Calling sampling with conditions and without num_rows crashes

[csala]

Environment Details

• SDV version: 0.12.1

Error Description

When calling the sample method of a tabular model without passing any arguments, the model produces
as many rows as it saw during the fit phase:

In [1]: import sdv
   ...: 
   ...: data = sdv.demo.load_tabular_demo('student_placements')
   ...: 
   ...: gc = sdv.tabular.GaussianCopula()
   ...: gc.fit(data.head(10))
   ...: 
   ...: gc.sample()
Out[1]: 
   student_id gender  second_perc  high_perc high_spec  degree_perc  ... mba_perc   salary  placed  start_date   end_date  duration
0       17270      M        59.18      60.63  Commerce        78.04  ...    51.58      NaN   False         NaT        NaT       NaN
1       17266      M        57.90      49.20   Science        69.00  ...    52.84      NaN   False         NaT        NaT       NaN
2       17266      M        58.67      64.73   Science        71.14  ...    62.49  22400.0    True  2020-01-12 2020-05-18       4.0
3       17268      M        79.76      58.37   Science        70.44  ...    56.47  33400.0    True  2020-03-23 2020-07-22       3.0
4       17269      M        76.12      67.72      Arts        69.53  ...    57.44  28800.0    True  2020-02-02 2020-07-17       4.0
5       17270      M        74.14      68.20  Commerce        60.42  ...    55.82  27000.0    True  2020-03-11 2020-08-14       6.0
6       17268      F        57.94      64.93  Commerce        68.47  ...    51.58      NaN   False         NaT        NaT       NaN
7       17268      M        53.20      54.04   Science        71.75  ...    51.58      NaN   False         NaT        NaT       NaN
8       17271      F        49.14      58.24  Commerce        76.70  ...    56.37      NaN   False         NaT        NaT       NaN
9       17268      F        68.63      49.20   Science        74.70  ...    51.58      NaN   False         NaT        NaT       NaN

[10 rows x 17 columns]

When sampling conditionally passing a dict AND a number of rows to sample, everything works as expected:

In [2]: gc.sample(num_rows=5, conditions={'gender': 'M'})
Out[2]: 
   student_id gender  second_perc  high_perc high_spec  degree_perc  ... mba_perc   salary  placed  start_date   end_date  duration
0       17267      M        62.39      81.37      Arts        75.29  ...    58.84  24000.0    True  2020-01-30 2020-07-11       5.0
1       17270      M        76.24      64.03   Science        77.89  ...    52.53  29100.0    True  2020-01-28 2020-07-04       4.0
2       17268      M        70.18      87.62  Commerce        62.72  ...    62.80  26100.0    True  2020-03-02 2020-08-23       4.0
3       17267      M        75.90      72.41   Science        74.46  ...    58.08  27800.0    True  2020-02-05 2020-07-14       3.0
4       17269      M        71.36      58.67   Science        63.50  ...    62.55  26000.0    True  2020-02-01 2020-05-20       3.0

[5 rows x 17 columns]

However, when sampling with a conditions dict WITHOUT giving num_rows, the sample call crashes.

In [3]: gc.sample(conditions={'gender': 'M'})
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
/home/ubuntu/.virtualenvs/sdv-test/lib/python3.7/site-packages/sdv/tabular/base.py in _make_conditions_df(self, conditions, num_rows)
    345             try:
--> 346                 conditions = pd.DataFrame(conditions)
    347             except ValueError:

/home/ubuntu/.virtualenvs/sdv-test/lib/python3.7/site-packages/pandas/core/frame.py in __init__(self, data, index, columns, dtype, copy)
    467         elif isinstance(data, dict):
--> 468             mgr = init_dict(data, index, columns, dtype=dtype)
    469         elif isinstance(data, ma.MaskedArray):

/home/ubuntu/.virtualenvs/sdv-test/lib/python3.7/site-packages/pandas/core/internals/construction.py in init_dict(data, index, columns, dtype)
    282         ]
--> 283     return arrays_to_mgr(arrays, data_names, index, columns, dtype=dtype)
    284 

/home/ubuntu/.virtualenvs/sdv-test/lib/python3.7/site-packages/pandas/core/internals/construction.py in arrays_to_mgr(arrays, arr_names, index, columns, dtype, verify_integrity)
     77         if index is None:
---> 78             index = extract_index(arrays)
     79         else:

/home/ubuntu/.virtualenvs/sdv-test/lib/python3.7/site-packages/pandas/core/internals/construction.py in extract_index(data)
    386         if not indexes and not raw_lengths:
--> 387             raise ValueError("If using all scalar values, you must pass an index")
    388 

ValueError: If using all scalar values, you must pass an index

During handling of the above exception, another exception occurred:

TypeError                                 Traceback (most recent call last)
<ipython-input-4-d253bdf9697f> in <module>
----> 1 gc.sample(conditions={'gender': 'M'})

/home/ubuntu/.virtualenvs/sdv-test/lib/python3.7/site-packages/sdv/tabular/base.py in sample(self, num_rows, max_retries, max_rows_multiplier, conditions, float_rtol, graceful_reject_sampling)
    443 
    444         # convert conditions to dataframe
--> 445         conditions = self._make_conditions_df(conditions, num_rows)
    446 
    447         # validate columns

/home/ubuntu/.virtualenvs/sdv-test/lib/python3.7/site-packages/sdv/tabular/base.py in _make_conditions_df(self, conditions, num_rows)
    346                 conditions = pd.DataFrame(conditions)
    347             except ValueError:
--> 348                 conditions = pd.DataFrame([conditions] * num_rows)
    349 
    350         elif not isinstance(conditions, pd.DataFrame):

TypeError: can't multiply sequence by non-int of type 'NoneType'

Expected Behavior

Since sampling without any arguments produces the same number of rows as seen in the input data, one would expect that passing a conditions dict without num_rows would achieve the same result.