Postgres-MADlib predictions is taking longer than training

Question

I am training my data using following code:

start_time := clock_timestamp();
  PERFORM madlib.create_nb_prepared_data_tables( 'nb_training',
                                                 'class', 
                                                 'attributes', 
                                                 'ARRAY[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57]', 
                                                 57, 
                                                 'categ_feature_probs', 
                                                 'numeric_attr_params', 
                                                 'class_priors' 
                                               );
  training_time := 1000* (extract(epoch FROM clock_timestamp()) - extract(epoch FROM start_time));

And my prediction code goes as follows:

start_time := clock_timestamp();
  PERFORM madlib.create_nb_probs_view( 'categ_feature_probs', 
                                       'class_priors', 
                                       'nb_testing', 
                                       'id', 
                                       'attributes', 
                                       57, 
                                       'numeric_attr_params', 
                                       'probs_view' );

select * from probs_view
prediction_time := 1000 * (extract(epoch FROM clock_timestamp()) - extract(epoch FROM start_time));

The training data is containing 450000 records were as testing dataset contains 50000 records.

Still, my average training_time is around 17173 ms where as prediction_time is 26481 ms. As per my understanding of naive bayes, the prediction_time should be less than training_time. What am I doing wrong here?

Answer 1

Naive Bayes classification is in early stage for MADlib which means that interface and implementation are preliminary at this stage. There are a bunch of open JIRAs which tells me it needs some effort before being promoted to a top level module.