Create sample_remaining_columns() method

[npatki]

Problem Description

Let's make sampling more user friendly. We can create multiple methods for different user needs.

A new sample_remaining_columns() method can address conditional sampling with a given Dataframe.

Expected behavior

Parameters:

• (required) known_columns: A pandas.DataFrame with the columns that are already known -- this specifies number of rows
• max_tries: renamed from existing max_retries param (default: 100)
• batch_size_per_try: Number of rows to sample per try (default: 10x requested num)
• randomize_samples will determine whether or not there should be a fixed seed (default: True)

>>> model.sample_remaining_columns(known_columns=my_dataframe)

Error Handling

Running out of tries

# Always gracefully reject sample (ie return any rows that are sampled)
>>> synthetic_data = model.sample_remaining_columns(known_columns=my_dataframe)
Warning: Only able to sample 75 of the requested rows. To sample more rows, try increasing max_tries
(currently: 100) or increasing batch_size_per_try (currently: 10000). Note that increasing these values will also
increase the sampling time.

# Error if we weren't able to sample any rows
>>> synthetic_data = model.sample_remaining_columns(known_columns=my_dataframe)
Error: Unable to sample any rows for the given conditions. Try increasing max_tries
(currently: 100) or increasing batch_size_per_try (currently: 10000). Note that increasing these values will also
increase the sampling time.

Checking for invalid input

# Unexpected column name
>>> synthetic_data = model.sample_remaining_columns(known_columns=invalid_dataframe)
Error: Unexpected column name 'New_Column'. Use a column name that was present in the original data.
 
# Unexpected categorical column value
>>> synthetic_data = model.sample_remaining_columns(known_columns=invalid_dataframe_2)
Error: Unexpected value 'NEW_VALUE' in column 'New_Column'. Use a value that was present in the original data.