mljar / mljar-supervised

Python package for AutoML on Tabular Data with Feature Engineering, Hyper-Parameters Tuning, Explanations and Automatic Documentation
https://mljar.com
MIT License
3.05k stars 406 forks source link

Problems when loading a stacked ensemble #318

Closed RafaD5 closed 3 years ago

RafaD5 commented 3 years ago

I installed the dev branch yesterday and trained a model with the following parameters

automl = AutoML(
    mode='Perform', 
    results_path=str(model_dir),
    total_time_limit=60*60*16,
    eval_metric='mape',
    stack_models=True
)

But when I try to load the trained model I get an error. The best model is an Ensemble_Stacked and there is no file Ensemble_Stacked/framework.json. I went to the file best_model.txt and replaced Ensemble_Stacked to Ensemble and it did load that model, but then I got another error: image

pplonski commented 3 years ago

That's because of an update in the Xgboost package. There is a need to retrain the AutoML. Apologize for this! I'm doing a lot of changes in the package. That's why the package is not backward compatible.

RafaD5 commented 3 years ago

I trained with the new version of mljar but to load the model I was using my local version. I'm sorry!

To be clear, all is working fine.