Skip to main content
Version: 0.91

snorkelflow.ingest.import_utterance_ground_truth

snorkelflow.ingest.import_utterance_ground_truth(node, filepath, label_col, conversation_id_col='context_uid', utterance_id_col='utterance_idx', metadata=None, user_format=False)

Uploads ground truth for utterance classification applications. Requires input in the form of [conversation_id, utterance_id, label] triplets.

Parameters:

  • node (int) – Node uid for the model node to which GTs are to be uploaded

  • filepath (str) – Path to the CSV file which contains GTs. Local path and MinIO path are supported.

  • label_col (str) – Name of the column which contains labels. If labels are strings, user_format must be True

  • conversation_id_col (str, default: 'context_uid') – Name of the column containing conversation_ids. Defaults to ‘context_uid’ if using Snorkelflow generated UIDs.

  • utterance_id_col (str, default: 'utterance_idx') – Name of the column containing utterance_ids. Defaults to ‘utterance_idx’.

  • metadata (Optional[Dict[str, Any]], default: None) – Metadata to register as the provenance of ground truth.

  • user_format (bool, default: False) – True if the ground truth labels are strings, False otherwise.

Return type:

None