Spaces:
Build error
Build error
File size: 7,014 Bytes
6b35cc5 572b22d 6b35cc5 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 |
import argparse
import re
import torch
from tqdm.auto import tqdm
from network import EntNet
from utils import read_conll_ner, split_conll_docs, create_context_data, extract_spans
use_cuda = torch.cuda.is_available()
device = torch.device("cuda" if use_cuda else "cpu")
def classify(model, sents, pos, batch_size):
model.eval()
result = []
for i in tqdm(range(0, len(sents), batch_size), desc='classifying... '):
tag_seqs = model(sentences=sents[i:i + batch_size],
pos=pos[i:i + batch_size])
result.extend(tag_seqs['pred_tags'])
# f1, p, r
return [[[w, t] for w, t in zip(s, r)] for s, r in zip(sents, result)]
def entities_from_token_classes(tokens):
ENTITY_BEGIN_REGEX = r"^B" # -(\w+)"
ENTITY_MIDDLE_REGEX = r"^I" # -(\w+)"
entities = []
current_entity = None
start_index_of_current_entity = 0
end_index_of_current_entity = 0
for i, kls in enumerate(tokens):
m = re.match(ENTITY_BEGIN_REGEX, kls)
if m is not None:
if current_entity is not None:
entities.append({
"type": current_entity,
"index": [start_index_of_current_entity,
end_index_of_current_entity]
})
# start of entity
current_entity = m.string.split('-')[1] if '-' in m.string else ''
start_index_of_current_entity = i
end_index_of_current_entity = i
continue
m = re.match(ENTITY_MIDDLE_REGEX, kls)
if current_entity is not None:
if m is None:
# after the end of this entity
entities.append({
"type": current_entity,
"index": [start_index_of_current_entity,
end_index_of_current_entity]
})
current_entity = None
continue
# in the middle of this entity
end_index_of_current_entity = i
# Add any remaining entity
if current_entity is not None:
entities.append({
"type": current_entity,
"index": [start_index_of_current_entity,
end_index_of_current_entity]
})
return entities
def calc_f1(targs, preds):
stat_dict = {
'overall': {'unl_tp': 0, 'lab_tp': 0, 'targs': 0, 'preds': 0}
}
for sent_targs, sent_preds in zip(targs, preds):
stat_dict['overall']['targs'] += len(sent_targs)
stat_dict['overall']['preds'] += len(sent_preds)
for pred in sent_preds:
if pred['type'] not in stat_dict.keys():
stat_dict[pred['type']] = {'lab_tp': 0, 'targs': 0, 'preds': 0}
stat_dict[pred['type']]['preds'] += 1
for targ in sent_targs:
if targ['type'] not in stat_dict.keys():
stat_dict[targ['type']] = {'lab_tp': 0, 'targs': 0, 'preds': 0}
stat_dict[targ['type']]['targs'] += 1
# is there a span that matches exactly?
for pred in sent_preds:
if pred['index'][0] == targ['index'][0] and pred['index'][1] == targ['index'][1]:
stat_dict['overall']['unl_tp'] += 1
# if so do the tags match exactly?
if pred['type'] == targ['type']:
stat_dict['overall']['lab_tp'] += 1
stat_dict[targ['type']]['lab_tp'] += 1
for k in stat_dict.keys():
if k == 'overall':
stat_dict[k]['unl_p'] = stat_dict[k]['unl_tp'] / stat_dict[k]['preds'] if stat_dict[k]['preds'] else 0
stat_dict[k]['unl_r'] = stat_dict[k]['unl_tp'] / stat_dict[k]['targs'] if stat_dict[k]['targs'] else 0
stat_dict[k]['unl_f1'] = 2 * stat_dict[k]['unl_p'] * stat_dict[k]['unl_r'] / (
stat_dict[k]['unl_p'] + stat_dict[k]['unl_r']) if (
stat_dict[k]['unl_p'] + stat_dict[k]['unl_r']) else 0
stat_dict[k]['lab_p'] = stat_dict[k]['lab_tp'] / stat_dict[k]['preds'] if stat_dict[k]['preds'] else 0
stat_dict[k]['lab_r'] = stat_dict[k]['lab_tp'] / stat_dict[k]['targs'] if stat_dict[k]['targs'] else 0
stat_dict[k]['lab_f1'] = 2 * stat_dict[k]['lab_p'] * stat_dict[k]['lab_r'] / (
stat_dict[k]['lab_p'] + stat_dict[k]['lab_r']) if (stat_dict[k]['lab_p'] + stat_dict[k]['lab_r']) else 0
class_f1s = [v['lab_f1'] for k, v in stat_dict.items() if k != 'overall']
stat_dict['overall']['macro_lab_f1'] = sum(class_f1s) / len(class_f1s)
return stat_dict
def main(args):
global device
device = torch.device('cuda' if use_cuda else 'cpu')
test_columns = read_conll_ner(args.test_path)
test_docs = split_conll_docs(test_columns[0])
test_data = create_context_data(test_docs, args.context_size)
sents = [td[0] for td in test_data]
pos = [td[1] for td in test_data]
if len(args.model_path) > 1 or args.span_model_path is not None:
model = StagedEnsemble(model_paths=args.model_path, span_model_paths=args.span_model_path, device=device)
else:
model = EntNet.load_model(args.model_path[0], device=device)
model.to(device)
BATCH_SIZE = args.batch_size
res = classify(model, sents, pos, BATCH_SIZE)
targets = [td[2] for td in test_data]
targ_tags = [entities_from_token_classes(td[2]) for td in test_data]
pred_tags = [entities_from_token_classes([t[1] for t in r]) for r in res]
result = calc_f1(targ_tags, pred_tags)
print(f'Overall unlabelled - F1:{result["overall"]["unl_f1"]}, '
f'P:{result["overall"]["unl_p"]}, '
f'R:{result["overall"]["unl_r"]}')
print(f'Overall labelled - Micro F1:{result["overall"]["lab_f1"]}, '
f'P:{result["overall"]["lab_p"]}, '
f'R:{result["overall"]["lab_r"]}')
print(f'Overall labelled - Macro F1:{result["overall"]["macro_lab_f1"]}')
for k, v in result.items():
if k == 'overall':
continue
print(f'{k} - F1:{v["lab_f1"]}, P:{v["lab_p"]}, R:{v["lab_r"]}')
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--model_path', type=str, nargs='+', default=None, required=True, help='')
parser.add_argument('--span_model_path', type=str, nargs='*', default=None, help='')
# parser.add_argument('--network_type', type=str,
# choices=['span', 'entity', 'joint'], required=True,
# default=None, help='If entity is chosen, a path to a '
# 'span model is required also')
parser.add_argument('--test_path', type=str, default=None, help='')
parser.add_argument('--context_size', type=int, default=1, help='')
parser.add_argument('--batch_size', type=int, default=8, help='')
# parser.add_argument('--cuda_id', type=int, default=0, help='')
args = parser.parse_args()
main(args)
|