-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathinit_parameter.py
169 lines (114 loc) · 8.65 KB
/
init_parameter.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
from argparse import ArgumentParser
def init_model():
parser = ArgumentParser()
parser.add_argument("--data_dir", default='data', type=str,
help="The input data dir. Should contain the .csv files (or other data files) for the task.")
parser.add_argument("--api_key", type=str, default=None,
help="The key for Openai API.")
parser.add_argument("--save_results_path", type=str, default='outputs',
help="The path to save results.")
parser.add_argument("--bert_model", default="bert-base-uncased", type=str,
help="The path or name for the pre-trained bert model.")
parser.add_argument("--tokenizer", default="bert-base-uncased", type=str,
help="The path or name for the tokenizer")
parser.add_argument("--feat_dim", default=768, type=int,
help="Bert feature dimension.")
parser.add_argument("--warmup_proportion", default=0.1, type=float,
help="Warmup proportion for optimizer.")
parser.add_argument("--gpu_id", type=str, default='0', help="Select the GPU id.")
parser.add_argument("--save_premodel", action="store_true", help="Save Pretrained model.")
parser.add_argument("--save_model", action="store_true", help="Save Pretrained model.")
parser.add_argument("--disable_pretrain", action="store_true", help="Disable Pretrain.")
parser.add_argument("--save_model_path", default='./model', type=str,
help="Path to save model checkpoints. Set to None if not save.")
parser.add_argument("--dataset", default=None, type=str, required=True,
help="Name of dataset.")
parser.add_argument('--seed', type=int, default=0,
help="Random seed.")
parser.add_argument("--method", type=str, default='CLNN',
help="The name of method.")
parser.add_argument("--rtr_prob", default=0.25, type=float,
help="Probability for random token replacement")
parser.add_argument("--pretrain_batch_size", default=64, type=int,
help="Batch size for pre-training")
parser.add_argument("--train_batch_size", default=96, type=int,
help="Batch size for training.") # original: 128
parser.add_argument("--eval_batch_size", default=64, type=int,
help="Batch size for evaluation.")
parser.add_argument("--wait_patient", default=20, type=int,
help="Patient steps for Early Stop in pretraining.")
parser.add_argument("--num_pretrain_epochs", default=100, type=int,
help="The pre-training epochs.")
parser.add_argument("--num_train_epochs", default=50, type=int,
help="The training epochs.")
parser.add_argument("--lr_pre", default=5e-5, type=float,
help="The learning rate for pre-training.")
parser.add_argument("--lr", default=1e-5, type=float,
help="The learning rate for training.")
parser.add_argument("--temp", default=0.07, type=float,
help="Temperature for contrastive loss")
parser.add_argument("--view_strategy", default="rtr", type=str,
help="Choose from rtr|shuffle|none")
parser.add_argument("--report_pretrain", action="store_true",
help="Enable reporting performance right after pretrain")
parser.add_argument("--topk", default=50, type=int,
help="Select topk nearest neighbors")
parser.add_argument("--grad_clip", default=1, type=float,
help="Value for gradient clipping.")
# DistillLoss
parser.add_argument('--memax_weight', type=float, default=1)
parser.add_argument('--warmup_teacher_temp', default=0.07, type=float, help='Initial value for the teacher temperature.')
parser.add_argument('--teacher_temp', default=0.04, type=float, help='Final value (after linear warmup)of the teacher temperature.')
parser.add_argument('--warmup_teacher_temp_epochs', default=5, type=int, help='Number of warmup epochs for the teacher temperature.')
parser.add_argument("--ce_weight", default=1, type=float, help="The weight of cross entropy loss.")
parser.add_argument("--cl_weight", default=1, type=float, help="The weight of contrative loss.")
parser.add_argument('--sup_weight', type=float, default=1, help='The weight for supervised loss.')
parser.add_argument("--weight_ce_unsup", default=0, type=float, help="The weight of supervised pseudo oracle cross entropy loss.")
parser.add_argument('--print_freq', default=10, type=int)
# Method
parser.add_argument("--architecture", default="SimGCD", type=str, help="Choose method and architecture from SimGCD|Loop")
parser.add_argument("--running_method", default=" ", type=str, help="Specify the running method.")
parser.add_argument("--experiment_name", default=" ", type=str, help="Specify massive experiment name.")
# Setting
parser.add_argument("--known_cls_ratio", default=0.1, type=float, required=True, help="The ratio of known classes.")
parser.add_argument("--label_setting", default="ratio", type=str, help="Choose from shot|ratio")
parser.add_argument('--labeled_shot', default=10, type=int)
parser.add_argument("--labeled_ratio", default=0.1, type=float, help="The ratio of labeled samples.")
parser.add_argument('--query_samples', default=500, type=int)
parser.add_argument("--options", default=5, type=int, help="# Options in querying llm.")
parser.add_argument("--update_per_epoch", default=1, type=int,
help="Update neighbor contrastive learning data after certain amount of epochs")
# Cluster Interpritability
parser.add_argument("--interpret_sampling_strategy", default="nearest_center", type=str, help="Choose from random|nearest_center|nearest_sub_kmeans_centriods")
parser.add_argument("--interpret_num_representatives", default=10, type=int, help="# representatives from each cluster to query llm.")
parser.add_argument("--llm", default="gpt-4o-mini", type=str, help="Choose from gpt-3.5-turbo|gpt-4o-mini|gpt-4o|gpt-4-turbo|gpt-4")
# Self-Adaptive Querying
parser.add_argument("--sampling_strategy", default="highest", type=str, help="Choose from random|highest|equal_highest|adaptive_difficulty")
parser.add_argument("--allocation_degree", default=1, type=float, help="degree of convexity or concavity for the allocation function")
# Cluster Instance Alignment Learning
parser.add_argument("--weight_cluster_instance_cl", default=0, type=float, help="The weight of alignment loss.")
parser.add_argument("--options_cluster_instance_ratio", default=0.5, type=float, help="# Options in querying llm.")
# LLM Feedback Caching & Replaying
parser.add_argument("--feedback_cache", action="store_true", help="Save all feedback.")
# LLM Feedback Enhancement and Filtering for Instance-Level Feedback
parser.add_argument("--flag_demo", action="store_true", help="Enable demo in prompt.") # default: False
parser.add_argument("--known_demo_num_per_class", default=5, type=int, help="# demo per known class.")
parser.add_argument("--flag_filtering", action="store_true", help="Enable feedback filtering.") # default: False
parser.add_argument("--filter_threshold", default=0.8, type=float, help="Threshold for filtering feedback.")
# LLM Feedback Enhancement and Filtering for Cluster-Level Feedback
parser.add_argument("--flag_demo_c", action="store_true", help="Enable demo in prompt.") # default: False
parser.add_argument("--known_demo_num_per_class_c", default=2, type=int, help="# demo per known class.")
parser.add_argument("--flag_filtering_c", action="store_true", help="Enable feedback filtering.") # default: False
parser.add_argument("--filter_threshold_c", default=0.85, type=float, help="Threshold for filtering feedback.")
# args.flag_demo = True
# args.known_demo_num_per_class = 5
# args.flag_filtering = True
# args.filter_threshold = 0.8
# args.flag_demo_c = True
# args.known_demo_num_per_class_c = 2
# args.flag_filtering_c = True
# args.filter_threshold_c = 0.8
# Ablation
parser.add_argument("--prompt_ablation", default="full", type=str, help="Choose from full|wo_demo|wo_name|wo_description")
parser.add_argument("--component_ablation", default="full", type=str, help="Choose from full|wo_ce|wo_cl_1|wo_cl_all|wo_instance_feedback|wo_cluster_feedback|wo_both_feedback")
return parser