Blocked Algorithms still run for Azure AutoML Experiment

Question

I have been trying to run an AutoML Forecasting Experiment with only allowing one algorithm (FBProphet) to run and no other supported algorithms. The issue I run into is that even though I specify the blocked algorithms, they still run in the experiment taking up unnecessary runtime. For eg, my experiment should run only for 1-2 hours but it ends up running for 24-30 hours because it still runs the undesired algorithms. Is there any way I can stop making the blocked algorithms from running in my experiment so I can save up on significant runtime? I have attached a screenshot and my AutoML config code to help understand this issue better.

Code:

n_test_periods = 60
blocked_algos = ['ExtremeRandomTrees','DecisionTree','ElasticNet','LassoLars']
time_series_settings = {
    'time_column_name': time_column_name,
    'grain_column_names': grain_column_names ,
    'forecast_horizon': n_test_periods
}

automl_config = AutoMLConfig(task='forecasting',
                             debug_log='Logs/prophet_forecasting_errors.log',
                             primary_metric='normalized_mean_absolute_error',
                             training_data=train_data,
                             label_column_name=target_column_name,
                             compute_target=compute_target,
                             featurization= 'off',
                             blocked_model = blocked_algos,
                             allowed_models = ['Prophet'],
                             n_cross_validations= 3,
                             verbosity=logging.INFO,
                             max_cores_per_iteration=6,
                             **time_series_settings)

remote_run = experiment.submit(automl_config, show_output=True)

Screenshot of the Experiment: (This took 32h when it should ideally take 56 mins)

github1

Can you please add more details about the python version and Azure ML SDK version. — Ram, Nov 23 '20 at 07:49
Hi, the python version is 3.6 and the Azure ML SDK Version is 1.9.0. — Megha Lokanadham, Nov 24 '20 at 18:05

Ram · Answer 1 · 2020-11-26T03:18:12.363

0

It's always good to have the latest version if possible, latest version is 1.18.0 as of today

Correct the name of the parameter( blocked_model) to "blocked_models" and keep both "blocked_models" and "allowed_models".

edited Nov 26 '20 at 03:18

answered Nov 25 '20 at 10:52

Ram

2,459
1
7
14

Blocked Algorithms still run for Azure AutoML Experiment

Code:

Screenshot of the Experiment: (This took 32h when it should ideally take 56 mins)

1 Answers1