Report fitting progress for the HMASynthesizer

[npatki]

Problem Description

The HMASynthesizer does not offer a verbosity option today. If there is a large dataset, it may take some time to learn the model (i.e. fit the data) and I am just waiting with a blank screen. It would be better if there was a progress report I could see.

Expected behavior

Add a verbose parameter to the initialization of the HMASynthesizer. If set to True, then it should print out the progress during all fitting stages. Let's set the parameter to True by default. Users can always turn this off.

>>> synth = HMASynthesizer(metadata, verbose=True)
>>> synth.fit(data)

Preprocess Tables: 100%|█████████████████████████████| 3/3 [00:33<00:10, 229.00it/s]

Learning relationships:
(1/2) Tables 'A' and 'B' ('user_id'): 100%|█████████████████████████████| 1000/1000 [00:33<00:10, 229.00it/s]
(2/2) Tables 'B' and 'C' ('session_id'): 100%|█████████████████████████████| 1200/1200 [00:33<00:10, 229.00it/s]

Modeling Tables: 100%|█████████████████████████████| 1/1 [00:33<00:10, 229.00it/s]

Additional context (fit)

Let r be the total number of relationships in the multi table schema. Then the total number of steps is r+2. Each step will have a new progress bar.

• Step 1: The first step will be preprocessing. The progress bar should increment for each table that we preprocess. (https://github.com/sdv-dev/SDV/blob/master/sdv/multi_table/base.py#L307)
• Next r steps: For each relationship, show a new progress bar. The bar should increment as we iterate over each parent row to compute the extension over a child’s foreign key (https://github.com/sdv-dev/SDV/blob/master/sdv/multi_table/hma.py#L63)
• Final step: The final step is modeling each root table. The progress bar should increment as we iterate over each root table that is being fitted (https://github.com/sdv-dev/SDV/blob/master/sdv/multi_table/hma.py#L186)

Additional Context (preprocress)

When preprocessing, only step 1 is performed.

>>> synth = HMASynthesizer(metadata, verbose=True)
>>> processed_data = synth.preprocess(data)

Preprocess Tables: 100%|█████████████████████████████| 3/3 [00:33<00:10, 229.00it/s]

Additional Context (fit_processed_data)

When fitting processed data, the preprocessing is already done, so there are only r+1 remaining steps.

>>> synth.fit_processed_data(processed_data)
Learning relationships:
(1/2) Tables 'A' and 'B' ('user_id'): 100%|█████████████████████████████| 1000/1000 [00:33<00:10, 229.00it/s]
(2/2) Tables 'B' and 'C' ('session_id'): 100%|█████████████████████████████| 1200/1200 [00:33<00:10, 229.00it/s]

Modeling Tables: 100%|█████████████████████████████| 1/1 [00:33<00:10, 229.00it/s]