refactor: reorganize evaluation scripts structure - #127
Conversation
- Move evaluator.py to src/evaluation/base/ - Move metrics files to src/evaluation/metrics/ - Move result files to src/evaluation/results/ - Update all imports to reflect new structure - Rename complete_hierarchy_results.json to hierarchical_evaluation_results.json - Update output paths in evaluation scripts Resolves #124
| from src.evaluation.evaluator import EvaluationConfig, ModelEvaluator | ||
| from src.evaluation.base.evaluator import EvaluationConfig, ModelEvaluator | ||
| from src.inference import MobileNetInference, ResNetInference | ||
| from src.onnx_model import OnnxClassifierInferenceBase as OnnxModel |
There was a problem hiding this comment.
could you please import as is, not as OnnxModel? Just use OnnxClassifierInferenceBase
There was a problem hiding this comment.
Did you make sure that the results remain the same?
There was a problem hiding this comment.
What do you mean? Are the results the same? I just arranged the folders, I didn't touch the logic at all.
And we haven't yet integrated into the project that the model will predict better with a black background, so there shouldn't be a change, and indeed, when we do that, we'll run the RUN_EVALOTION again and see the best results.
There was a problem hiding this comment.
Please note that it's a part of the ticket to run the code before your changes, and run after your changes, and at the end to compare the results. The goal behind it is to ensure we didn't break anything - but not to improve results etc. Please do that
There was a problem hiding this comment.
Did you make sure that the results remain the same?
…Base - Remove 'as OnnxModel' alias as requested in code review - Use OnnxClassifierInferenceBase directly for better readability Addresses reviewer feedback
- Updated evaluation_results.json with PyTorch model performance - ResNet18: 26.2% accuracy, 84.9ms latency - MobileNet: 27.2% accuracy, 55.1ms latency - Results generated after dataset/preparation and src/evaluation refactoring
2 participants
Resolves #124