-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathemb_prompt_qa.py
217 lines (207 loc) · 9.67 KB
/
emb_prompt_qa.py
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
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
from tqdm import tqdm
from openprompt.data_utils.text_classification_dataset import AgnewsProcessor, DBpediaProcessor, ImdbProcessor, AmazonProcessor, MicrosoftProcessor
from openprompt.data_utils.glue_dataset import SST2Processor, MNLIProcessor, QNLIProcessor, COLAProcessor, MRPCProcessor, QQPProcessor, RTEProcessor, STSBProcessor
from openprompt.data_utils.huggingface_dataset import YahooAnswersTopicsProcessor, QAProcessor
import torch
import argparse
from openprompt import PromptDataLoader
from openprompt.prompts import ManualVerbalizer, EmbVerbalizerQA
from openprompt.prompts import ManualTemplate
from openprompt import PromptForClassification
from openprompt.utils.reproduciblity import set_seed
from openprompt.plms import load_plm
from sklearn.metrics import classification_report, matthews_corrcoef
from scipy.stats import pearsonr
parser = argparse.ArgumentParser("")
parser.add_argument("--shot", type=int, default=0)
parser.add_argument("--seed", type=int, default=144)
parser.add_argument("--plm_eval_mode", default=True, action="store_true")
parser.add_argument("--model", type=str, default='roberta')
parser.add_argument("--model_name_or_path", default='roberta-large')
parser.add_argument("--result_file", type=str, default="qa_result.txt")
parser.add_argument("--openprompt_path", type=str, default="./")
parser.add_argument("--verbalizer", type=str, default='ept')
parser.add_argument("--calibration", default=False, action="store_true")
parser.add_argument("--nocut", default=False, action="store_true")
parser.add_argument("--filter", default="tfidf_filter", type=str)
parser.add_argument("--template_id", default=0, type=int)
parser.add_argument("--max_token_split", default=-1, type=int)
parser.add_argument("--dataset", default="qa", type=str)
parser.add_argument("--select", default=12, type=int)
parser.add_argument("--truncate", default=-1., type=float)
parser.add_argument("--sumprob", default=False, action="store_true")
parser.add_argument("--verbose", default=True, type=bool)
parser.add_argument("--write_filter_record", default=True, action="store_true")
args = parser.parse_args()
set_seed(args.seed)
use_cuda = True
plm, tokenizer, model_config, WrapperClass = load_plm(args.model, args.model_name_or_path)
dataset = {}
if args.dataset == "agnews":
dataset['train'] = AgnewsProcessor().get_train_examples(f"{args.openprompt_path}/datasets/TextClassification/agnews/")
dataset['test'] = AgnewsProcessor().get_test_examples(f"{args.openprompt_path}/datasets/TextClassification/agnews/")
class_labels = AgnewsProcessor().get_labels()
scriptsbase = "TextClassification/agnews"
scriptformat = "txt"
cutoff = 0.5 if (not args.nocut) else 0.0
max_seq_l = 128
batch_s = 800
num_labels = [i for i in range(4)]
elif args.dataset == "dbpedia":
dataset['train'] = DBpediaProcessor().get_train_examples(f"{args.openprompt_path}/datasets/TextClassification/dbpedia/")
dataset['test'] = DBpediaProcessor().get_test_examples(f"{args.openprompt_path}/datasets/TextClassification/dbpedia/")
class_labels = DBpediaProcessor().get_labels()
scriptsbase = "TextClassification/dbpedia"
scriptformat = "txt"
cutoff = 0.5 if (not args.nocut) else 0.0
max_seq_l = 128
batch_s = 800
num_labels = [i for i in range(14)]
elif args.dataset == "imdb":
# dataset['train'] = ImdbProcessor().get_train_examples(f"{args.openprompt_path}/datasets/TextClassification/imdb/")
dataset['test'] = ImdbProcessor().get_test_examples(f"{args.openprompt_path}/datasets/TextClassification/imdb/")
class_labels = ImdbProcessor().get_labels()
scriptsbase = "TextClassification/imdb"
scriptformat = "txt"
cutoff = 0
max_seq_l = 512
batch_s = 60
num_labels = [i for i in range(2)]
elif args.dataset == "amazon":
# dataset['train'] = AmazonProcessor().get_train_examples(f"{args.openprompt_path}/datasets/TextClassification/amazon/")
dataset['test'] = AmazonProcessor().get_test_examples(f"{args.openprompt_path}/datasets/TextClassification/amazon/")
class_labels = AmazonProcessor().get_labels()
scriptsbase = "TextClassification/amazon"
scriptformat = "txt"
cutoff = 0
max_seq_l = 512
batch_s = 200
num_labels = [i for i in range(2)]
elif args.dataset == "sst2":
# dataset['train'] = SST2Processor().get_examples('train')
dataset['test'] = SST2Processor().get_examples('validation')
class_labels = SST2Processor().get_labels()
scriptsbase = "GLUE/sst2"
scriptformat = "txt"
cutoff = 0.5 if (not args.nocut) else 0.0
max_seq_l = 128
batch_s = 300
num_labels = [i for i in range(2)]
elif args.dataset == "mnli-m":
# dataset['train'] = MNLIProcessor('mnli-m').get_examples('train')
dataset['test'] = MNLIProcessor('mnli_matched').get_examples('validation')
class_labels = MNLIProcessor('mnli_matched').get_labels()
scriptsbase = "GLUE/mnli-m"
scriptformat = "txt"
cutoff = 0.5 if (not args.nocut) else 0.0
max_seq_l = 256
batch_s = 500
num_labels = [i for i in range(3)]
elif args.dataset == "mnli-mm":
# dataset['train'] = MNLIProcessor('mnli-m').get_examples('train')
dataset['test'] = MNLIProcessor('mnli_mismatched').get_examples('validation')
class_labels = MNLIProcessor('mnli_mismatched').get_labels()
scriptsbase = "GLUE/mnli-mm"
scriptformat = "txt"
cutoff = 0.5 if (not args.nocut) else 0.0
max_seq_l = 256
batch_s = 500
num_labels = [i for i in range(3)]
elif args.dataset == "qnli":
# dataset['train'] = MNLIProcessor('mnli-m').get_examples('train')
dataset['test'] = QNLIProcessor().get_examples('validation')
class_labels = QNLIProcessor().get_labels()
scriptsbase = "GLUE/qnli"
scriptformat = "txt"
cutoff = 0.5 if (not args.nocut) else 0.0
max_seq_l = 300
batch_s = 400
num_labels = [i for i in range(2)]
elif args.dataset == "cola":
# dataset['train'] = MNLIProcessor('mnli-m').get_examples('train')
dataset['test'] = COLAProcessor().get_examples('validation')
class_labels = QNLIProcessor().get_labels()
scriptsbase = "GLUE/cola"
scriptformat = "txt"
cutoff = 0.5 if (not args.nocut) else 0.0
max_seq_l = 64
batch_s = 300
num_labels = [i for i in range(2)]
elif args.dataset == "mrpc":
# dataset['train'] = MNLIProcessor('mnli-m').get_examples('train')
dataset['test'] = MRPCProcessor().get_examples('validation')
class_labels = MRPCProcessor().get_labels()
scriptsbase = "GLUE/mrpc"
scriptformat = "txt"
cutoff = 0.5 if (not args.nocut) else 0.0
max_seq_l = 128
batch_s = 300
num_labels = [i for i in range(2)]
elif args.dataset == "qqp":
# dataset['train'] = MNLIProcessor('mnli-m').get_examples('train')
dataset['test'] = QQPProcessor().get_examples('validation')
class_labels = QQPProcessor().get_labels()
scriptsbase = "GLUE/qqp"
scriptformat = "txt"
cutoff = 0.5 if (not args.nocut) else 0.0
max_seq_l = 256
batch_s = 300
num_labels = [i for i in range(2)]
elif args.dataset == "rte":
# dataset['train'] = MNLIProcessor('mnli-m').get_examples('train')
dataset['test'] = RTEProcessor().get_examples('validation')
class_labels = RTEProcessor().get_labels()
scriptsbase = "GLUE/rte"
scriptformat = "txt"
cutoff = 0.5 if (not args.nocut) else 0.0
max_seq_l = 512
batch_s = 100
num_labels = [i for i in range(2)]
elif args.dataset == "stsb":
# dataset['train'] = MNLIProcessor('mnli-m').get_examples('train')
dataset['test'] = STSBProcessor().get_examples('validation')
class_labels = STSBProcessor().get_labels()
scriptsbase = "GLUE/stsb"
scriptformat = "txt"
cutoff = 0.5 if (not args.nocut) else 0.0
max_seq_l = 128
batch_s = 300
num_labels = [i for i in range(2)]
elif args.dataset == "qa":
dataset['test'] = QAProcessor().get_examples('valid')
class_labels = QAProcessor().get_labels()
scriptsbase = "QA/commonsense_qa"
scriptformat = "txt"
cutoff = 0
max_seq_l = 128
batch_s = 1
num_labels = [i for i in range(5)]
else:
raise NotImplementedError
mytemplate = ManualTemplate(tokenizer=tokenizer).from_file(f"{args.openprompt_path}/scripts/{scriptsbase}/manual_template.txt", choice=args.template_id)
myverbalizer = EmbVerbalizerQA(tokenizer, model=plm, classes=class_labels, candidate_frac=cutoff, max_token_split=args.max_token_split, sumprob=args.sumprob, verbose=args.verbose).from_file(
select_num=args.select, truncate=args.truncate, dataset_name=args.dataset, path=f'{args.dataset}_{args.model}_cos.pt', tmodel=args.model)
# zero-shot test
test_dataloader = PromptDataLoader(dataset=dataset["test"], template=mytemplate, tokenizer=tokenizer,
tokenizer_wrapper_class=WrapperClass, max_seq_length=max_seq_l, decoder_max_length=3,
batch_size=batch_s, shuffle=False, teacher_forcing=False, predict_eos_token=False,
truncate_method="tail")
allpreds = []
allprobs = []
alllabels = []
pbar = tqdm(test_dataloader)
all_stat = []
for step, inputs in enumerate(pbar):
a = dataset['test'][step]
myverbalizer.change_cls(select_num=args.select, classes=a.meta['text'])
prompt_model = PromptForClassification(plm=plm, template=mytemplate, verbalizer=myverbalizer, freeze_plm=False, plm_eval_mode=args.plm_eval_mode)
if use_cuda:
inputs = inputs.cuda()
prompt_model = prompt_model.cuda()
stat = prompt_model(inputs) # batch_size * num_class, 30 * 6
labels = inputs['label']
alllabels.extend(labels.cpu().tolist())
allpreds.extend(torch.argmax(stat, dim=-1).cpu().tolist())
allprobs.append(torch.softmax(stat, dim=-1).cpu())
acc = sum([int(i == j) for i, j in zip(allpreds, alllabels)]) / len(allpreds)
print("ACC", acc)