-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathrun_embedded_lstm_early_pred.py
175 lines (148 loc) · 7.07 KB
/
run_embedded_lstm_early_pred.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
import argparse
import numpy
import os
import re
import pandas
import tensorflow as tf
from kddcup_dataset import KDDCupDataset
from quick_experiment import utils
from models import kdd_embedded_lstm
def parse_arguments():
parser = argparse.ArgumentParser()
parser.add_argument('--base_logs_dirname', type=str, default=None,
help='Path to directory to store tensorboard info')
parser.add_argument('--input_directory', type=str,
help='The path to the pickled file with the processed'
'sequences.')
parser.add_argument('--test_prediction_dir', type=str,
help='The path to the directory to store the '
'predictions')
parser.add_argument('--training_epochs', type=int, default=500,
help='The number of epochs to run.')
parser.add_argument('--runs', type=int, default=1,
help='Number of times to run the experiment with'
'different samples')
parser.add_argument('--hidden_layer_size', type=int, default=50,
help='Number of cells in the recurrent layer.')
parser.add_argument('--batch_size', type=int, default=100,
help='Number if instances to process at the same time.')
parser.add_argument('--log_values', type=int, default=50,
help='How many training epochs to wait before logging'
'the accuracy in validation.')
parser.add_argument('--max_num_steps', type=int, default=100,
help='Number of time steps to unroll the network.')
parser.add_argument('--dropout_ratio', type=float, default=0.3,
help='Dropout for the input layer and the recurrent '
'layer.')
parser.add_argument('--embedding_size', type=int, default=100,
help='Number of units in the embedding layer.')
parser.add_argument('--course_number', type=str,
help='Number of the course to identify predictions.')
parser.add_argument('--model', type=str, default='elstm',
help='Name of the model to run. The variation is in the'
'difference function between co-embeddings. '
'Possible values are elstm and ebilstm.')
return parser.parse_args()
MODELS = {
'elstm': kdd_embedded_lstm.KDDCupEmbeddedLSTMModel,
'ebilstm': kdd_embedded_lstm.KDDCupEmbedBiLSTMModel,
}
def read_configuration(args):
config = {
'hidden_layer_size': args.hidden_layer_size,
'batch_size': args.batch_size,
'log_values': args.log_values,
'max_num_steps': args.max_num_steps,
'dropout_ratio': args.dropout_ratio,
'embedding_size': args.embedding_size,
'embedding_model': None,
'name': args.model
}
dataset_config = {'train': 0.85, 'test': 1, 'validation': 0.15}
return config, dataset_config
def transform_input(train_sequences, test_sequences):
"""Removes the second column from all sequences. Leaves only the module id.
"""
return (
numpy.array([numpy.array(x)[:,0] for x in train_sequences]),
numpy.array([numpy.array(x)[:,0] for x in test_sequences])
)
def get_periods(args):
is_valid = lambda filepath: (
os.path.isfile(filepath) and 'period' in filepath
and 'c{}_'.format(args.course_number) in filepath)
for filename in os.listdir(args.input_directory):
filepath = os.path.join(args.input_directory, filename)
if not is_valid(filepath):
continue
period = int(re.search('.*_period(\d+)_.*', filename).group(1))
train_df, test_df = utils.pickle_from_file(filepath)
yield period, train_df, test_df
def evaluate_period(args, experiment_config, kddcup_dataset, period):
predictions = []
performances = []
for run in range(args.runs):
print('Running iteration {} of {}'.format(run + 1, args.runs))
kddcup_dataset.set_current_sample(run)
if args.base_logs_dirname:
tf.reset_default_graph()
logs_dirname = os.path.join(
args.base_logs_dirname,
'c{}_p{}_run{}'.format(args.course_number, period, run))
utils.safe_mkdir(logs_dirname)
experiment_config['logs_dirname'] = logs_dirname
model = MODELS[args.model](kddcup_dataset, **experiment_config)
model.fit(partition_name='train',
training_epochs=args.training_epochs, close_session=False)
true, predicted = model.predict('test')
prediction_df = pandas.DataFrame(true, columns=['true'])
prediction_df.loc[:, 'predicted'] = predicted
prediction_df.loc[:, 'run'] = run
predictions.append(prediction_df)
train_performance = pandas.DataFrame(
model.training_performance, columns=['epoch', 'mse'])
train_performance.loc[:, 'run'] = run
train_performance.loc[:, 'dataset'] = 'train'
val_performance = pandas.DataFrame(
model.validation_performance, columns=['epoch', 'mse'])
val_performance.loc[:, 'dataset'] = 'validation'
val_performance.loc[:, 'run'] = run
performances.extend([train_performance, val_performance])
return pandas.concat(predictions), pandas.concat(performances)
def main():
args = parse_arguments()
experiment_config, partitions = read_configuration(args)
print('Dataset Configuration')
print(partitions)
print('Experiment Configuration')
print(experiment_config)
if args.base_logs_dirname:
utils.safe_mkdir(args.base_logs_dirname)
utils.safe_mkdir(args.test_prediction_dir)
print('Reading dataset')
predictions, performances = [], []
for period, train_sequences, test_sequences in get_periods(args):
print('Creating samples')
train_instances, test_instances = transform_input(
train_sequences.sequence.values, test_sequences.sequence.values)
kddcup_dataset = KDDCupDataset(padding_value=-1)
kddcup_dataset.create_fixed_samples(
train_instances, train_sequences.dropout.values,
test_instances, test_sequences.dropout.values,
partition_sizes=partitions, samples_num=args.runs)
period_predictions, period_performances = evaluate_period(
args, experiment_config, kddcup_dataset, period)
period_predictions.loc[:, 'period'] = period
period_performances.loc[:, 'period'] = period
performances.append(period_performances)
predictions.append(period_predictions)
prediction_filename = os.path.join(
args.test_prediction_dir,
'predictions_c{}.p'.format(args.course_number))
utils.pickle_to_file(pandas.concat(predictions), prediction_filename)
utils.pickle_to_file(pandas.concat(performances), os.path.join(
args.test_prediction_dir,
'performances_c{}.p'.format(args.course_number)))
print('All operations finished')
if __name__ == '__main__':
main()