-
Notifications
You must be signed in to change notification settings - Fork 27
/
experiments.conf
236 lines (198 loc) · 6.96 KB
/
experiments.conf
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
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
# Word embeddings.
dummy {
path = ""
size = 0
format = txt
lowercase = false
}
glove_300d {
path = embeddings/glove.840B.300d.txt
size = 300
format = txt
lowercase = false
}
glove_300d_05_filtered {
path = embeddings/glove.840B.300d.05.filtered
size = 300
format = txt
lowercase = false
}
glove_300d_v5_filtered {
path = embeddings/glove.840B.300d.v5.filtered
size = 300
format = txt
lowercase = false
}
glove_300d_2w {
path = embeddings/glove_50_300_2.txt
size = 300
format = txt
lowercase = false
}
# Main configurations.
best {
# Computation limits.
batch_size = 40
max_tokens_per_batch = 700
# Model hyperparameters.
filter_widths = [3, 4, 5]
filter_size = 50
char_embedding_size = 8
char_vocab_path = "embeddings/char_vocab.english.txt"
context_embeddings = ${glove_300d_v5_filtered}
head_embeddings = ${glove_300d_2w}
contextualizer = lstm
contextualization_size = 200
contextualization_layers = 3
ffnn_size = 150
ffnn_depth = 2
feature_size = 20
max_span_width = 30
use_features = true
model_heads = true
num_attention_heads = 1
lm_path = "elmo/english.lm_embeddings.skip.hdf5"
lm_layers = 3
lm_size = 1024
# SRL-specific.
max_arg_width = 30
argument_ratio = 0.8
predicate_ratio = 0.4
srl_labels = ["R-ARGM-COM", "C-ARGM-NEG", "C-ARGM-TMP", "R-ARGM-DIR", "ARGM-LOC", "R-ARG2", "ARGM-GOL", "ARG5", "ARGM-EXT", "R-ARGM-ADV", "C-ARGM-MNR", "ARGA", "C-ARG4", "C-ARG2", "C-ARG3", "C-ARG0", "C-ARG1", "ARGM-ADV", "ARGM-NEG", "R-ARGM-MNR", "C-ARGM-EXT", "R-ARGM-PRP", "C-ARGM-ADV", "R-ARGM-MOD", "C-ARGM-ADJ", "ARGM-LVB", "R-ARGM-PRD", "ARGM-MNR", "ARGM-ADJ", "C-ARGM-CAU", "ARGM-CAU", "C-ARGM-MOD", "R-ARGM-EXT", "C-ARGM-COM", "ARGM-COM", "R-ARGM-GOL", "R-ARGM-TMP", "R-ARG4", "ARGM-MOD", "R-ARG1", "R-ARG0", "R-ARG3", "V", "ARGM-REC", "C-ARGM-DSP", "R-ARG5", "ARGM-DIS", "ARGM-DIR", "R-ARGM-LOC", "C-ARGM-DIS", "ARG0", "ARG1", "ARG2", "ARG3", "ARG4", "ARGM-TMP", "C-ARGM-DIR", "ARGM-PRD", "R-ARGM-PNC", "ARGM-PRX", "ARGM-PRR", "R-ARGM-CAU", "C-ARGM-LOC", "ARGM-PNC", "ARGM-PRP", "C-ARGM-PRP", "ARGM-DSP"]
enforce_srl_constraint = false
filter_v_args = true
use_gold_predicates = false
# Learning hyperparameters.
max_gradient_norm = 5.0
lexical_dropout_rate = 0.5
dropout_rate = 0.2
lstm_dropout_rate = 0.4
optimizer = adam
learning_rate = 0.001
decay_rate = 0.999
decay_frequency = 100
# Dataset/Other.
train_path = "data/srl/train.english.mtl.jsonlines"
eval_path = "data/srl/dev.english.mtl.jsonlines"
lm_path = "elmo/english.lm_embeddings.skip.hdf5"
lm_layers = 3
lm_size = 1024
main_metrics = srl
srl_conll_eval_path = ""
eval_frequency = 1000
report_frequency = 250
log_root = logs
eval_sleep_secs = 1200
}
# CoNLL2012
conll2012_best = ${best} {
main_metrics = srl
ner_conll_eval_path = ""
include_c_v = false
}
conll2012_noelmo = ${conll2012_best} {
lm_path = ""
lm_path_dev = ""
}
conll2012_final = ${conll2012_best} {
context_embeddings = ${glove_300d}
head_embeddings = ${glove_300d_2w}
eval_path = "data/srl/test.english.mtl.jsonlines"
lm_path_dev = "https://tfhub.dev/google/elmo/1"
#lm_path_dev = "elmo/ontonotes5.test.english.lm_embeddings.skip.hdf5"
srl_conll_eval_path = ""
}
conll2012_noelmo_final = ${conll2012_final} {
lm_path = ""
lm_path_dev = ""
}
conll2012_goldprops = ${conll2012_best} {
# Using larger train and dev split following previous work.
train_path = "data/srl/train.english.v5.jsonlines"
eval_path = "data/srl/dev.english.v5.jsonlines"
context_embeddings = ${glove_300d_v5_filtered}
lm_path = "elmo/ontonotes5.train.english.lm_embeddings.skip.hdf5"
lm_path_dev = "elmo/ontonotes5.dev.english.lm_embeddings.skip.hdf5"
use_gold_predicates = true
}
conll2012_goldprops_final = ${conll2012_goldprops} {
context_embeddings = ${glove_300d}
head_embeddings = ${glove_300d_2w}
eval_path = "data/srl/test.english.mtl.jsonlines"
lm_path_dev = "https://tfhub.dev/google/elmo/1"
#lm_path_dev = "elmo/ontonotes5.test.english.lm_embeddings.skip.hdf5"
}
conll2012_goldprops_noelmo = ${conll2012_goldprops} {
lm_path = ""
lm_path_dev = ""
}
conll2012_goldprops_noelmo_final = ${conll2012_goldprops_final} {
lm_path = ""
lm_path_dev = ""
}
# CoNLL-05 Experiments.
conll05_best = ${conll2012_best} {
srl_labels = ["R-A4", "C-AM-DIR", "R-A0", "R-A1", "AM-MNR", "R-A3", "V", "C-AM-MNR", "R-AM-MNR", "R-AM-TMP", "AM-PRD", "R-AM-DIR", "C-AM-CAU", "R-A2", "C-AM-TMP", "AM-EXT", "R-AM-CAU", "A1", "A0", "A3", "A2", "A5", "A4", "R-AM-EXT", "C-V", "AM-DIR", "AM-DIS", "AM-TMP", "AM-REC", "AA", "C-AM-DIS", "AM-TM", "AM-PNC", "AM-LOC", "C-A4", "AM", "R-AM-LOC", "C-AM-EXT", "AM-MOD", "AM-CAU", "C-AM-LOC", "R-AM-ADV", "C-AM-PNC", "C-AM-NEG", "C-A3", "C-A2", "C-A1", "C-A0", "R-AA", "C-A5", "R-AM-PNC", "AM-ADV", "C-AM-ADV", "AM-NEG"]
train_path = "./data/srl/train.english.conll05.jsonlines"
eval_path = "./data/srl/dev.english.conll05.jsonlines"
srl_conll_eval_path = "./data/srl/conll05.devel.props.gold.txt"
# Uses a smaller GloVe file, but not really necessary.
context_embeddings = ${glove_300d_05_filtered}
lm_path = "./elmo/conll05.train.elmo_embeddings.hdf5"
lm_path_dev = "./elmo/conll05.dev.elmo_embeddings.hdf5"
}
conll05_goldprops = ${conll05_best} {
use_gold_predicates = true
}
conll05_noelmo = ${conll05_best} {
lm_path = ""
lm_path_dev = ""
}
conll05_goldprops_noelmo = ${conll05_best} {
use_gold_predicates = true
include_c_v = true # Due to historical reasons.
lm_path = ""
lm_path_dev = ""
}
conll05_final_wsj = ${conll05_best} {
context_embeddings = ${glove_300d}
head_embeddings = ${glove_300d_2w}
eval_path = "./data/srl/test_wsj.english.conll05.jsonlines"
srl_conll_eval_path = "./data/srl/conll05.test.wsj.props.gold.txt"
#lm_path_dev = "elmo/conll05.test_wsj.elmo_embeddings.hdf5"
lm_path_dev = "https://tfhub.dev/google/elmo/1"
}
conll05_final_brown = ${conll05_final_wsj} {
eval_path = "./data/srl/test_brown.conll05.jsonlines"
srl_conll_eval_path = "./data/srl/conll05.test.brown.props.gold.txt"
#lm_path_dev = "elmo/conll05.test_brown.elmo_embeddings.hdf5"
lm_path_dev = "https://tfhub.dev/google/elmo/1"
}
conll05_noelmo_final_wsj = ${conll05_final_wsj} {
lm_path = ""
}
conll05_noelmo_final_brown = ${conll05_final_brown} {
lm_path = ""
}
conll05_goldprops_final_wsj = ${conll05_final_wsj} {
eval_path = "./data/srl/test_wsj.conll05.jsonlines"
use_gold_predicates = true
}
conll05_goldprops_final_brown = ${conll05_final_brown} {
eval_path = "./data/srl/test_brown.conll05.jsonlines"
use_gold_predicates = true
}
conll05_goldprops_noelmo_final_wsj = ${conll05_final_wsj} {
lm_path = ""
lm_path_dev = ""
eval_path = "./data/srl/test_wsj.conll05.jsonlines"
use_gold_predicates = true
include_c_v = true # Due to historical reasons.
}
conll05_goldprops_noelmo_final_brown = ${conll05_final_brown} {
lm_path = ""
lm_path_dev = ""
eval_path = "./data/srl/test_brown.conll05.jsonlines"
use_gold_predicates = true
include_c_v = true # Due to historical reasons.
}