def evaluate(schema_name, use_gold_edu=False):
pipeline = schemas.create_pipeline(schema_name)
cdtb = dataset.load_cdtb_by_config()
parses = []
if use_gold_edu:
print("parsing with gold EDU")
else:
print("parsing with auto discourse segmenter")
for gold in tqdm(cdtb.test, desc="parsing for evaluation"):
if use_gold_edu:
discourse = pipeline.annotate(gold.strip())
else:
discourse = pipeline(gold.label, gold.text, gold.span[0],
gold.span[1], gold.info)
parses.append(discourse)
metrics = CDTBMetrics(golds=cdtb.test, parses=parses)
if not use_gold_edu:
print(metrics.segmenter_report())
print(metrics.parser_report())
print(metrics.nuclear_report())
print(metrics.relation_report())
def main():
cdtb = dataset.load_cdtb_by_config()
model = train(cdtb.train + cdtb.validate)
model_dir = config.get("segmenter.svm", "model_dir")
save(model, model_dir)
evaluate(cdtb.test, model)
def main():
cdtb = dataset.load_cdtb_by_config()
train(cdtb)