_template.yaml

v0.10.x - Universal Quickstart (Optional)

One-command training for demos and onboarding:

  • [ ] Quickstart Command - kladml quickstart --data <path>
  • [ ] Auto-workflow - Detect → Generate config → Train → Evaluate
  • [ ] Pre-configured pipelines for common cases

Note: This is a thin wrapper around existing commands, primarily for marketing and demos.

v1.0.0 - Stable Release

First stable release with:

  • [ ] Comprehensive documentation
  • [ ] Full test coverage
  • [ ] API stability guarantees
  • [ ] Migration guides from previous versions

Future Ideas

These are being considered but not yet planned:

Feature Description
Hyperparameter Tuning Optuna integration for automated HPO
Distributed Training Multi-GPU support
Remote Execution Run on cloud infrastructure
Model Versioning Track model versions with metadata
Experiment Templates Reusable experiment configurations

Contributing

Want to help shape the roadmap?

Changelog

See CHANGELOG.md for detailed version history.