def main():
torch.multiprocessing.set_sharing_strategy("file_system")
parser = argparse.ArgumentParser("Visualize BERTScore")
parser.add_argument(
"--lang",
type=str,
default="en",
help="two-letter abbreviation of the language (e.g., en)")
parser.add_argument("-m",
"--model",
default=None,
help="BERT model name (default: bert-base-uncased)")
parser.add_argument("-l",
"--num_layers",
type=int,
default=None,
help="use first N layer in BERT (default: 8)")
parser.add_argument("-v",
"--verbose",
action="store_true",
help="increase output verbosity")
parser.add_argument("-r",
"--ref",
type=str,
required=True,
help="reference sentence")
parser.add_argument("-c",
"--cand",
type=str,
required=True,
help="candidate sentence")
parser.add_argument("-f",
"--file",
type=str,
default="visualize.png",
help="name of file to save output matrix in")
parser.add_argument(
"--rescale_with_baseline",
action="store_true",
help="Rescaling the numerical score with precomputed baselines")
parser.add_argument("--baseline_path",
default=None,
type=str,
help="path of custom baseline csv file")
args = parser.parse_args()
bert_score.plot_example(
args.cand,
args.ref,
model_type=args.model,
lang=args.lang,
num_layers=args.num_layers,
fname=args.file,
rescale_with_baseline=args.rescale_with_baseline,
baseline_path=args.baseline_path,
)
def main():
torch.multiprocessing.set_sharing_strategy('file_system')
parser = argparse.ArgumentParser('Visualize BERTScore')
parser.add_argument(
'--lang',
type=str,
default='en',
help='two-letter abbreviation of the language (e.g., en)')
parser.add_argument('-m',
'--model',
default=None,
help='BERT model name (default: bert-base-uncased)')
parser.add_argument('-l',
'--num_layers',
type=int,
default=None,
help='use first N layer in BERT (default: 8)')
parser.add_argument('-v',
'--verbose',
action='store_true',
help='increase output verbosity')
parser.add_argument('-r',
'--ref',
type=str,
required=True,
help='reference sentence')
parser.add_argument('-c',
'--cand',
type=str,
required=True,
help='candidate sentence')
parser.add_argument('-f',
'--file',
type=str,
default='visualize.png',
help='name of file to save output matrix in')
parser.add_argument(
'--rescale-with-baseline',
action='store_true',
help='Rescaling the numerical score with precomputed baselines')
args = parser.parse_args()
bert_score.plot_example(args.cand,
args.ref,
model_type=args.model,
lang=args.lang,
num_layers=args.num_layers,
fname=args.file,
rescale_with_baseline=args.rescale_with_baseline)
def main():
torch.multiprocessing.set_sharing_strategy('file_system')
parser = argparse.ArgumentParser('Calculate BERTScore')
parser.add_argument('--bert',
default='bert-base-multilingual-cased',
choices=bert_score.bert_types,
help='BERT model name (default: bert-base-uncased)')
parser.add_argument('-l',
'--num_layers',
default=8,
help='use first N layer in BERT (default: 8)')
parser.add_argument('-v',
'--verbose',
action='store_true',
help='increase output verbosity')
parser.add_argument('-r',
'--ref',
required=True,
help='reference sentence')
parser.add_argument('-c',
'--cand',
required=True,
help='candidate sentence')
parser.add_argument('-o',
'--output_file_name',
default='',
help='output file name')
args = parser.parse_args()
cand = args.cand
ref = args.ref
fname = args.output_file_name
bert_score.plot_example(cand,
ref,
verbose=args.verbose,
bert=args.bert,
num_layers=args.num_layers,
fname=fname)
def word_alignment(sentence1, sentence2, outfile):
batches = batchify(sentence1, sentence2, batch_size=500)
# output = []
for batch in batches:
b0 = [x.sent for x in batch[0]]
b1 = [x.sent for x in batch[1]]
otokens1 = [[x.tokens[y].word for y in x.tokens if y != 0]
for x in batch[0]]
otokens2 = [[x.tokens[y].word for y in x.tokens if y != 0]
for x in batch[1]]
# otokens1 = [b.split(' ') for b in b0]
# otokens2 = [b.split(' ') for b in b1]
btokens1, btokens2, sim = plot_example(b0, b1)
sim = compress_similarity_matrix(otokens1, otokens2, btokens1,
btokens2, sim)
for i in range(len(batch[0])):
try:
sim[i].shape
cont = True
except:
cont = False
if cont:
try:
reordered1, reordered2 = get_sentence_ordering_dep(
batch[0][i], batch[1][i], sim[i])
if reordered1 is None or reordered2 is None:
continue
r1 = " ".join(reordered1)
r2 = " ".join(reordered2)
t1 = " ".join(otokens1[i])
t2 = " ".join(otokens2[i])
outfile.write(t1 + '\n')
outfile.write(t2 + '\n')
outfile.write(r1 + '\n')
outfile.write(r2 + '\n')
outfile.write('\n')
except:
continue
def plot_similarity_matrix(cand, ref, lang):
plot_example(cand, ref, lang=lang)
def bert_viz_heatmap(ref, pred):
from bert_score import plot_example
ref = ' '.join(ref[0]).strip()
pred = ' '.join(pred[0]).strip()
plot_example(pred, ref, lang='en', rescale_with_baseline=True)
from bert_score import score, plot_example
import matplotlib.pyplot as plt
import os
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
with open("../data/models/hyps.txt") as f:
cands = [line.strip() for line in f]
with open("../data/models/refs.txt") as f:
refs = [line.strip() for line in f]
P, R, F1 = score(cands, refs, lang='en')
print(f"System level F1 score: {F1.mean():.3f}")
plt.hist(F1, bins=20)
plt.show()
print(cands[0])
print(f'P={P[0]:.6f} R={R[0]:.6f} F={F1[0]:.6f}')
plot_example(cands[0], refs[0], lang="en")
