Documentação do Oracle Cloud Infrastructure

Integração com o Label Studio

O Data Labeling Service da Oracle está sendo preterido. Recomendamos que você migre seus conjuntos de dados rotulados para o Label Studio, uma ferramenta de rotulagem de código aberto e suportada pelo marketplace.

Com a descontinuação do serviço Data Labeling, siga estas etapas para converter exportações de snapshot do serviço Data Labeling para importação do Label Studio e formatos de exportação JSON brutos do serviço Label Studio. Você pode usar esses formatos para anotação adicional no Label Studio ou treinamento direto do modelo.

Fluxos de trabalho de rotulagem para NER personalizado ou TXTC personalizado

Anote um novo conjunto de dados no Label Studio.
  1. Faça upload de um conjunto de dados, seja um arquivo de texto ou conjuntos de dados pré-anotados convertidos do serviço Data Labeling.
  2. (Opcional) Se necessário, anote os intervalos da entidade.
  3. Exporte os dados anotados.
    • O conjunto de dados é exportado no formato Raw JSON do Label Studio.
    • Esses dados exportados podem ser carregados no Object Storage para processamento posterior.

Use os dados no workflow de treinamento do modelo como parte do pipeline de Idioma.

Migrando Dados Anotados pelo Serviço Data Labeling Existentes para o Label Studio

Se você já tiver anotado dados no serviço Data Labeling, será fornecido um script de migração para converter o conjunto de dados existente em um formato compatível com o Label Studio.
  1. Exporte o conjunto de dados do serviço Data Labeling no formato JSONL para um bucket de armazenamento de objetos.
  2. Execute o script de conversão.
    Um script Python processa o conjunto de dados e o converte em:
    • Formato de Importação do Label Studio para permitir o uso direto no Label Studio para obter mais anotações.
    • Label Studio Export Format, estruturado para uso direto em fluxos de trabalho de treinamento sem modificações adicionais.
  3. Faça upload do arquivo de formato de importação convertido no Label Studio para qualquer atualização de anotação.
  4. Exporte o conjunto de dados atualizado no formato RAW JSONL para o armazenamento de objetos.
Use os dados para treinamento em Linguagem.

Formatos de Dados

CNER

Os dados CNER podem ser armazenados nos seguintes formatos:
  • Formato do Data Labeling
  • Formato JSNOL portátil (texto independente)
  • Formato de Exportação JSON Bruta do Label Studio

Formato de Rotulagem de Dados (JSONL Portátil)

A principal diferença entre o Data Labeling e o formato JSONL portátil é como o texto é armazenado:

• Portable JSONL: O texto real está presente no mesmo arquivo em sourceDetails['text'].

• Formato de Rotulagem de Dados: Em vez de texto, uma sourcePath é especificada em sourceDetails['path'], apontando para um local externo onde o texto é armazenado.

Exemplos de diferentes formatos:
Formato JSONL portátil CNER
{"labelsSet": [{"name": "PER"}, {"name": "MISC"}, {"name": "LOC"}, {"name": "ORG"}], "annotationFormat": "ENTITY_EXTRACTION", "datasetFormatDetails": {"formatType": "TEXT"}}
{"sourceDetails": {"text": "His current band , Gigi , was formed in 1994 with Baron ( guitar ) , Thomas ( bass ) , Armand ( vocal ) and Ronald ( drum ) ."}, "annotations": [{"entities": [{"entityType": "TEXTSELECTION", "labels": [{"label_name": "ORG"}], "textSpan": {"offset": 19, "length": 4}}, {"entityType": "TEXTSELECTION", "labels": [{"label_name": "PER"}], "textSpan": {"offset": 50, "length": 5}}, {"entityType": "TEXTSELECTION", "labels": [{"label_name": "PER"}], "textSpan": {"offset": 69, "length": 6}}, {"entityType": "TEXTSELECTION", "labels": [{"label_name": "PER"}], "textSpan": {"offset": 87, "length": 6}}, {"entityType": "TEXTSELECTION", "labels": [{"label_name": "PER"}], "textSpan": {"offset": 108, "length": 6}}]}]}
{"sourceDetails": {"text": "Since then , Gigi has released six albums ."}, "annotations": [{"entities": [{"entityType": "TEXTSELECTION", "labels": [{"label_name": "ORG"}], "textSpan": {"offset": 13, "length": 4}}]}]}
{"sourceDetails": {"text": "He has also released his own solo albums : Nusa Damai ; Gitarku ; Samsara ; and Home , a tribute album to the December 2004 tsunami victims ."}, "annotations": [{"entities": [{"entityType": "TEXTSELECTION", "labels": [{"label_name": "MISC"}], "textSpan": {"offset": 43, "length": 10}}, {"entityType": "TEXTSELECTION", "labels": [{"label_name": "MISC"}], "textSpan": {"offset": 56, "length": 7}}, {"entityType": "TEXTSELECTION", "labels": [{"label_name": "MISC"}], "textSpan": {"offset": 66, "length": 7}}, {"entityType": "TEXTSELECTION", "labels": [{"label_name": "MISC"}], "textSpan": {"offset": 80, "length": 4}}]}]}
{"sourceDetails": {"text": "Since Dewa Budjana started his professional career as a musician , he has mainly used a Parker Fly Delux as his main guitar , occasionally using Klein and Gibson guitars SG series instead ."}, "annotations": [{"entities": [{"entityType": "TEXTSELECTION", "labels": [{"label_name": "PER"}], "textSpan": {"offset": 6, "length": 12}}, {"entityType": "TEXTSELECTION", "labels": [{"label_name": "MISC"}], "textSpan": {"offset": 88, "length": 16}}, {"entityType": "TEXTSELECTION", "labels": [{"label_name": "ORG"}], "textSpan": {"offset": 145, "length": 16}}]}]}
{"sourceDetails": {"text": "Budjana also owns a double neck Klein guitar which was used on his latest album , Home , and pictured on the CD cover ."}, "annotations": [{"entities": [{"entityType": "TEXTSELECTION", "labels": [{"label_name": "PER"}], "textSpan": {"offset": 0, "length": 7}}, {"entityType": "TEXTSELECTION", "labels": [{"label_name": "ORG"}], "textSpan": {"offset": 32, "length": 5}}, {"entityType": "TEXTSELECTION", "labels": [{"label_name": "MISC"}], "textSpan": {"offset": 82, "length": 4}}]}]}
Formato de Importação do Label Studio
[{'data': {'text': 'His current band , Gigi , was formed in 1994 with Baron ( guitar ) , Thomas ( bass ) , Armand ( vocal ) and Ronald ( drum ) .'},
  'predictions': [{'result': [{'value': {'start': 19,
       'end': 23,
       'text': 'Gigi',
       'labels': ['ORG']},
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels'},
     {'value': {'start': 50, 'end': 55, 'text': 'Baron', 'labels': ['PER']},
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels'},
     {'value': {'start': 69, 'end': 75, 'text': 'Thomas', 'labels': ['PER']},
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels'},
     {'value': {'start': 87, 'end': 93, 'text': 'Armand', 'labels': ['PER']},
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels'},
     {'value': {'start': 108, 'end': 114, 'text': 'Ronald', 'labels': ['PER']},
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels'}]}]},
 {'data': {'text': 'Since then , Gigi has released six albums .'},
  'predictions': [{'result': [{'value': {'start': 13,
       'end': 17,
       'text': 'Gigi',
       'labels': ['ORG']},
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels'}]}]},
 {'data': {'text': 'He has also released his own solo albums : Nusa Damai ; Gitarku ; Samsara ; and Home , a tribute album to the December 2004 tsunami victims .'},
  'predictions': [{'result': [{'value': {'start': 43,
       'end': 53,
       'text': 'Nusa Damai',
       'labels': ['MISC']},
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels'},
     {'value': {'start': 56, 'end': 63, 'text': 'Gitarku', 'labels': ['MISC']},
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels'},
     {'value': {'start': 66, 'end': 73, 'text': 'Samsara', 'labels': ['MISC']},
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels'},
     {'value': {'start': 80, 'end': 84, 'text': 'Home', 'labels': ['MISC']},
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels'}]}]},
 {'data': {'text': 'Since Dewa Budjana started his professional career as a musician , he has mainly used a Parker Fly Delux as his main guitar , occasionally using Klein and Gibson guitars SG series instead .'},
  'predictions': [{'result': [{'value': {'start': 6,
       'end': 18,
       'text': 'Dewa Budjana',
       'labels': ['PER']},
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels'},
     {'value': {'start': 88,
       'end': 104,
       'text': 'Parker Fly Delux',
       'labels': ['MISC']},
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels'},
     {'value': {'start': 145,
       'end': 161,
       'text': 'Klein and Gibson',
       'labels': ['ORG']},
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels'}]}]}]
Formato de Exportação do Label Studio (JSON bruto)
[{'id': 141,
  'annotations': [{'id': 42,
    'completed_by': 4,
    'result': [{'value': {'start': 19,
       'end': 23,
       'text': 'Gigi',
       'labels': ['ORG']},
      'id': '467b95b0',
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels',
      'origin': 'prediction'},
     {'value': {'start': 50, 'end': 55, 'text': 'Baron', 'labels': ['PER']},
      'id': '8f12fa43',
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels',
      'origin': 'prediction'},
     {'value': {'start': 69, 'end': 75, 'text': 'Thomas', 'labels': ['PER']},
      'id': '936f63f0',
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels',
      'origin': 'prediction'},
     {'value': {'start': 87, 'end': 93, 'text': 'Armand', 'labels': ['PER']},
      'id': 'f02d2509',
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels',
      'origin': 'prediction'},
     {'value': {'start': 108, 'end': 114, 'text': 'Ronald', 'labels': ['PER']},
      'id': 'cb479df1',
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels',
      'origin': 'prediction'}],
    'was_cancelled': False,
    'ground_truth': False,
    'created_at': '2025-01-19T16:47:10.630009Z',
    'updated_at': '2025-01-19T16:47:10.630009Z',
    'draft_created_at': None,
    'lead_time': 0,
    'prediction': {'id': 95,
     'result': [{'value': {'start': 19,
        'end': 23,
        'text': 'Gigi',
        'labels': ['ORG']},
       'id': '467b95b0',
       'from_name': 'label',
       'to_name': 'text',
       'type': 'labels',
       'origin': 'prediction'},
      {'value': {'start': 50, 'end': 55, 'text': 'Baron', 'labels': ['PER']},
       'id': '8f12fa43',
       'from_name': 'label',
       'to_name': 'text',
       'type': 'labels',
       'origin': 'prediction'},
      {'value': {'start': 69, 'end': 75, 'text': 'Thomas', 'labels': ['PER']},
       'id': '936f63f0',
       'from_name': 'label',
       'to_name': 'text',
       'type': 'labels',
       'origin': 'prediction'},
      {'value': {'start': 87, 'end': 93, 'text': 'Armand', 'labels': ['PER']},
       'id': 'f02d2509',
       'from_name': 'label',
       'to_name': 'text',
       'type': 'labels',
       'origin': 'prediction'},
      {'value': {'start': 108,
        'end': 114,
        'text': 'Ronald',
        'labels': ['PER']},
       'id': 'cb479df1',
       'from_name': 'label',
       'to_name': 'text',
       'type': 'labels',
       'origin': 'prediction'}],
     'model_version': 'undefined',
     'created_ago': '0\xa0minutes',
     'score': None,
     'cluster': None,
     'neighbors': None,
     'mislabeling': 0.0,
     'created_at': '2025-01-19T16:47:10.630009Z',
     'updated_at': '2025-01-19T16:47:10.630009Z',
     'model': None,
     'model_run': None,
     'task': 141,
     'project': 14},
    'result_count': 5,
    'unique_id': '54a60d2b-6960-48c9-baae-bf24b15ce273',
    'import_id': None,
    'last_action': None,
    'task': 141,
    'project': 14,
    'updated_by': 4,
    'parent_prediction': 95,
    'parent_annotation': None,
    'last_created_by': None}],
  'file_upload': 'example_file.json',
  'drafts': [],
  'predictions': [95],
  'data': {'text': 'His current band , Gigi , was formed in 1994 with Baron ( guitar ) , Thomas ( bass ) , Armand ( vocal ) and Ronald ( drum ) .'},
  'meta': {},
  'created_at': '2025-01-19T16:47:10.630009Z',
  'updated_at': '2025-01-19T16:47:10.630009Z',
  'inner_id': 1,
  'total_annotations': 1,
  'cancelled_annotations': 0,
  'total_predictions': 1,
  'comment_count': 0,
  'unresolved_comment_count': 0,
  'last_comment_updated_at': None,
  'project': 14,
  'updated_by': 4,
  'comment_authors': []},
 {'id': 141,
  'annotations': [{'id': 60,
    'completed_by': 4,
    'result': [{'value': {'start': 13,
       'end': 17,
       'text': 'Gigi',
       'labels': ['ORG']},
      'id': 'ce521973',
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels',
      'origin': 'prediction'}],
    'was_cancelled': False,
    'ground_truth': False,
    'created_at': '2025-01-19T16:47:10.630104Z',
    'updated_at': '2025-01-19T16:47:10.630104Z',
    'draft_created_at': None,
    'lead_time': 0,
    'prediction': {'id': 95,
     'result': [{'value': {'start': 13,
        'end': 17,
        'text': 'Gigi',
        'labels': ['ORG']},
       'id': 'ce521973',
       'from_name': 'label',
       'to_name': 'text',
       'type': 'labels',
       'origin': 'prediction'}],
     'model_version': 'undefined',
     'created_ago': '0\xa0minutes',
     'score': None,
     'cluster': None,
     'neighbors': None,
     'mislabeling': 0.0,
     'created_at': '2025-01-19T16:47:10.630104Z',
     'updated_at': '2025-01-19T16:47:10.630104Z',
     'model': None,
     'model_run': None,
     'task': 141,
     'project': 14},
    'result_count': 1,
    'unique_id': '4a4b5631-4b51-4a09-99c4-ad8927074503',
    'import_id': None,
    'last_action': None,
    'task': 141,
    'project': 14,
    'updated_by': 4,
    'parent_prediction': 95,
    'parent_annotation': None,
    'last_created_by': None}],
  'file_upload': 'example_file.json',
  'drafts': [],
  'predictions': [95],
  'data': {'text': 'Since then , Gigi has released six albums .'},
  'meta': {},
  'created_at': '2025-01-19T16:47:10.630104Z',
  'updated_at': '2025-01-19T16:47:10.630104Z',
  'inner_id': 1,
  'total_annotations': 1,
  'cancelled_annotations': 0,
  'total_predictions': 1,
  'comment_count': 0,
  'unresolved_comment_count': 0,
  'last_comment_updated_at': None,
  'project': 14,
  'updated_by': 4,
  'comment_authors': []},
 {'id': 141,
  'annotations': [{'id': 96,
    'completed_by': 4,
    'result': [{'value': {'start': 43,
       'end': 53,
       'text': 'Nusa Damai',
       'labels': ['MISC']},
      'id': 'a013849e',
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels',
      'origin': 'prediction'},
     {'value': {'start': 56, 'end': 63, 'text': 'Gitarku', 'labels': ['MISC']},
      'id': '57423aa5',
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels',
      'origin': 'prediction'},
     {'value': {'start': 66, 'end': 73, 'text': 'Samsara', 'labels': ['MISC']},
      'id': 'eeee84bb',
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels',
      'origin': 'prediction'},
     {'value': {'start': 80, 'end': 84, 'text': 'Home', 'labels': ['MISC']},
      'id': '3df96d57',
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels',
      'origin': 'prediction'}],
    'was_cancelled': False,
    'ground_truth': False,
    'created_at': '2025-01-19T16:47:10.630126Z',
    'updated_at': '2025-01-19T16:47:10.630126Z',
    'draft_created_at': None,
    'lead_time': 0,
    'prediction': {'id': 95,
     'result': [{'value': {'start': 43,
        'end': 53,
        'text': 'Nusa Damai',
        'labels': ['MISC']},
       'id': 'a013849e',
       'from_name': 'label',
       'to_name': 'text',
       'type': 'labels',
       'origin': 'prediction'},
      {'value': {'start': 56,
        'end': 63,
        'text': 'Gitarku',
        'labels': ['MISC']},
       'id': '57423aa5',
       'from_name': 'label',
       'to_name': 'text',
       'type': 'labels',
       'origin': 'prediction'},
      {'value': {'start': 66,
        'end': 73,
        'text': 'Samsara',
        'labels': ['MISC']},
       'id': 'eeee84bb',
       'from_name': 'label',
       'to_name': 'text',
       'type': 'labels',
       'origin': 'prediction'},
      {'value': {'start': 80, 'end': 84, 'text': 'Home', 'labels': ['MISC']},
       'id': '3df96d57',
       'from_name': 'label',
       'to_name': 'text',
       'type': 'labels',
       'origin': 'prediction'}],
     'model_version': 'undefined',
     'created_ago': '0\xa0minutes',
     'score': None,
     'cluster': None,
     'neighbors': None,
     'mislabeling': 0.0,
     'created_at': '2025-01-19T16:47:10.630126Z',
     'updated_at': '2025-01-19T16:47:10.630126Z',
     'model': None,
     'model_run': None,
     'task': 141,
     'project': 14},
    'result_count': 4,
    'unique_id': 'd2b8ec9a-cc62-4881-a42b-833546df7953',
    'import_id': None,
    'last_action': None,
    'task': 141,
    'project': 14,
    'updated_by': 4,
    'parent_prediction': 95,
    'parent_annotation': None,
    'last_created_by': None}],
  'file_upload': 'example_file.json',
  'drafts': [],
  'predictions': [95],
  'data': {'text': 'He has also released his own solo albums : Nusa Damai ; Gitarku ; Samsara ; and Home , a tribute album to the December 2004 tsunami victims .'},
  'meta': {},
  'created_at': '2025-01-19T16:47:10.630126Z',
  'updated_at': '2025-01-19T16:47:10.630126Z',
  'inner_id': 1,
  'total_annotations': 1,
  'cancelled_annotations': 0,
  'total_predictions': 1,
  'comment_count': 0,
  'unresolved_comment_count': 0,
  'last_comment_updated_at': None,
  'project': 14,
  'updated_by': 4,
  'comment_authors': []},
 {'id': 141,
  'annotations': [{'id': 7,
    'completed_by': 4,
    'result': [{'value': {'start': 6,
       'end': 18,
       'text': 'Dewa Budjana',
       'labels': ['PER']},
      'id': '2d675f56',
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels',
      'origin': 'prediction'},
     {'value': {'start': 88,
       'end': 104,
       'text': 'Parker Fly Delux',
       'labels': ['MISC']},
      'id': '13f6536b',
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels',
      'origin': 'prediction'},
     {'value': {'start': 145,
       'end': 161,
       'text': 'Klein and Gibson',
       'labels': ['ORG']},
      'id': '283b9e25',
      'from_name': 'label',
      'to_name': 'text',
      'type': 'labels',
      'origin': 'prediction'}],
    'was_cancelled': False,
    'ground_truth': False,
    'created_at': '2025-01-19T16:47:10.630165Z',
    'updated_at': '2025-01-19T16:47:10.630165Z',
    'draft_created_at': None,
    'lead_time': 0,
    'prediction': {'id': 95,
     'result': [{'value': {'start': 6,
        'end': 18,
        'text': 'Dewa Budjana',
        'labels': ['PER']},
       'id': '2d675f56',
       'from_name': 'label',
       'to_name': 'text',
       'type': 'labels',
       'origin': 'prediction'},
      {'value': {'start': 88,
        'end': 104,
        'text': 'Parker Fly Delux',
        'labels': ['MISC']},
       'id': '13f6536b',
       'from_name': 'label',
       'to_name': 'text',
       'type': 'labels',
       'origin': 'prediction'},
      {'value': {'start': 145,
        'end': 161,
        'text': 'Klein and Gibson',
        'labels': ['ORG']},
       'id': '283b9e25',
       'from_name': 'label',
       'to_name': 'text',
       'type': 'labels',
       'origin': 'prediction'}],
     'model_version': 'undefined',
     'created_ago': '0\xa0minutes',
     'score': None,
     'cluster': None,
     'neighbors': None,
     'mislabeling': 0.0,
     'created_at': '2025-01-19T16:47:10.630165Z',
     'updated_at': '2025-01-19T16:47:10.630165Z',
     'model': None,
     'model_run': None,
     'task': 141,
     'project': 14},
    'result_count': 3,
    'unique_id': '21115ec1-4a96-43c4-b1ec-1734cd004160',
    'import_id': None,
    'last_action': None,
    'task': 141,
    'project': 14,
    'updated_by': 4,
    'parent_prediction': 95,
    'parent_annotation': None,
    'last_created_by': None}],
  'file_upload': 'example_file.json',
  'drafts': [],
  'predictions': [95],
  'data': {'text': 'Since Dewa Budjana started his professional career as a musician , he has mainly used a Parker Fly Delux as his main guitar , occasionally using Klein and Gibson guitars SG series instead .'},
  'meta': {},
  'created_at': '2025-01-19T16:47:10.630165Z',
  'updated_at': '2025-01-19T16:47:10.630165Z',
  'inner_id': 1,
  'total_annotations': 1,
  'cancelled_annotations': 0,
  'total_predictions': 1,
  'comment_count': 0,
  'unresolved_comment_count': 0,
  'last_comment_updated_at': None,
  'project': 14,
  'updated_by': 4,
  'comment_authors': []}]

TXTC Personalizado (CTXTC) - Classificação de Texto

Os conjuntos de dados CTXTC suportam vários formatos para flexibilidade na anotação e no processamento:
  • Formato do Data Labeling
  • Formato CSV
  • Formato do Label Studio
Formato do Data Labeling
CTXTC formato portátiljsonl
Formato CSV
Formato CSV CTXTC - RÓTULO ÚNICO
Formato CSV CTXTC - MULTI LABEL
Formato de Importação do Label Studio
Formato CSV CTXTC - RÓTULO ÚNICO
Formato CSV CTXTC - MULTI LABEL
[{'data': {'text': "I need to book a hotel in the east that has 4 stars. I can help you with that. What is your price range? That doesn't matter as long as it has free wifi and parking. If you'd like something cheap, I recommend the Allenbell. For something moderately priced, I would recommend the Warkworth House. Could you book the Wartworth for one night, 1 person? What day will you be staying? Friday and Can you book it for me and get a reference number ? Booking was successful. Reference number is : BMUKPTG6.  Can I help you with anything else today? I am looking to book a train that is leaving from Cambridge to Bishops Stortford on Friday. There are a number of trains leaving throughout the day.  What time would you like to travel? I want to get there by 19:45 at the latest. Okay! The latest train you can take leaves at 17:29, and arrives by 18:07. Would you like for me to book that for you? Yes please. I also need the travel time, departure time, and price. Reference number is : UIFV8FAS. The price is 10.1 GBP and the trip will take about 38 minutes. May I be of any other assistance? Yes. Sorry, but suddenly my plans changed. Can you change the Wartworth booking to Monday for 3 people and 4 nights? I have made that change and your reference number is YF86GE4J. Thank you very much, goodbye. You're welcome. Have a nice day!"},
  'predictions': [{'model_version': '1.3',
    'result': [{'id': '0',
      'from_name': 'textClassification',
      'to_name': 'text',
      'type': 'choices',
      'value': {'choices': ['hotel', 'train']}}]}]},
 {'data': {'text': 'Howdy, I need a train heading into cambridge. I would be happy to help you find a train.  Where are you departing from? I am departing from norwich.  I need to leave after 18:45 on Wednesday. I have several options for you. Where is your destination? I will be heading to cabridge. The earliest after 18:45 is the TR8658, leaving Norwich at 19:16. Can I reserve you one or more seats on this train? yeah, i need one ticket. Booking was successful, the total fee is 17.6 GBP payable at the station. Your reference number is AXH1NM1I. Do you need assistance with anything else? I am also looking for a multi sport in the East. It looks like there is The Cherry Hinton Village Centre.  Can I get you more information about it? I would like to get the phone number, please. their phone number is 01223576412. anything else? Oh, and what is their postcode, please? Sure, the postcode is cb19ej. Can I help you find any other information about Cambridge? That is all for now thank you. enjoy your time in Cambridge!'},
  'predictions': [{'model_version': '1.3',
    'result': [{'id': '1',
      'from_name': 'textClassification',
      'to_name': 'text',
      'type': 'choices',
      'value': {'choices': ['attraction', 'train']}}]}]},
 {'data': {'text': "What can you tell me about the Riverside Brasserie? It is a restaurant that serves modern european food near the centre of town. It is moderately priced. The phone number is 01223259988. Perfect. Can you help me with a reservation for 6 people at 14:30 this coming sunday? And please make sure I have a confirmation number to use. Your reservation is set! The table will be reserved for 15 minutes. Your reference number is LZLUDTVI. Is there anything else you need? I'm also looking for a place to stay.  In the south preferably. What price range were you thinking? No particular price range, but I would like it to be a 4 star hotel. There are no hotels that fit your criteria in the South, but there are two Guesthouses. Would you like to book one of those? Sure, that will work. Can you tell me more about them? Aylesbray Lodge Guesthouse and Rosa's Bed and Breakfast, both are rated at 4 stars and both include free parking and internet. Would you like a reservation for one of them? Can I get the postcode for both of them? Aylesbray postcode is cd17sr and Rosa's postcode is cb22ha. Is there anything else I can help you with today? No thanks. That's all the help I need. Take care. Bye. thank you! Enjoy your stay!"},
  'predictions': [{'model_version': '1.3',
    'result': [{'id': '2',
      'from_name': 'textClassification',
      'to_name': 'text',
      'type': 'choices',
      'value': {'choices': ['hotel', 'restaurant']}}]}]},
 {'data': {'text': "I am looking for a specific hotel, its name is express by holiday inn cambridge. I have the Express by Holiday Inn Cambridge located on 15-17 norman way, coldhams business park. Their phone number is 01223866800. Would you like to know anything else? Yes, could you book the hotel room for me for 7 people? Yes, of course. What day would you like to stay? Monday, please. There will be 7 of us and we'd like to stay for 4 days. Here is the booking information:Booking was successful. Reference number is : 5F8G6J1G. Thank you. I would also like to book a train, please. Sure, which stations will you be traveling between? I will be going from cambridge to birmingham new street. What time would you like to leave?  The trains depart every hour. Whenever will get me there by 17:30. I do need to leave on Friday and I will need the travel time please. There are 11 results. Would you prefer the earliest departure time or latest?"},
  'predictions': [{'model_version': '1.3',
    'result': [{'id': '3',
      'from_name': 'textClassification',
      'to_name': 'text',
      'type': 'choices',
      'value': {'choices': ['hotel', 'train']}}]}]}]
Formato de Exportação do Label Studio
Formato de exportação do CTXTC LABEL Studio - RÓTULO ÚNICO
Formato de Exportação do CTXTC LABEL Studio - MULTI LABEL
[{'id': 237,
  'annotations': [{'id': 110,
    'completed_by': 4,
    'result': [{'id': '0',
      'from_name': 'textClassification',
      'to_name': 'text',
      'type': 'choices',
      'value': {'choices': ['hotel', 'train']},
      'origin': 'prediction'}],
    'was_cancelled': False,
    'ground_truth': False,
    'created_at': '2025-01-19T19:15:56.447771Z',
    'updated_at': '2025-01-19T19:15:57.447771Z',
    'draft_created_at': None,
    'lead_time': 1.101,
    'prediction': {'id': 253,
     'result': [{'id': '0',
       'from_name': 'textClassification',
       'to_name': 'text',
       'type': 'choices',
       'value': {'choices': ['hotel', 'train']}}],
     'model_version': '1.3',
     'created_ago': '1\xa0minute',
     'score': None,
     'cluster': None,
     'neighbors': None,
     'mislabeling': 0.0,
     'created_at': '2025-01-19T19:15:56.447771Z',
     'updated_at': '2025-01-19T19:15:57.447771Z',
     'model': None,
     'model_run': None,
     'task': 237,
     'project': 22},
    'result_count': 1,
    'unique_id': '6aa5eb24-4c5b-4e7a-9770-eca304ee1420',
    'import_id': None,
    'last_action': None,
    'task': 237,
    'project': 22,
    'updated_by': 4,
    'parent_prediction': 253,
    'parent_annotation': None,
    'last_created_by': None}],
  'file_upload': '0606dc5a-multiLabel_ethos_train_LabelStudio_Upload.json',
  'drafts': [],
  'predictions': [253],
  'data': {'text': "I need to book a hotel in the east that has 4 stars. I can help you with that. What is your price range? That doesn't matter as long as it has free wifi and parking. If you'd like something cheap, I recommend the Allenbell. For something moderately priced, I would recommend the Warkworth House. Could you book the Wartworth for one night, 1 person? What day will you be staying? Friday and Can you book it for me and get a reference number ? Booking was successful. Reference number is : BMUKPTG6.  Can I help you with anything else today? I am looking to book a train that is leaving from Cambridge to Bishops Stortford on Friday. There are a number of trains leaving throughout the day.  What time would you like to travel? I want to get there by 19:45 at the latest. Okay! The latest train you can take leaves at 17:29, and arrives by 18:07. Would you like for me to book that for you? Yes please. I also need the travel time, departure time, and price. Reference number is : UIFV8FAS. The price is 10.1 GBP and the trip will take about 38 minutes. May I be of any other assistance? Yes. Sorry, but suddenly my plans changed. Can you change the Wartworth booking to Monday for 3 people and 4 nights? I have made that change and your reference number is YF86GE4J. Thank you very much, goodbye. You're welcome. Have a nice day!"},
  'meta': {},
  'created_at': '2025-01-19T19:15:56.447771Z',
  'updated_at': '2025-01-19T19:15:57.447771Z',
  'inner_id': 2,
  'total_annotations': 1,
  'cancelled_annotations': 0,
  'total_predictions': 1,
  'comment_count': 0,
  'unresolved_comment_count': 0,
  'last_comment_updated_at': None,
  'project': 22,
  'updated_by': 4,
  'comment_authors': []},
 {'id': 237,
  'annotations': [{'id': 110,
    'completed_by': 4,
    'result': [{'id': '1',
      'from_name': 'textClassification',
      'to_name': 'text',
      'type': 'choices',
      'value': {'choices': ['attraction', 'train']},
      'origin': 'prediction'}],
    'was_cancelled': False,
    'ground_truth': False,
    'created_at': '2025-01-19T19:15:56.447847Z',
    'updated_at': '2025-01-19T19:15:57.447847Z',
    'draft_created_at': None,
    'lead_time': 1.101,
    'prediction': {'id': 253,
     'result': [{'id': '1',
       'from_name': 'textClassification',
       'to_name': 'text',
       'type': 'choices',
       'value': {'choices': ['attraction', 'train']}}],
     'model_version': '1.3',
     'created_ago': '1\xa0minute',
     'score': None,
     'cluster': None,
     'neighbors': None,
     'mislabeling': 0.0,
     'created_at': '2025-01-19T19:15:56.447847Z',
     'updated_at': '2025-01-19T19:15:57.447847Z',
     'model': None,
     'model_run': None,
     'task': 237,
     'project': 22},
    'result_count': 1,
    'unique_id': 'adfa5ebc-bd45-4c9a-9f7e-1e797329dd85',
    'import_id': None,
    'last_action': None,
    'task': 237,
    'project': 22,
    'updated_by': 4,
    'parent_prediction': 253,
    'parent_annotation': None,
    'last_created_by': None}],
  'file_upload': '0606dc5a-multiLabel_ethos_train_LabelStudio_Upload.json',
  'drafts': [],
  'predictions': [253],
  'data': {'text': 'Howdy, I need a train heading into cambridge. I would be happy to help you find a train.  Where are you departing from? I am departing from norwich.  I need to leave after 18:45 on Wednesday. I have several options for you. Where is your destination? I will be heading to cabridge. The earliest after 18:45 is the TR8658, leaving Norwich at 19:16. Can I reserve you one or more seats on this train? yeah, i need one ticket. Booking was successful, the total fee is 17.6 GBP payable at the station. Your reference number is AXH1NM1I. Do you need assistance with anything else? I am also looking for a multi sport in the East. It looks like there is The Cherry Hinton Village Centre.  Can I get you more information about it? I would like to get the phone number, please. their phone number is 01223576412. anything else? Oh, and what is their postcode, please? Sure, the postcode is cb19ej. Can I help you find any other information about Cambridge? That is all for now thank you. enjoy your time in Cambridge!'},
  'meta': {},
  'created_at': '2025-01-19T19:15:56.447847Z',
  'updated_at': '2025-01-19T19:15:57.447847Z',
  'inner_id': 2,
  'total_annotations': 1,
  'cancelled_annotations': 0,
  'total_predictions': 1,
  'comment_count': 0,
  'unresolved_comment_count': 0,
  'last_comment_updated_at': None,
  'project': 22,
  'updated_by': 4,
  'comment_authors': []},
 {'id': 237,
  'annotations': [{'id': 110,
    'completed_by': 4,
    'result': [{'id': '2',
      'from_name': 'textClassification',
      'to_name': 'text',
      'type': 'choices',
      'value': {'choices': ['hotel', 'restaurant']},
      'origin': 'prediction'}],
    'was_cancelled': False,
    'ground_truth': False,
    'created_at': '2025-01-19T19:15:56.447870Z',
    'updated_at': '2025-01-19T19:15:57.447870Z',
    'draft_created_at': None,
    'lead_time': 1.101,
    'prediction': {'id': 253,
     'result': [{'id': '2',
       'from_name': 'textClassification',
       'to_name': 'text',
       'type': 'choices',
       'value': {'choices': ['hotel', 'restaurant']}}],
     'model_version': '1.3',
     'created_ago': '1\xa0minute',
     'score': None,
     'cluster': None,
     'neighbors': None,
     'mislabeling': 0.0,
     'created_at': '2025-01-19T19:15:56.447870Z',
     'updated_at': '2025-01-19T19:15:57.447870Z',
     'model': None,
     'model_run': None,
     'task': 237,
     'project': 22},
    'result_count': 1,
    'unique_id': '2b0c6f75-82a7-4a0f-a5d5-a5a697ef6798',
    'import_id': None,
    'last_action': None,
    'task': 237,
    'project': 22,
    'updated_by': 4,
    'parent_prediction': 253,
    'parent_annotation': None,
    'last_created_by': None}],
  'file_upload': '0606dc5a-multiLabel_ethos_train_LabelStudio_Upload.json',
  'drafts': [],
  'predictions': [253]]

Scripts de Conversão - CNER

Script: Converter Formato de Rotulagem de Dados em Formato do Label Studio
Um script Python chamado cner_export_to_LS.py é fornecido para converter o formato JSONL do serviço Data Labeling em formatos de Importação e Exportação do Label Studio: 
import os
import argparse
import json
import uuid
from datetime import datetime
import random
import glob
 
 
def convert_to_label_studio_import(data):
    """ Convert portable JSONL format to Label Studio import format. """
    dd = []
    for i in data[1:]:
        ents = i['annotations'][0]['entities']
        ee = []
        text = i['sourceDetails']['text']
        for e in ents:
            start = e['textSpan']['offset']
            end = e['textSpan']['offset'] + e['textSpan']['length']
            ee.append({
                'value': {
                    'start': start,
                    'end': end,
                    'text': text[start:end],
                    'labels': [e['labels'][0]['label_name']]
                },
                'from_name': 'label',
                'to_name': 'text',
                'type': 'labels'
            })
        dd.append({
            'data': {'text': i['sourceDetails']['text']},
            'predictions': [{'result': ee}]
        })
    return dd
 
 
def convert_to_label_studio_raw_export(input_data):
    """ Convert portable JSONL format to Label Studio raw JSON export format. """
    # Extract text and annotations
    text = input_data['sourceDetails']['text']
    entities = input_data['annotations'][0]['entities']
     
    # Initialize variables
    task_id = 141
    project_id = 14
    now = datetime.now().isoformat() + "Z"
    unique_id = str(uuid.uuid4())
     
    # Process entities
    results = []
    for entity in entities:
        offset = entity['textSpan']['offset']
        length = entity['textSpan']['length']
        label_name = entity['labels'][0]['label_name']
        entity_text = text[offset:offset + length]
         
        result = {
            'value': {
                'start': offset,
                'end': offset + length,
                'text': entity_text,
                'labels': [label_name]
            },
            'id': str(uuid.uuid4())[:8],
            'from_name': 'label',
            'to_name': 'text',
            'type': 'labels',
            'origin': 'prediction'
        }
        results.append(result)
     
    # Construct the final output
    output = {
        'id': task_id,
        'annotations': [{
            'id': random.randint(1, 100),
            'completed_by': 4,
            'result': results,
            'was_cancelled': False,
            'ground_truth': False,
            'created_at': now,
            'updated_at': now,
            'draft_created_at': None,
            'lead_time': 0,
            'prediction': {
                'id': 95,
                'result': results,
                'model_version': 'undefined',
                'created_ago': '0\xa0minutes',
                'score': None,
                'cluster': None,
                'neighbors': None,
                'mislabeling': 0.0,
                'created_at': now,
                'updated_at': now,
                'model': None,
                'model_run': None,
                'task': task_id,
                'project': project_id
            },
            'result_count': len(results),
            'unique_id': unique_id,
            'import_id': None,
            'last_action': None,
            'task': task_id,
            'project': project_id,
            'updated_by': 4,
            'parent_prediction': 95,
            'parent_annotation': None,
            'last_created_by': None
        }],
        'file_upload': 'example_file.json',
        'drafts': [],
        'predictions': [95],
        'data': {'text': text},
        'meta': {},
        'created_at': now,
        'updated_at': now,
        'inner_id': 1,
        'total_annotations': 1,
        'cancelled_annotations': 0,
        'total_predictions': 1,
        'comment_count': 0,
        'unresolved_comment_count': 0,
        'last_comment_updated_at': None,
        'project': project_id,
        'updated_by': 4,
        'comment_authors': []
    }
     
    return output
 
 
def load_jsonl(file_path):
    """ Load a JSONL file and return a list of parsed JSON objects. """
    with open(file_path, 'r', encoding='utf-8') as f:
        return [json.loads(line) for line in f]
 
 
def save_jsonl(data, file_path):
    """ Save a list of JSON objects to a JSONL file. """
    with open(file_path, 'w', encoding='utf-8') as f:
        for entry in data:
            f.write(json.dumps(entry) + '\n')
 
 
def save_json(data, file_path):
    """ Save a JSON object to a JSON file. """
    with open(file_path, 'w', encoding='utf-8') as f:
        json.dump(data, f, indent=4)
 
 
def process_folder(input_folder, output_folder):
    """ Process all test, train, and dev files in the folder. """
    for split in ["test", "train", "dev"]:
        input_file = os.path.join(input_folder, f"{split}.jsonl")
        if not os.path.exists(input_file):
            print(f"Skipping {input_file}: File not found.")
            continue
 
        # Load input data
        data = load_jsonl(input_file)
         
        # Convert and save import format
        import_output_file = os.path.join(output_folder, f"{split}_import.jsonl")
        import_data = convert_to_label_studio_import(data)
        save_jsonl(import_data, import_output_file)
        print(f"Saved Label Studio import data to {import_output_file}")
         
        # Convert and save export format
        export_output_file = os.path.join(output_folder, f"{split}_export.json")
        export_data = [convert_to_label_studio_raw_export(item) for item in data]
        save_json(export_data, export_output_file)
        print(f"Saved Label Studio export data to {export_output_file}")
 
 
if __name__ == "__main__":
    parser = argparse.ArgumentParser(description="Convert JSONL files to Label Studio formats.")
    parser.add_argument("input_folder", type=str, help="Folder containing input JSONL files (test, train, dev).")
    parser.add_argument("output_folder", type=str, help="Folder to save converted files.")
    args = parser.parse_args()
 
    # Ensure output folder exists
    os.makedirs(args.output_folder, exist_ok=True)
 
    # Process input folder
    process_folder(args.input_folder, args.output_folder)

Instruções de uso

  1. Coloque todos os arquivos JSONL de Rotulagem de Dados (teste, treinamento, dev) em um diretório de entrada.
  2. Execute o script, cner_conversion.ipynb: 
    import glob
 
 
def list_folders_in_directory(directory_path):
    """List all folders in a given directory using glob."""
    folder_paths = glob.glob(os.path.join(directory_path, "*/"))
    # folder_names = [os.path.basename(os.path.normpath(folder)) for folder in folder_paths]
    return folder_paths
 
 
def list_files_in_directory(directory_path, extension="*"):
    """List all files in a directory with an optional extension filter."""
    file_paths = glob.glob(os.path.join(directory_path, f"*.{extension}"))
    # file_names = [os.path.basename(file) for file in file_paths]
    return file_paths
 
directory_path = "/home/niksoni/cner_labelstudio_integeration/datasets/cner"
output_folder = "/home/niksoni/cner_labelstudio_integeration/datasets/cner/label_studio_formats"
folders = list_folders_in_directory(directory_path)
for folder in folders:
    folder_name = os.path.basename(os.path.normpath(folder))
    output_folder_path = f"{output_folder}/{folder_name}"
    os.makedirs(output_folder_path, exist_ok=True)
    files = list_files_in_directory(folder)
    for input_file in files:
        if not os.path.exists(input_file):
                    print(f"Skipping {input_file}: File not found.")
                    continue 
        data = load_jsonl(input_file)
        # Convert and save import format
        file_name = os.path.basename(input_file)
        import_output_file = os.path.join(output_folder_path, f"{file_name}_LS_import.json")
        import_data = convert_to_label_studio_import(data)
        save_json(import_data, import_output_file)
        print(f"Saved Label Studio import data to {import_output_file}")
         
        export_output_file = os.path.join(output_folder_path, f"{file_name}_LS_export.json")
        export_data = [convert_to_label_studio_raw_export(item) for item in data[1:]]
        save_json(export_data, export_output_file)
        print(f"Saved Label Studio export data to {export_output_file}")

    O script gera:

    • *_import.jsonl para importação direta no Label Studio
    • *_export.json para uso direto em workflows de treinamento
    Após a conversão, a pasta de saída contém conteúdo com as seguintes estruturas:
    /input_folder/
│── test.jsonl
│── train.jsonl
│── dev.jsonl

/output_folder/
│── test_import.json
│── test_export.json
│── train_import.json
│── train_export.json
│── dev_import.json
│── dev_export.json

Scripts de Conversão - CTXTC

Script: Converter Formato CSV em Formato do Label Studio
Um script Python chamado ctxtc_export_to_LS.py é fornecido para converter o formato CSV em formatos de Importação e Exportação do Label Studio:
import os
import argparse
import json
import uuid
# from datetime import datetime
import random
import glob
from datetime import datetime, timedelta
 
import pandas as pd
 
def load_jsonl(file_path):
    """ Load a JSONL file and return a list of parsed JSON objects. """
    with open(file_path, 'r', encoding='utf-8') as f:
        return [json.loads(line) for line in f]
 
 
def save_jsonl(data, file_path):
    """ Save a list of JSON objects to a JSONL file. """
    with open(file_path, 'w', encoding='utf-8') as f:
        for entry in data:
            f.write(json.dumps(entry) + '\n')
             
             
def save_to_json(data, file_path):
    with open(file_path, 'w') as f:
        json.dump(data, f, indent=4)
    print(f"Data successfully saved to {file_path}")
 
def load_from_json(file_path):
    with open(file_path, 'r') as f:
        data = json.load(f)
    print(f"Data successfully loaded from {file_path}")
    return data
 
 
# Function to convert dataframe to Label Studio format
def convert_to_label_studio(df, label_name="textClassification"):
    data = []
     
    for idx, row in df.iterrows():
        text = row['text']
        label = str(row['labels']).split("|")
         
        data_entry = {
            "data": {
                "text": text
            },
            "predictions": [
                {
                    "model_version": "1.3",  # This can be adjusted if needed
                    "result": [
                        {
                            "id": str(idx),
                            "from_name": label_name,  # Label name changed here
                            "to_name": "text",
                            "type": "choices",
                            "value": {
#                                 "score": 1.0,  # You can adjust the score based on confidence if needed
                                "choices": label
                            }
                        }
                    ]
                }
            ]
        }
        data.append(data_entry)
     
    return json.dumps(data, indent=4)
 
 
 
 
def convert_prediction_to_label_studio_export_format(input_data):
    # Generate placeholders
    current_time = datetime.utcnow()
    created_at = current_time.isoformat() + "Z"
    updated_at = (current_time + timedelta(seconds=1)).isoformat() + "Z"
    prediction_id = 253
    task_id = 237
    project_id = 22
    annotation_id = 110
    completed_by = 4
 
    # Extract values from input
    text = input_data['data']['text']
    predictions = input_data['predictions'][0]
    model_version = predictions['model_version']
    result = predictions['result']
 
    # Build the output structure
    converted_data = {
        'id': task_id,
        'annotations': [{
            'id': annotation_id,
            'completed_by': completed_by,
            'result': [
                {
                    **item,
                    'origin': 'prediction'
                }
                for item in result
            ],
            'was_cancelled': False,
            'ground_truth': False,
            'created_at': created_at,
            'updated_at': updated_at,
            'draft_created_at': None,
            'lead_time': 1.101,
            'prediction': {
                'id': prediction_id,
                'result': result,
                'model_version': model_version,
                'created_ago': '1\xa0minute',
                'score': None,
                'cluster': None,
                'neighbors': None,
                'mislabeling': 0.0,
                'created_at': created_at,
                'updated_at': updated_at,
                'model': None,
                'model_run': None,
                'task': task_id,
                'project': project_id
            },
            'result_count': 1,
            'unique_id': str(uuid.uuid4()),
            'import_id': None,
            'last_action': None,
            'task': task_id,
            'project': project_id,
            'updated_by': completed_by,
            'parent_prediction': prediction_id,
            'parent_annotation': None,
            'last_created_by': None
        }],
        'file_upload': '0606dc5a-multiLabel_ethos_train_LabelStudio_Upload.json',
        'drafts': [],
        'predictions': [prediction_id],
        'data': {
            'text': text
        },
        'meta': {},
        'created_at': created_at,
        'updated_at': updated_at,
        'inner_id': 2,
        'total_annotations': 1,
        'cancelled_annotations': 0,
        'total_predictions': 1,
        'comment_count': 0,
        'unresolved_comment_count': 0,
        'last_comment_updated_at': None,
        'project': project_id,
        'updated_by': completed_by,
        'comment_authors': []
    }
    return converted_data

Instruções de uso

  1. Coloque todos os arquivos JSONL de Rotulagem de Dados (teste, treinamento, dev) em um diretório de entrada.
  2. Execute o script, ctxtc_conversion.ipynb: 
    import json
from ctxtc_conversion_to_LS import *
import glob
 
def list_folders_in_directory(directory_path):
    """List all folders in a given directory using glob."""
    folder_paths = glob.glob(os.path.join(directory_path, "*/"))
    # folder_names = [os.path.basename(os.path.normpath(folder)) for folder in folder_paths]
    return folder_paths
 
 
def list_files_in_directory(directory_path, extension="*"):
    """List all files in a directory with an optional extension filter."""
    file_paths = glob.glob(os.path.join(directory_path, f"*.{extension}"))
    # file_names = [os.path.basename(file) for file in file_paths]
    return file_paths
 
# cls_types = ["singleLabel", "multilabel"]
cls_types = ["multilabel"]
directory_path = "/home/niksoni/cner_labelstudio_integeration/datasets/ctxtc"
output_folder = "/home/niksoni/cner_labelstudio_integeration/datasets/ctxtc/label_studio_formats"
for cls_type in cls_types:
    cls_directory_path = f"{directory_path}/{cls_type}"
    folders = list_folders_in_directory(cls_directory_path)
    for folder in folders:
        folder_name = os.path.basename(os.path.normpath(folder))
        output_folder_path = f"{output_folder}/{cls_type}/{folder_name}"
        os.makedirs(output_folder_path, exist_ok=True)
        files = list_files_in_directory(folder)
        # print(files)
        for input_file in files:
            if not os.path.exists(input_file):
                        print(f"Skipping {input_file}: File not found.")
                        continue 
            df = pd.read_csv(input_file)
            json_data = convert_to_label_studio(df)
            json_data = json.loads(json_data)
            file_name = os.path.basename(input_file)
            # save_to_json(json_data,file_name)
            import_output_file = os.path.join(output_folder_path, f"{file_name}_LS_import.json")
            # import_data = convert_to_label_studio(data)
            save_to_json(json_data, import_output_file)
            print(f"Saved Label Studio import data to {import_output_file}")
 
            export_output_file = os.path.join(output_folder_path, f"{file_name}_LS_export.json")
            export_data = [convert_prediction_to_label_studio_export_format(item) for item in json_data]
            save_to_json(export_data, export_output_file)
            print(f"Saved Label Studio export data to {export_output_file}")

    O script gera:

    • *_import.jsonl para importação direta no Label Studio
    • *_export.json para uso direto em workflows de treinamento
    Após a conversão, a pasta de saída contém conteúdo com as seguintes estruturas:
    /input_folder/
│── test.jsonl
│── train.jsonl
│── dev.jsonl

/output_folder/
│── test_import.json
│── test_export.json
│── train_import.json
│── train_export.json
│── dev_import.json
│── dev_export.json

Integração com Idioma

Siga este fluxo de treinamento com dados convertidos:
  1. Faça upload de um conjunto de dados, anotado manualmente usando o label studio ou convertido do Data Labeling.
  2. Selecione o conjunto de dados armazenado no Object Storage para treinamento de modelo.
    O idioma processa o conjunto de dados e continua o pipeline de treinamento.

    O modelo é validado, treinado e implantado da mesma maneira que antes.

Conclusão

  • O script de migração garante a transição perfeita do serviço Data Labeling para o Label Studio.
  • Você pode continuar anotando dados existentes no Label Studio.
  • O conjunto de dados exportado é totalmente compatível com pipelines de treinamento de Idioma.
  • O formato JSONL existente ainda é suportado para compatibilidade com versões anteriores.
Para obter mais detalhes, consulte a documentação do Label Studio: Consulte também: