[BUG] Token Classification fails on value error text_column

[jmccrae]

Prerequisites

• I have read the documentation.
• I have checked other issues for similar problems.

Backend

Local

Interface Used

CLI

CLI Command

from autotrain.params import TokenClassificationParams
from autotrain.project import AutoTrainProject


params = TokenClassificationParams(
    model="FacebookAI/roberta-base",
    data_path="data")    

backend = "local"
project = AutoTrainProject(params=params, backend=backend, process=True)
project.create()

UI Screenshots & Parameters

No response

Error Logs

Traceback (most recent call last):
  File "/home/jmccrae/scratch/wikilinks_autotrain/apply_autotrain.py", line 13, in <module>
    project.create()
  File "/home/jmccrae/.cache/pypoetry/virtualenvs/wikilinks-autotrain-NvKt9JsM-py3.12/lib/python3.12/site-packages/autotrain/project.py", line 567, in create
    self.params = self._process_params_data()
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/jmccrae/.cache/pypoetry/virtualenvs/wikilinks-autotrain-NvKt9JsM-py3.12/lib/python3.12/site-packages/autotrain/project.py", line 559, in _process_params_data
    return token_clf_munge_data(self.params, self.local)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/jmccrae/.cache/pypoetry/virtualenvs/wikilinks-autotrain-NvKt9JsM-py3.12/lib/python3.12/site-packages/autotrain/project.py", line 265, in token_clf_munge_data
    params.text_column = "autotrain_text"
    ^^^^^^^^^^^^^^^^^^
  File "/home/jmccrae/.cache/pypoetry/virtualenvs/wikilinks-autotrain-NvKt9JsM-py3.12/lib/python3.12/site-packages/pydantic/main.py", line 884, in __setattr__
    raise ValueError(f'"{self.__class__.__name__}" object has no field "{name}"')

Additional Information

Data is formatted as in https://huggingface.co/docs/autotrain/en/tasks/token_classification

I also tried commenting out the offending lines and then run into this error

  File "/home/jmccrae/.cache/pypoetry/virtualenvs/wikilinks-autotrain-NvKt9JsM-py3.12/lib/python3.12/site-packages/autotrain/trainers/common.py", line 212, in wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/home/jmccrae/.cache/pypoetry/virtualenvs/wikilinks-autotrain-NvKt9JsM-py3.12/lib/python3.12/site-packages/autotrain/trainers/token_classification/__main__.py", line 89, in train
    label_list = train_data.features[config.tags_column].feature.names
                 ~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^
KeyError: 'tags'

[abhishekkrthakur]

could you print column names in your dataset and the output of print(params) ?

[jmccrae]

You can run the reproduction here: https://colab.research.google.com/drive/1shka-nlusipnN6TTAlQPhcXhrvgehNF8?usp=sharing

This is the output of print(params)

{'data_path': 'data', 'model': 'FacebookAI/roberta-base', 'lr': 5e-05, 'epochs': 3, 
'max_seq_length': 128, 'batch_size': 8, 'warmup_ratio': 0.1, 
'gradient_accumulation': 1, 'optimizer': 'adamw_torch', 'scheduler': 'linear', 
'weight_decay': 0.0, 'max_grad_norm': 1.0, 'seed': 42, 'train_split': 'train',
 'valid_split': None, 'tokens_column': 'tokens', 'tags_column': 'tags',
 'logging_steps': -1, 'project_name': 'project-name', 
'auto_find_batch_size': False, 'mixed_precision': None, 'save_total_limit': 1,
 'token': None, 'push_to_hub': False, 'eval_strategy': 'epoch', 'username': None,
 'log': 'none', 'early_stopping_patience': 5, 'early_stopping_threshold': 0.01}

Also, I noted that the CSV on this page is broken as there is a space after the comma that breaks CSV parsing

[abhishekkrthakur]

fixed.

pip install -U autotrain-advanced

code:

import os

from autotrain.params import TokenClassificationParams
from autotrain.project import AutoTrainProject


if not os.path.exists("data"):
  os.makedirs("data")

with open("data/train.csv", "w") as f:
  print("tokens,tags", file=f)
  print("\"['I', 'love', 'Paris']\",\"['O', 'O', 'B-LOC']\"", file=f)
  print("\"['I', 'live', 'in', 'New', 'York']\",\"['O', 'O', 'O', 'B-LOC', 'I-LOC']\"", file=f)

with open("data/valid.csv", "w") as f:
  print("tokens,tags", file=f)
  print("\"['I', 'love', 'Paris']\",\"['O', 'O', 'B-LOC']\"", file=f)
  print("\"['I', 'live', 'in', 'New', 'York']\",\"['O', 'O', 'O', 'B-LOC', 'I-LOC']\"", file=f)


params = TokenClassificationParams(
    model="FacebookAI/roberta-base",
    data_path="data")

backend = "local"
project = AutoTrainProject(params=params, backend=backend, process=True)
project.create()

Note: ive changed the test filename to valid, otherwise, you need to specify valid_split in params.

apologies for the inconvenience.