Training Configurations Add/Update implementation #4873
Conversation
github-actions
bot
added
TEST
Docker Image Sizes
|
📊 Test coverage report
|
There was a problem hiding this comment.
Pull Request Overview
This PR implements GET and POST operations for training configurations, adding a dedicated service layer and endpoints. The implementation replaces placeholder endpoints with full functionality that supports retrieving and updating training configurations for projects and specific model architectures.
- Added complete service layer with
TrainingConfigurationServicefor managing training configurations - Implemented database repository and schema for persisting training configurations
- Created dedicated API endpoints with proper error handling and validation
Reviewed Changes
Copilot reviewed 11 out of 11 changed files in this pull request and generated 3 comments.
Show a summary per file
| File | Description |
|---|---|
test_training_configurations.py |
Unit tests for training configuration endpoints |
test_training_configuration_service.py |
Integration tests for the training configuration service |
test_initial_schema_migration.py |
Updated test to verify new training_configurations table |
training_configuration_service.py |
Core service implementing configuration retrieval and update logic |
training_configuration.py |
Enhanced schema with new factory methods and detailed examples |
training_configuration_repo.py |
Repository for database operations on training configurations |
schema.py |
Added TrainingConfigurationDB model with unique constraints |
training_configurations.py |
New dedicated API endpoints for training configuration operations |
projects.py |
Removed placeholder training configuration endpoints |
dependencies.py |
Added dependency injection for training configuration service |
da385d690aae_schema.py |
Database migration to create training_configurations table |
Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.
application/backend/app/alembic/versions/da385d690aae_schema.py
Outdated
Show resolved
Hide resolved
application/backend/app/repositories/training_configuration_repo.py
Outdated
Show resolved
Hide resolved
application/backend/tests/integration/services/test_training_configuration_service.py
Outdated
Show resolved
Hide resolved
application/backend/tests/integration/services/test_training_configuration_service.py
Outdated
Show resolved
Hide resolved
| """ | ||
| existing = self.get_by_project_and_model_architecture(project_id, model_architecture_id) | ||
|
|
||
| if existing: |
There was a problem hiding this comment.
Do we need existing check - will it not be able to detect whether to generate update/insert based on provided ids by default?
application/backend/app/services/training_configuration_service.py
Outdated
Show resolved
Hide resolved
| "evaluation": {"config": {"metrics": ["accuracy", "precision", "recall"], "validation_split": 0.2}}, | ||
| "dataset_preparation": { | ||
| "augmentation": { | ||
| "topdown_affine": None, |
There was a problem hiding this comment.
Null-valued parameters should be filtered out from the response, because in this context None represents a parameter that's not applicable to the algorithm.
I believe Geti already implements a similar feature. This is an example of response, as you can see there are some empty sections (-> this algorithm doesn't support any augmentation)
| "gaussian_noise": {"enable": False, "mean": 0, "sigma": 0.1, "probability": 0.5}, | ||
| "tiling": {"enable": False, "adaptive_tiling": True, "tile_size": 400, "tile_overlap": 0.2}, | ||
| } | ||
| }, |
There was a problem hiding this comment.
I think the subset_split category is missing?
| } | ||
| }, | ||
| "training": { | ||
| "max_epochs": 200, |
There was a problem hiding this comment.
The REST endpoint should return a richer representation of each parameter, including the name, description, type, min/max and default values, etc... These details are necessary to the UI to correctly display the parameter.
For reference, here is how the max_epochs parameter looks like in Geti response:
…nfiguration retrieval, and use model_copy for model updating
3 participants
Summary
Add the implementation for GET and POST of the training configuration
How to test
http://geti-tune.localhost/api/projects/165f79f9-fba5-415b-adda-7e6548ed2c40/training_configuration
Checklist