Summary

A collection of code quality improvements that make the codebase more maintainable, debuggable, and self-documenting. No behavioral changes — all improvements are structural.

Changes

1. Replace print() with logging module

• Converted all print() calls in tinyml-modelmaker to use Python's logging module
• Enables log level control, file output, and integration with external logging frameworks

2. Extract duplicated path resolution

• Factored out identical path resolution logic from timeseries/runner.py and vision/runner.py into shared resolve_paths() utility
• Eliminates 120+ lines of duplicated code

3. Replace magic strings with named constants

• Added TRAINING_DEVICE_CPU, TRAINING_DEVICE_CUDA, TRAINING_DEVICE_MPS constants
• Replaced hardcoded strings like 'cpu', 'cuda', 'tinyml_tinyverse' with constants

4. Replace assert/sys.exit/raise string with proper exceptions

• assert statements used for input validation → raise ValueError
• sys.exit() calls → raise RuntimeError
• Bare string raises → typed exceptions with descriptive messages

5. Protocol definitions for component interfaces

• Added protocols.py with typing.Protocol definitions for ModelRunner, Trainer, DatasetLoader
• Enables static type checking and documents expected interfaces

6. Architecture documentation

• Added ARCHITECTURE.md with codebase structure diagram, component descriptions, and improvement analysis

Files Changed (22 files)

All within tinyml-modelmaker

Testing

• All changes are refactoring with no behavioral impact
• Existing test suite passes without modification