def view_scores(references: PathOrPathsOrDictOfStrList,
hypothesis: Optional[PathOrPathsOrDictOfStrList],
metrics: List[str],
tags: Optional[PathOrPathsOrDictOfStrList] = None):
_ref = VizSeqDataSources(references)
_hypo = VizSeqDataSources(hypothesis)
_tags, tag_set = None, []
if tags is not None:
_tags = VizSeqDataSources(tags, text_merged=True)
tag_set = sorted(_tags.unique())
_tags = _tags.text
models = _hypo.names
all_metrics = get_scorer_ids()
_metrics = []
for s in metrics:
if s in all_metrics:
_metrics.append(s)
else:
logger.warn(f'"{s}" is not a valid metric.')
scores = {
s: {
m: get_scorer(s)(corpus_level=True,
sent_level=False).score(_hypo.data[i].text,
_ref.text,
tags=_tags)
for i, m in enumerate(models)
}
for s in _metrics
}
corpus_scores = {
s: {m: scores[s][m].corpus_score
for m in models}
for s in _metrics
}
group_scores = {
s: {
t: {m: scores[s][m].group_scores[t]
for m in models}
for t in tag_set
}
for s in _metrics
}
metrics_and_names = [[s, get_scorer_name(s)] for s in _metrics]
html = env.get_template('ipynb_scores.html').render(
metrics_and_names=metrics_and_names,
models=models,
tag_set=tag_set,
corpus_scores=corpus_scores,
group_scores=group_scores,
corpus_and_group_score_latex=VizSeqWebView.latex_corpus_group_scores(
corpus_scores, group_scores),
corpus_and_group_score_csv=VizSeqWebView.csv_corpus_group_scores(
corpus_scores, group_scores),
)
return HTML(html)
def get_enum_metrics_and_names(self):
return [[i, s, get_scorer_name(s)] for i, s in enumerate(self.metrics)]
def scores_to_latex(scores: Dict) -> str:
return VizSeqTableExporter.to_latex(
{get_scorer_name(s): scores for s, scores in scores.items()}
)