Error in Preprocessing the data

[AchintyaX]

I was using the colab notesbook for training a model using wav2vec2forclassification. In the preprocessing step when I am running the following code -

train_dataset = train_dataset.map(
    preprocess_function,
    batch_size=10,
    batched=True,
   
)
eval_dataset = eval_dataset.map(
    preprocess_function,
    batch_size=10,
    batched=True,
    
)

I am getting into the following error, I guess it has something to do with hugging face datasets -

0%|          | 0/1765 [00:00<?, ?ba/s]
---------------------------------------------------------------------------
ArrowInvalid                              Traceback (most recent call last)
/tmp/ipykernel_29222/3011913806.py in <module>
      2 
      3 
----> 4 train_dataset = train_dataset.map(
      5     preprocess_function,
      6     batch_size=10,

~/.pyenv/versions/3.8.7/envs/bg_classifier/lib/python3.8/site-packages/datasets/arrow_dataset.py in map(self, function, with_indices, input_columns, batched, batch_size, drop_last_batch, remove_columns, keep_in_memory, load_from_cache_file, cache_file_name, writer_batch_size, features, disable_nullable, fn_kwargs, num_proc, suffix_template, new_fingerprint, desc)
   1667 
   1668         if num_proc is None or num_proc == 1:
-> 1669             return self._map_single(
   1670                 function=function,
   1671                 with_indices=with_indices,

~/.pyenv/versions/3.8.7/envs/bg_classifier/lib/python3.8/site-packages/datasets/arrow_dataset.py in wrapper(*args, **kwargs)
    183         }
    184         # apply actual function
--> 185         out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
    186         datasets: List["Dataset"] = list(out.values()) if isinstance(out, dict) else [out]
    187         # re-apply format to the output

~/.pyenv/versions/3.8.7/envs/bg_classifier/lib/python3.8/site-packages/datasets/fingerprint.py in wrapper(*args, **kwargs)
    395             # Call actual function
    396 
--> 397             out = func(self, *args, **kwargs)
    398 
    399             # Update fingerprint of in-place transforms + update in-place history of transforms

~/.pyenv/versions/3.8.7/envs/bg_classifier/lib/python3.8/site-packages/datasets/arrow_dataset.py in _map_single(self, function, with_indices, input_columns, batched, batch_size, drop_last_batch, remove_columns, keep_in_memory, load_from_cache_file, cache_file_name, writer_batch_size, features, disable_nullable, fn_kwargs, new_fingerprint, rank, offset, disable_tqdm, desc)
   2036                             else:
   2037                                 batch = cast_to_python_objects(batch)
-> 2038                                 writer.write_batch(batch)
   2039                 if update_data and writer is not None:
   2040                     writer.finalize()  # close_stream=bool(buf_writer is None))  # We only close if we are writing in a file

~/.pyenv/versions/3.8.7/envs/bg_classifier/lib/python3.8/site-packages/datasets/arrow_writer.py in write_batch(self, batch_examples, writer_batch_size)
    401             typed_sequence = OptimizedTypedSequence(batch_examples[col], type=col_type, try_type=col_try_type, col=col)
    402             typed_sequence_examples[col] = typed_sequence
--> 403         pa_table = pa.Table.from_pydict(typed_sequence_examples)
    404         self.write_table(pa_table, writer_batch_size)
    405 

~/.pyenv/versions/3.8.7/envs/bg_classifier/lib/python3.8/site-packages/pyarrow/table.pxi in pyarrow.lib.Table.from_pydict()

~/.pyenv/versions/3.8.7/envs/bg_classifier/lib/python3.8/site-packages/pyarrow/array.pxi in pyarrow.lib.asarray()

~/.pyenv/versions/3.8.7/envs/bg_classifier/lib/python3.8/site-packages/pyarrow/array.pxi in pyarrow.lib.array()

~/.pyenv/versions/3.8.7/envs/bg_classifier/lib/python3.8/site-packages/pyarrow/array.pxi in pyarrow.lib._handle_arrow_array_protocol()

~/.pyenv/versions/3.8.7/envs/bg_classifier/lib/python3.8/site-packages/datasets/arrow_writer.py in __arrow_array__(self, type)
    105                 out = numpy_to_pyarrow_listarray(self.data)
    106             else:
--> 107                 out = pa.array(self.data, type=type)
    108             if trying_type and out[0].as_py() != self.data[0]:
    109                 raise TypeError(

~/.pyenv/versions/3.8.7/envs/bg_classifier/lib/python3.8/site-packages/pyarrow/array.pxi in pyarrow.lib.array()

~/.pyenv/versions/3.8.7/envs/bg_classifier/lib/python3.8/site-packages/pyarrow/array.pxi in pyarrow.lib._sequence_to_array()

~/.pyenv/versions/3.8.7/envs/bg_classifier/lib/python3.8/site-packages/pyarrow/error.pxi in pyarrow.lib.pyarrow_internal_check_status()

~/.pyenv/versions/3.8.7/envs/bg_classifier/lib/python3.8/site-packages/pyarrow/error.pxi in pyarrow.lib.check_status()

ArrowInvalid: Can only convert 1-dimensional array values

[AchintyaX]

@m3hrdadfi I feel it might have something to do with the version of hugging face datasets being used in this implementation, can you share the version since directly installing from git might result in some conflicts

[hamzakhalem]

i have the same problem here ... if you've got the solution please share it with me

[Alymostafa]

@AchintyaX
@hamzakhalem
The problem is the torchaudio.load you will find that it returns 2d array like that (2,82585) but it should be like that (82585,). So the solution is to change this code block :

def speech_file_to_array_fn(path):
    speech_array, sampling_rate = torchaudio.load(path)
    resampler = torchaudio.transforms.Resample(sampling_rate, target_sampling_rate)
    speech = resampler(speech_array).squeeze().numpy()
    return speech

To that:

def speech_file_to_array_fn(path):
    speech_array, sampling_rate = torchaudio.load(path)
    speech_array = speech_array[0].numpy().squeeze()
    resampler = librosa.resample(speech_array, sampling_rate, target_sampling_rate)
    speech = resampler.squeeze()
    return speech

it works for me.

[eaedkbaamtu]

I have the same bug but the first and the second array are different, are you sure that select the [0] is not a splitting of the sample ?

[Alymostafa]

Yes, it's not splitting. To be sure try librosa.load instead of torch.audio, it will output a single array that is equal [0].

I have the same bug but the first and the second array are different, are you sure that select the [0] is not a splitting of the sample ?

[eaedkbaamtu]

ok, but when I checked the both arrays were different that why I asked.