update model_training process (train and export)

[sunya-ch]

** Rebase from the PR #172 **

Also, refer to archived pipeline that should be merged on kepler-model-db first by sustainable-computing-io/kepler-model-db#16

Need

• fix extractor/profileisolator bug and update test case #172
• [U] models (v0.6 nx12 std_v0.6) kepler-model-db#16

to be merged first.

---

This PR includes

• document (README) auto-generation on export script (auto-generate document mentioned in enrich PR template for adding trained pipeline kepler-model-db#14)
• sort metadata with feature group first
• handle missing trainer class in the list
• update get_url to also handle weight url
• add assure_path disable option for getting remote machine path
• add trainers_with_weight list to list model with weight
• update train script to use only stressng benchmark
• update stressng benchmark according to the result in community votes (19/09/2023)

Signed-off-by: Sunyanan Choochotkaew [email protected]

[marceloamaral]

Is this selecting the data only for training?

If it is also for the validation: should we also use the coremark results for testing to verify the accuracy of the model with different workload?

[sunya-ch]

It will be shuffled and use 10% of it for validation.

kepler-model-server/src/train/trainer/__init__.py

Line 28 in 6b4f716

def normalize_and_split(X_values, y_values, scaler, test_size=0.1):

We need to refactor the code to have a fixed validation dataset. Should be created for a separate issue.

[rootfs]

@sunya-ch the CI passed now.