/
experiments.conf
115 lines (104 loc) · 2.65 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
# Word embeddings.
glove_300d {
path = glove.840B.300d.txt
size = 300
}
glove_300d_filtered {
path = glove.840B.300d.txt.filtered
size = 300
}
glove_300d_2w {
path = glove_50_300_2.txt
size = 300
}
# Main configuration.
best {
# Computation limits.
max_top_antecedents = 50
max_training_sentences = 50
top_span_ratio = 0.4
# Model hyperparameters.
filter_widths = [3, 4, 5]
filter_size = 50
char_embedding_size = 8
char_vocab_path = "char_vocab.english.txt"
context_embeddings = ${glove_300d_filtered}
head_embeddings = ${glove_300d_filtered}
contextualization_size = 200
contextualization_layers = 3
ffnn_size = 150
ffnn_depth = 1
feature_size = 10
max_span_width = 6
use_metadata = true
use_features = true
model_heads = false #attention
coref_depth = 1
lm_layers = 3
lm_size = 1024
coarse_to_fine = false
# Learning hyperparameters.
max_gradient_norm = 5.0
lstm_dropout_rate = 0.4
lexical_dropout_rate = 0.5
dropout_rate = 0.2
optimizer = adam
learning_rate = 0.001
decay_rate = 0.999
decay_frequency = 100
# Other.
train_path = data/train.english.jsonlines
eval_path = data/dev.english.jsonlines
conll_eval_path = data/dev.english.wo_singleton_v4_gold_conll
lm_path = elmo_cache.hdf5
genres = ["bc", "bn", "mz", "nw", "pt", "tc", "wb"]
eval_frequency = 300
report_frequency = 100
log_root = logs
#video_feature
scene_embedding_dir = "data/scene_embedding"
scene_emb_size = 2048
use_video = false
#logic_rule_feature
use_gender_logic_rule = true
logic_rule_reg_C = 400.0
logic_rule_lambda = 1.0
logic_rule_imitation_alpha = 0.9998
logic_rule_pi_zero = 0.5
}
# For evaluation. Do not use for training (i.e. only for predict.py, evaluate.py, and demo.py). Rename `best` directory to `final`.
final = ${best} {
context_embeddings = ${glove_300d}
head_embeddings = ${glove_300d}
lm_path = ""
#eval_path = data/bigbang.english.jsonlines
#conll_eval_path = data/bigbang.english.wo_singleton_v4_gold_conll
eval_path = data/friendsnew.english.jsonlines
conll_eval_path = data/friendsnew.english.wo_singleton_v4_gold_conll
#eval_path = data/test.english.jsonlines
#conll_eval_path = data/test.english.wo_singleton_v4_gold_conll
}
# Baselines.
c2f_100_ant = ${best} {
max_top_antecedents = 100
}
c2f_250_ant = ${best} {
max_top_antecedents = 250
}
c2f_1_layer = ${best} {
coref_depth = 2
}
c2f_3_layer = ${best} {
coref_depth = 3
}
distance_50_ant = ${best} {
max_top_antecedents = 50
coarse_to_fine = false
coref_depth = 1
}
distance_100_ant = ${distance_50_ant} {
max_top_antecedents = 100
}
distance_250_ant = ${distance_50_ant} {
max_top_antecedents = 250
}