Sampler: IndexError out of bounds

[primus852]

• CTGAN version: 0.2.2
• Python version: 3.7 (Anaconda)
• Operating System: Windows 10

Description

I tried to load a sample tsv file (instead of the adult demo). Unfortunately it crashes with various variations of the file (always the same structure).

What I Did

data = pd.read_csv('https://github.com/DCMLab/ABC/raw/master/data/tsv/op.%20127%20No.%2012/op127_no12_mov1.tsv', delimiter='\t')
    # data = pd.read_csv('https://github.com/DCMLab/ABC/raw/master/data/tsv/op.%2018%20No.%203/op18_no3_mov3.tsv', delimiter='\t')
    # data = pd.read_csv('https://github.com/DCMLab/ABC/raw/master/data/tsv/op.%2059%20No.%209/op59_no9_mov3.tsv', delimiter='\t')

    discrete_columns = [
        'chord',
        'altchord',
        'timesig',
        'global_key',
        'local_key',
        'pedal',
        'numeral',
        'form',
        'relativeroot'
    ]

    ctgan = CTGANSynthesizer()
    ctgan.fit(data, data.columns, epochs=5)

    samples = ctgan.sample(1000)
    print(samples)

Traceback

Traceback (most recent call last):
  File "D:/Projects/Master/pythonProject/main.py", line 30, in <module>
    ctgan.fit(data, data.columns, epochs=5)
  File "C:\Users\XXX\anaconda3\lib\site-packages\ctgan\synthesizer.py", line 157, in fit
    data_sampler = Sampler(train_data, self.transformer.output_info)
  File "C:\Users\XXX\anaconda3\lib\site-packages\ctgan\sampler.py", line 28, in __init__
    tmp.append(np.nonzero(data[:, st + j])[0])
IndexError: index 1086 is out of bounds for axis 1 with size 1086

index varies from file to file

[fealho]

Hi @primus852, for this scenario, you should use CopulaGAN from SDV which is more user friendly and provides additional functionality on top of CTGAN.