def __init__(self, dataset, target_dois):
self.metrics = \
[dfk.EVAL_MEAN_PREC, dfk.EVAL_MEAN_REC, dfk.EVAL_MEAN_F1,
dfk.EVAL_CI_PREC, dfk.EVAL_CI_REC, dfk.EVAL_CI_F1,
dfk.EVAL_DOC_METRICS]
self.metrics_groups = \
[(dfk.EVAL_MEAN_PREC, dfk.EVAL_CI_PREC, dfk.EVAL_PREC),
(dfk.EVAL_MEAN_REC, dfk.EVAL_CI_REC, dfk.EVAL_REC),
(dfk.EVAL_MEAN_F1, dfk.EVAL_CI_F1, dfk.EVAL_F1)]
target_dois_norm = [doi_normalize(doi) for doi in target_dois]
dataset_split = {doi: [] for doi in target_dois_norm}
for d in dataset:
doi_gt = get_target_gt_doi(d)
if isinstance(doi_gt, list):
doi_gt = doi_gt[0]
doi_gt = doi_normalize(doi_gt)
if doi_gt in target_dois_norm:
dataset_split[doi_gt].append(d)
doi_test = doi_normalize(get_target_test_doi(d))
if doi_test != doi_gt and doi_test in target_dois_norm:
dataset_split[doi_test].append(d)
results_by_doc = \
[(doi,
TargetDocLinkMetricsResults(dataset_split[doi_normalize(doi)],
doi))
for doi in target_dois]
self.results = {}
for av, ci, m in self.metrics_groups:
if results_by_doc:
self.results[av] = mean([r.get(m) for _, r in results_by_doc])
self.results[ci] = \
confidence_interval([r.get(m) for _, r in results_by_doc],
.95)
else:
self.results[ci] = None
self.results[av] = None
doc_metrics = {'doc': [d for d, _ in results_by_doc]}
for metric in [dfk.EVAL_PREC, dfk.EVAL_REC, dfk.EVAL_F1]:
doc_metrics[metric] = [r.get(metric) for _, r in results_by_doc]
self.results[dfk.EVAL_DOC_METRICS] = pd.DataFrame(doc_metrics)
def doi_test_same(item, doi):
return doi_same(doi, get_target_test_doi(item))
def doi_equals(item):
return doi_same(get_target_test_doi(item), get_target_gt_doi(item))
def doi_test_null(item):
return get_target_test_doi(item) is None
def test_target_test_doi(self):
assert get_target_test_doi({}) is None
assert get_target_test_doi(ITEM) == '10.14195/2182-708'