AutoPytorch selects only Dummy model.

[shabir1]

I run AutoPytorch for 10 hours on a dataset with 1.5 Lac rows and 13 columns, but it was not able to find the model it selects the dummy model.

Code

model = TabularRegressionTask(
                    ensemble_size=5,
                    ensemble_nbest=5,
                    max_models_on_disc=5,
                )
model.search(
                optimize_metric='r2',
                X_train=X_train,
                y_train=y_train,
                total_walltime_limit=36000,
                memory_limit=5092,
                enable_traditional_pipeline=False
            )

Output

After 10 hours it gives me

Dummy Model

Local environment

• Ubuntu
• Python: 3.8
• autopytorch:0.2

[SaleemAkhtarAngstrom]

Also getting this, not sure why

[ravinkohli]

You'll get the dummy model in case no other configuration finished successfully. So, could you check the log file to see what is going wrong?

[SaleemAkhtarAngstrom]

Mine was a CUDA error, removed GPU access from the container I was developing in and it was fine. Also I wasn't allowing enough predict time for my dataset, changing the above two things fixed my issues

[shabir1]

Is there any option to set GPU off in AutoPytorch?