I’m trying to replicate the extract_hidden_states from Chapter 2 of the Transformers text and I’m getting the following error and I’m not sure what the issue is exactly:
TypeError: can't convert np.ndarray of type numpy.str_. The only supported types are: float64, float32, float16, complex64, complex128, int64, int32, int16, int8, uint8, and bool..
The code is exactly similar to the text and the only difference is I’m using local dataset usig csv’s rather than the a in-built dataset.
Here’s the code which causes the error and the full error-trace:
dataset_encoded.set_format("torch", columns=["input_ids", "attention_mask", "sentiment"])
dataset_hidden = dataset_encoded.map(extract_hidden_states, batched=True)
Error trace:
--------------------------------------------------------------------------
TypeError Traceback (most recent call last)
/home/harish/Desktop/SeaWord/action_extraction/hf/nbs/text_classification.ipynb Cell 29' in <cell line: 5>()
1 # converting the input_ids and attention_mask columns to the "torch" format
2 dataset_encoded.set_format("torch",
3 columns=["input_ids", "attention_mask", "sentiment"])
----> 5 dataset_hidden = dataset_encoded.map(extract_hidden_states, batched=True)
File ~/miniconda3/envs/dl/lib/python3.9/site-packages/datasets/dataset_dict.py:494, in DatasetDict.map(self, function, with_indices, input_columns, batched, batch_size, remove_columns, keep_in_memory, load_from_cache_file, cache_file_names, writer_batch_size, features, disable_nullable, fn_kwargs, num_proc, desc)
491 if cache_file_names is None:
492 cache_file_names = {k: None for k in self}
493 return DatasetDict(
--> 494 {
495 k: dataset.map(
496 function=function,
497 with_indices=with_indices,
498 input_columns=input_columns,
499 batched=batched,
500 batch_size=batch_size,
501 remove_columns=remove_columns,
502 keep_in_memory=keep_in_memory,
503 load_from_cache_file=load_from_cache_file,
504 cache_file_name=cache_file_names[k],
505 writer_batch_size=writer_batch_size,
506 features=features,
507 disable_nullable=disable_nullable,
508 fn_kwargs=fn_kwargs,
509 num_proc=num_proc,
510 desc=desc,
511 )
512 for k, dataset in self.items()
513 }
514 )
File ~/miniconda3/envs/dl/lib/python3.9/site-packages/datasets/dataset_dict.py:495, in <dictcomp>(.0)
491 if cache_file_names is None:
492 cache_file_names = {k: None for k in self}
493 return DatasetDict(
494 {
--> 495 k: dataset.map(
496 function=function,
497 with_indices=with_indices,
498 input_columns=input_columns,
499 batched=batched,
500 batch_size=batch_size,
501 remove_columns=remove_columns,
502 keep_in_memory=keep_in_memory,
503 load_from_cache_file=load_from_cache_file,
504 cache_file_name=cache_file_names[k],
505 writer_batch_size=writer_batch_size,
506 features=features,
507 disable_nullable=disable_nullable,
508 fn_kwargs=fn_kwargs,
509 num_proc=num_proc,
510 desc=desc,
511 )
512 for k, dataset in self.items()
513 }
514 )
File ~/miniconda3/envs/dl/lib/python3.9/site-packages/datasets/arrow_dataset.py:2092, in Dataset.map(self, function, with_indices, with_rank, 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)
2089 disable_tqdm = bool(logging.get_verbosity() == logging.NOTSET) or not utils.is_progress_bar_enabled()
2091 if num_proc is None or num_proc == 1:
-> 2092 return self._map_single(
2093 function=function,
2094 with_indices=with_indices,
2095 with_rank=with_rank,
2096 input_columns=input_columns,
2097 batched=batched,
2098 batch_size=batch_size,
2099 drop_last_batch=drop_last_batch,
2100 remove_columns=remove_columns,
2101 keep_in_memory=keep_in_memory,
2102 load_from_cache_file=load_from_cache_file,
2103 cache_file_name=cache_file_name,
2104 writer_batch_size=writer_batch_size,
2105 features=features,
2106 disable_nullable=disable_nullable,
2107 fn_kwargs=fn_kwargs,
2108 new_fingerprint=new_fingerprint,
2109 disable_tqdm=disable_tqdm,
2110 desc=desc,
2111 )
2112 else:
2114 def format_cache_file_name(cache_file_name, rank):
File ~/miniconda3/envs/dl/lib/python3.9/site-packages/datasets/arrow_dataset.py:518, in transmit_tasks.<locals>.wrapper(*args, **kwargs)
516 self: "Dataset" = kwargs.pop("self")
517 # apply actual function
--> 518 out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
519 datasets: List["Dataset"] = list(out.values()) if isinstance(out, dict) else [out]
520 for dataset in datasets:
521 # Remove task templates if a column mapping of the template is no longer valid
File ~/miniconda3/envs/dl/lib/python3.9/site-packages/datasets/arrow_dataset.py:485, in transmit_format.<locals>.wrapper(*args, **kwargs)
478 self_format = {
479 "type": self._format_type,
480 "format_kwargs": self._format_kwargs,
481 "columns": self._format_columns,
482 "output_all_columns": self._output_all_columns,
483 }
484 # apply actual function
--> 485 out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
486 datasets: List["Dataset"] = list(out.values()) if isinstance(out, dict) else [out]
487 # re-apply format to the output
File ~/miniconda3/envs/dl/lib/python3.9/site-packages/datasets/fingerprint.py:411, in fingerprint_transform.<locals>._fingerprint.<locals>.wrapper(*args, **kwargs)
405 kwargs[fingerprint_name] = update_fingerprint(
406 self._fingerprint, transform, kwargs_for_fingerprint
407 )
409 # Call actual function
--> 411 out = func(self, *args, **kwargs)
413 # Update fingerprint of in-place transforms + update in-place history of transforms
415 if inplace: # update after calling func so that the fingerprint doesn't change if the function fails
File ~/miniconda3/envs/dl/lib/python3.9/site-packages/datasets/arrow_dataset.py:2461, in Dataset._map_single(self, function, with_indices, with_rank, 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, cache_only)
2459 if drop_last_batch and i + batch_size > input_dataset.num_rows:
2460 continue
-> 2461 batch = input_dataset._getitem(
2462 slice(i, i + batch_size),
2463 decoded=False,
2464 )
2465 indices = list(
2466 range(*(slice(i, i + batch_size).indices(input_dataset.num_rows)))
2467 ) # Something simpler?
2468 try:
File ~/miniconda3/envs/dl/lib/python3.9/site-packages/datasets/arrow_dataset.py:1900, in Dataset._getitem(self, key, decoded, **kwargs)
1898 formatter = get_formatter(format_type, features=self.features, decoded=decoded, **format_kwargs)
1899 pa_subtable = query_table(self._data, key, indices=self._indices if self._indices is not None else None)
-> 1900 formatted_output = format_table(
1901 pa_subtable, key, formatter=formatter, format_columns=format_columns, output_all_columns=output_all_columns
1902 )
1903 return formatted_output
File ~/miniconda3/envs/dl/lib/python3.9/site-packages/datasets/formatting/formatting.py:539, in format_table(table, key, formatter, format_columns, output_all_columns)
537 else:
538 pa_table_to_format = pa_table.drop(col for col in pa_table.column_names if col not in format_columns)
--> 539 formatted_output = formatter(pa_table_to_format, query_type=query_type)
540 if output_all_columns:
541 if isinstance(formatted_output, MutableMapping):
File ~/miniconda3/envs/dl/lib/python3.9/site-packages/datasets/formatting/formatting.py:284, in Formatter.__call__(self, pa_table, query_type)
282 return self.format_column(pa_table)
283 elif query_type == "batch":
--> 284 return self.format_batch(pa_table)
File ~/miniconda3/envs/dl/lib/python3.9/site-packages/datasets/formatting/torch_formatter.py:67, in TorchFormatter.format_batch(self, pa_table)
65 def format_batch(self, pa_table: pa.Table) -> dict:
66 batch = self.numpy_arrow_extractor().extract_batch(pa_table)
---> 67 return self.recursive_tensorize(batch)
File ~/miniconda3/envs/dl/lib/python3.9/site-packages/datasets/formatting/torch_formatter.py:55, in TorchFormatter.recursive_tensorize(self, data_struct)
54 def recursive_tensorize(self, data_struct: dict):
---> 55 return map_nested(self._recursive_tensorize, data_struct, map_list=False)
File ~/miniconda3/envs/dl/lib/python3.9/site-packages/datasets/utils/py_utils.py:261, in map_nested(function, data_struct, dict_only, map_list, map_tuple, map_numpy, num_proc, types, disable_tqdm)
259 num_proc = 1
260 if num_proc <= 1 or len(iterable) <= num_proc:
--> 261 mapped = [
262 _single_map_nested((function, obj, types, None, True))
263 for obj in utils.tqdm(iterable, disable=disable_tqdm)
264 ]
265 else:
266 split_kwds = [] # We organize the splits ourselve (contiguous splits)
File ~/miniconda3/envs/dl/lib/python3.9/site-packages/datasets/utils/py_utils.py:262, in <listcomp>(.0)
259 num_proc = 1
260 if num_proc <= 1 or len(iterable) <= num_proc:
261 mapped = [
--> 262 _single_map_nested((function, obj, types, None, True))
263 for obj in utils.tqdm(iterable, disable=disable_tqdm)
264 ]
265 else:
266 split_kwds = [] # We organize the splits ourselve (contiguous splits)
File ~/miniconda3/envs/dl/lib/python3.9/site-packages/datasets/utils/py_utils.py:197, in _single_map_nested(args)
195 # Singleton first to spare some computation
196 if not isinstance(data_struct, dict) and not isinstance(data_struct, types):
--> 197 return function(data_struct)
199 # Reduce logging to keep things readable in multiprocessing with tqdm
200 if rank is not None and logging.get_verbosity() < logging.WARNING:
File ~/miniconda3/envs/dl/lib/python3.9/site-packages/datasets/formatting/torch_formatter.py:52, in TorchFormatter._recursive_tensorize(self, data_struct)
50 if data_struct.dtype == np.object: # pytorch tensors cannot be instantied from an array of objects
51 return [self.recursive_tensorize(substruct) for substruct in data_struct]
---> 52 return self._tensorize(data_struct)
File ~/miniconda3/envs/dl/lib/python3.9/site-packages/datasets/formatting/torch_formatter.py:44, in TorchFormatter._tensorize(self, value)
41 elif np.issubdtype(value.dtype, np.floating):
42 default_dtype = {"dtype": torch.float32}
---> 44 return torch.tensor(value, **{**default_dtype, **self.torch_tensor_kwargs})
TypeError: can't convert np.ndarray of type numpy.str_. The only supported types are: float64, float32, float16, complex64, complex128, int64, int32, int16, int8, uint8, and bool.