def main(args: argparse.Namespace) -> None:
# Word-level measures.
correct = 0
incorrect = 0
# Label-level measures.
total_edits = 0
total_length = 0
# Aux measures
ignorance_edits = 0.
dominance_edits = 0.
no_feature_edits = 0.
ped_edits = 0.
# Since the edit distance algorithm is quadratic, let's do this with
# multiprocessing.
with multiprocessing.Pool(args.cores) as pool:
gen = pool.starmap(evallib.score, evallib.tsv_reader(args.tsv_path))
for (edits, length, (ig, dom, nof, ped)) in gen:
if edits == 0:
correct += 1
else:
incorrect += 1
total_edits += edits
total_length += length
ignorance_edits += ig
dominance_edits += dom
no_feature_edits += nof
ped_edits += ped
print(f"WER:\t{100 * incorrect / (correct + incorrect):.2f}")
print(f"LER:\t{100 * total_edits / total_length:.2f}")
print(f"PED_IG:\t{100 * ignorance_edits / total_length:.2f}")
print(f"PED_DOM:\t{100 * dominance_edits / total_length:.2f}")
print(f"PED_NOF:\t{100 * no_feature_edits / total_length:.2f}")
print(f"PED:\t{100 * ped_edits / total_length:.2f}")
def main(args: argparse.Namespace) -> None:
wers = []
lers = []
for tsv_path in args.tsv_paths:
# Word-level measures.
correct = 0
incorrect = 0
# Label-level measures.
total_edits = 0
total_length = 0
# Since the edit distance algorithm is quadratic, let's do this with
# multiprocessing.
with multiprocessing.Pool(args.cores) as pool:
gen = pool.starmap(evallib.score, evallib.tsv_reader(tsv_path))
for (edits, length) in gen:
if edits == 0:
correct += 1
else:
incorrect += 1
total_edits += edits
total_length += length
wer = 100 * incorrect / (correct + incorrect)
ler = 100 * total_edits / total_length
wers.append(wer)
lers.append(ler)
print(f"{tsv_path}:\tWER:\t{wer:.2f}\tLER:\t{ler:.2f}")
wer = statistics.mean(wers)
ler = statistics.mean(lers)
print(f"Macro-average:\tWER:\t{wer:.2f}\tLER:\t{ler:.2f}")
def main(args: argparse.Namespace) -> None:
correct = 0
incorrect = 0
for (gold, hypo) in evallib.tsv_reader(args.tsv_path):
if gold == hypo:
correct += 1
else:
incorrect += 1
wer = evallib.wer(correct, incorrect)
print(f"WER:\t{wer:5.2f}")
def main(args: argparse.Namespace) -> None:
wers = []
for tsv_path in args.tsv_paths:
correct = 0
incorrect = 0
for (gold, hypo) in evallib.tsv_reader(tsv_path):
if gold == hypo:
correct += 1
else:
incorrect += 1
wer = evallib.wer(correct, incorrect)
wers.append(wer)
print(f"{tsv_path} WER:\t{wer:5.2f}")
wer = statistics.mean(wers)
print(f"Macro-average WER:\t{wer:5.2f}")
def main(args: argparse.Namespace) -> None:
# Word-level measures.
correct = 0
incorrect = 0
# Label-level measures.
total_edits = 0
total_length = 0
# Since the edit distance algorithm is quadratic, let's do this with
# multiprocessing.
with multiprocessing.Pool(args.cores) as pool:
gen = pool.starmap(evallib.score, evallib.tsv_reader(args.tsv_path))
for (edits, length) in gen:
if edits == 0:
correct += 1
else:
incorrect += 1
total_edits += edits
total_length += length
print(f"WER:\t{100 * incorrect / (correct + incorrect):.2f}")
print(f"LER:\t{100 * total_edits / total_length:.2f}")