-
Notifications
You must be signed in to change notification settings - Fork 15
/
Copy pathguild.yml
155 lines (155 loc) · 5.17 KB
/
guild.yml
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
- model: rnn_net
description: PyTorch RNN net hyperparameter tuning
operations:
train_rnn:
description: "Evaluate model trained with given parameters."
sourcecode:
root: ../../..
select:
- modeling
- categorical_variables_embeddings
main: modeling.models.rnn.rnn_tuning_guild
requires:
- file: ../../../data
- file: ../../../submissions
flags:
run_type:
description: "Choose whether you want to evaluate, or predict by training from scratch/using already trained model."
choices: ["evaluate", "train_predict", "load_predict"]
default: "evaluate"
submission_name:
description: "Filename for the submission, if preparing one"
default: "submission.csv.gz"
embedding_type:
description: "Type of embedding to use"
choices: ["starspace", "wiki", "gensim"]
default: "starspace"
embedding_size:
description: "Size of the embeddings"
default: 50
embedding_epochs:
description: "Number of epochs for embeddings training"
default: 20
embedding_lr:
description: "Learning rate for embeddings training"
default: 0.02
average_dense_sets:
description: "Whether to average dense sets or concatenate them"
choices: [0, 1]
default: 1
num_epochs:
description: "Number of epochs"
default: 1
batch_size:
description: "Size of a batch"
default: 256
learning_rate:
description: "Learning rate"
default: 1e-5
pre_rnn_layers_num:
description: "Number of fully connected layers before RNN input"
default: 2
pre_rnn_dim:
description: "Size of pre RNN FC layers"
default: 376
rnn_module:
description: "RNN module to be used"
choices: ["rnn", "lstm", "gru"]
default: "gru"
rnn_layers_num:
description: "Number of inner RNN layers"
default: 1
rnn_input_dim:
description: "Size of rnn input"
default: 188
rnn_hidden_output_dim:
description: "Size of RNN hidden/output layers"
default: 188
rnn_initialize_memory_gate_bias:
description: "Whether to use the trick to make RNN remember more at the beggining"
choices: [0, 1]
default: 1
post_rnn_layers_num:
description: "Number of fully connected layers after RNN output"
default: 2
post_rnn_dim:
description: "Size of post RNN FC layers"
default: 188
pre_output_dim:
description: "Size of the last layer before the output"
default: 94
tune_rnn_iter_1:
sourcecode: no
steps:
- run: train_rnn
embedding_type=["gensim","wiki"]
embedding_size=[1:100]
embedding_epochs=[5:50]
embedding_lr=loguniform[1e-3:1.0]
average_dense_sets=[0,1]
num_epochs=10
batch_size=512
learning_rate=loguniform[1e-6:1e-3]
pre_rnn_layers_num=[1:5]
pre_rnn_dim=[20:500]
rnn_module=["gru","lstm","rnn"]
rnn_layers_num=[1:5]
rnn_input_dim=[20:500]
rnn_hidden_output_dim=[20:500]
rnn_initialize_memory_gate_bias=[0,1]
post_rnn_layers_num=[1:5]
post_rnn_dim=[20:500]
pre_output_dim=[20:500]
run_type="evaluate"
--max-trials 150
--optimizer gp
tune_rnn_iter_2:
sourcecode: no
steps:
- run: train_rnn
embedding_type="wiki"
embedding_size=300
embedding_epochs=1
embedding_lr=1
average_dense_sets=1
num_epochs=1
batch_size=[750:1000]
learning_rate=[0.0003:0.0004]
pre_rnn_layers_num=1
pre_rnn_dim=[400:1200]
rnn_module="gru"
rnn_layers_num=1
rnn_input_dim=[250:800]
rnn_hidden_output_dim=[500:1500]
rnn_initialize_memory_gate_bias=1
post_rnn_layers_num=1
post_rnn_dim=[1:100]
pre_output_dim=[1:20]
run_type="evaluate"
--max-trials 150
--optimizer gp
tune_rnn_iter_3:
sourcecode: no
steps:
- run: train_rnn
embedding_type="wiki" \
embedding_size=300 \
embedding_epochs=1 \
embedding_lr=1 \
average_dense_sets=1 \
num_epochs=15 \
batch_size=256 \
learning_rate=[1e-6:1e-5] \
pre_rnn_layers_num=1 \
pre_rnn_dim=[400:1200] \
rnn_module="gru" \
rnn_layers_num=1 \
rnn_input_dim=[250:800] \
rnn_hidden_output_dim=[500:1500] \
rnn_initialize_memory_gate_bias=1 \
post_rnn_layers_num=1 \
post_rnn_dim=[500:1500] \
pre_output_dim=[1:20] \
run_type="evaluate" \
--max-trials 150 \
--optimizer gp