Azure Machine Learning best model's update paramete with python pickle

윤 광섭 1

i am trying to Azure Machine Learning best model's update parameter with python pickle type

best model trained by xgboost.

i load model through pickle and tried to change parameter.

import joblib

a=joblib.load("model.pkl")

a.pipeline.set_params(XGBoostClassifier__base_score=0.6)

but it doesn't work.

however

a.pipeline.set_params(memory='n')

this code is works.

let me know how to change model's parameter.

romungi-MSFT 42,986 Reputation points Microsoft Employee

2020-06-11T11:28:22.917+00:00

@87568746 Are you following any steps that you can share to create this model and load it? Is there any error when you use the first parameter? Essentially, it looks like the model returns the python object stored in it. In this case, the model that it is trained based on which the parameters that are to be set depend on it.
윤 광섭 1 Reputation point

2020-06-12T00:32:30.66+00:00

the pickle file made by Azure ML automl service.

i tried to develop model by tunning some parameter by set_params method

as you know this pickle file has 'Pipeline', 'y_transformer', 'y_transformer_name' key.

and Pipeline has parameter setting

but as i wrote,

a.pipeline.set_params (memory = 'n')

this code is work.

memory parameter is changed.

but

a.pipeline.set_params (XGBoostClassifier__base_score = 0.6)

this code is not work.

please let me know how to change model's hyperparameter
romungi-MSFT 42,986 Reputation points Microsoft Employee

2020-06-23T08:10:04.547+00:00

@윤 광섭 Based on some available info on github forums I came across this issue where xgboost is not installed by default in training which caused issues to users after training. Could you please check if your case is similar?

Also, is it possible to check get_params() is available to check if this param can be set?