Fairseq Integration APIs

#vizseq.ipynb.fairseq_viz.*

VizSeq can directly import and analyze model predictions generated by fairseq-generate or fairseq-interactive in Jupyter Notebook. The APIs are almost the same as the normal Jupyter Notebook APIs.

#view_stats()

#Arguments

• log_path: str: The path to fairseq-generate or fairseq-interactive log file.

#view_scores()

#Arguments

• log_path: str: The path to fairseq-generate or fairseq-interactive log file.
• metrics: List[str]: List of scorer IDs. Use available_scorers() to check all the available ones.

#view_examples()

#Arguments

• log_path: str: The path to fairseq-generate or fairseq-interactive log file.
• metrics: Optional[List[str]] = None: List of scorer IDs. Default to None. Use available_scorers() to check all the available ones.
• query: str = '': The keyword(s) for example filtering. Default to ''.
• page_sz: int = 10: Page size. Default to 10.
• page_no: int = 1: Page number. Default to 1.
• sorting: VizSeqSortingType = VizSeqSortingType.original
• need_g_translate: bool = False: To show Google Translate results or not. Default to False.
• disable_alignment: bool = False: Not to show source-reference and reference-hypothesis alignments for rendering speedup. Default to False.

#view_n_grams()

#Arguments

• log_path: str: The path to fairseq-generate or fairseq-interactive log file.
• k: int = 64: Number of n-grams to be shown. Default to 64.