[WIP] Support local xgboost regressor #1007

rootfs · 2023-10-16T20:32:15Z

the xgboost library PR #1041 needs to be merged first

marceloamaral · 2023-10-17T08:23:50Z

pkg/model/estimator/local/lr_test.go

@@ -243,7 +244,7 @@ var _ = Describe("Test LR Weight Unit", func() {
 			// Estimate Container Components Power using Linear Regression
 			modelWeightFilepath := config.GetDefaultPowerModelURL(types.DynPower.String(), types.ComponentEnergySource)
 			initModelURL := "https://raw.githubusercontent.com/sustainable-computing-io/kepler-model-db/main/models/v0.6/nx12/std_v0.6/rapl/DynPower/BPFOnly/SGDRegressorTrainer_1.json"


Probably need to update the model here, this model is SGD, not sure it it if compatible with XGBoost.
Did you try to locally?

Might be also good to change the name of the files and other references from lr to xgb.

marceloamaral

Looks very promising!

sunya-ch · 2023-11-01T05:48:47Z

@rootfs I updated the xgboost to the main pipeline and upload the trained model on my local repo. Please use it first for testing the WIP code by the follow config.

apiVersion: v1
kind: ConfigMap
metadata:
  name: kepler-cfm
  namespace: kepler
data:
  MODEL_CONFIG: |
    NODE_TOTAL_INIT_URL =https://raw.githubusercontent.com/sunya-ch/kepler-model-db/trl-nx12/models/v0.7/nx12/std_v0.7/acpi/AbsPower/BPFOnly/XgboostFitTrainer_1.json
    NODE_COMPONENTS_INIT_URL=https://raw.githubusercontent.com/sunya-ch/kepler-model-db/trl-nx12/models/v0.7/nx12/std_v0.7/rapl/AbsPower/BPFOnly/XgboostFitTrainer_1.json

The json of xgboost is in NormalizedNumericalFeature.Weight
You may redefine NormalizedNumericalFeature.Weight to interface{} instead of float and handle it separately for each type.
Not sure but you may need to temporarily save the extracted Weight in json format can read it from the function C.LoadModelFromJSON.

rootfs · 2023-11-01T14:13:17Z

@rootfs I updated the xgboost to the main pipeline and upload the trained model on my local repo. Please use it first for testing the WIP code by the follow config.
apiVersion: v1
kind: ConfigMap
metadata:
  name: kepler-cfm
  namespace: kepler
data:
  MODEL_CONFIG: |
    NODE_TOTAL_INIT_URL =https://raw.githubusercontent.com/sunya-ch/kepler-model-db/trl-nx12/models/v0.7/nx12/std_v0.7/acpi/AbsPower/BPFOnly/XgboostFitTrainer_1.json
    NODE_COMPONENTS_INIT_URL=https://raw.githubusercontent.com/sunya-ch/kepler-model-db/trl-nx12/models/v0.7/nx12/std_v0.7/rapl/AbsPower/BPFOnly/XgboostFitTrainer_1.json
The json of xgboost is in NormalizedNumericalFeature.Weight You may redefine NormalizedNumericalFeature.Weight to interface{} instead of float and handle it separately for each type. Not sure but you may need to temporarily save the extracted Weight in json format can read it from the function C.LoadModelFromJSON.

thank you @sunya-ch

Signed-off-by: Huamin Chen <[email protected]>

Bumps [tj-actions/verify-changed-files](https://github.com/tj-actions/verify-changed-files) from 11.1 to 13.1. - [Release notes](https://github.com/tj-actions/verify-changed-files/releases) - [Changelog](https://github.com/tj-actions/verify-changed-files/blob/main/HISTORY.md) - [Commits](tj-actions/verify-changed-files@v11.1...v13.1) --- updated-dependencies: - dependency-name: tj-actions/verify-changed-files dependency-type: direct:production update-type: version-update:semver-major ... Signed-off-by: dependabot[bot] <[email protected]>