dizzy_layer/runFilterProblem.py at master · Pastromhaug/dizzy_layer · GitHub

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
import numpy as np
import tensorflow as tf
import sys
from tensorflow.python.client import timeline

from data.genFilterProblemData import genData, genEpochs, genBatch, getTestData
from utils.buildRNNCells import buildRNNCells
from utils.regularizeSpread import regularizeSpread

#global config variables
num_epochs = 1
num_steps = 4        # number of truncated backprop steps ('n' in the discussion above)
batch_size = 50
summary_name = sys.argv[1]
state_size = int(sys.argv[2])
layer_type = int(sys.argv[3])
learning_rate = float(sys.argv[4])
num_data_points = 1500
num_stacked = int(sys.argv[5])
num_test_runs = batch_size
indices = [40,15,8,3]
num_classes = len(indices)+1
Lambda = 0
trace = sys.argv[6] == "True" or sys.argv[6] == "true"
num_rots = state_size-1
if (layer_type == 10 or layer_type == 12) and len(sys.argv) >= 7:
    num_rots = int(sys.argv[6])
    trace = sys.argv[7] == "True" or sys.argv[7] == "true"
if layer_type == 8:
    Lambda = float(sys.argv[6])

rnn = buildRNNCells(layer_type, state_size, num_stacked, num_rots)

# model
x = tf.placeholder(tf.int32, [batch_size, num_steps], name='input_placeholder')
y = tf.placeholder(tf.int32, [batch_size, num_steps], name='labels_placeholder')
lr = tf.placeholder(tf.float32, name='learning_rate')
init_state = rnn.zero_state(batch_size, tf.float32)

x_one_hot = tf.one_hot(x, num_classes)
rnn_inputs = tf.unpack(x_one_hot, axis=1)

rnn_outputs, final_state = tf.nn.rnn(rnn, rnn_inputs, initial_state=init_state)
sigmas = None
if layer_type == 8:
    sigma = rnn.get_sigmas()

# [tf.histogram_summary('hidden state %d' % i, output[:,0]) for i, output in enumerate(rnn_outputs)]


with tf.variable_scope('softmax'):
    gauss =  tf.random_normal(shape=[state_size, num_classes], mean=0.0, stddev = 1/np.sqrt(num_classes))
    W = tf.get_variable('W', initializer=gauss)
    b = tf.get_variable('v', [num_classes], initializer=tf.constant_initializer(0.0))
logits = [tf.matmul(rnn_output, W) + b for rnn_output in rnn_outputs]

predictions = [tf.nn.softmax(logit) for logit in logits]

y_as_list = [tf.squeeze(i, squeeze_dims=[1]) for i in tf.split(1, num_steps, y)]

losses = [tf.nn.sparse_softmax_cross_entropy_with_logits(logit,label) for \
            logit, label in zip(logits, y_as_list)]

total_loss = tf.reduce_mean(losses)
regularization_loss = 0
if layer_type == 8:
    regularization_loss = tf.reduce_mean([regularizeSpread(sigma, Lambda) for sigma in sigmas])

pred_labels = [tf.argmax(log,1) for log in predictions]
y_as_list = tf.pack(y_as_list)
pred_labels = tf.cast(tf.pack(pred_labels), tf.int32)
correct_prediction = tf.equal(pred_labels, y_as_list)
accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
train_accuracy_summary = tf.scalar_summary('acc_train', accuracy)
train_loss_summary = tf.scalar_summary('loss_train', total_loss)
test_accuracy_summary = tf.scalar_summary('acc_test', accuracy)
test_loss_summary = tf.scalar_summary('loss_test', total_loss)

if layer_type == 8:
    regularization_loss_summary = tf.scalar_summary("regularization_loss", regularization_loss)
    sigmas_summary = tf.histogram_summary("sigmas", sigmas)
    train_summaries = tf.merge_summary([train_accuracy_summary, train_loss_summary, regularization_loss_summary, sigmas_summary])
    test_summaries = tf.merge_summary([test_accuracy_summary, test_loss_summary])
else:
    train_summaries = tf.merge_summary([train_accuracy_summary, train_loss_summary])
    test_summaries = tf.merge_summary([test_accuracy_summary, test_loss_summary])
# sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))
sess = tf.Session()
train_writer = tf.train.SummaryWriter('./test_shite/' + summary_name, sess.graph)

train_step = tf.train.AdagradOptimizer(learning_rate).minimize(total_loss + regularization_loss)

if trace:
    run_options = tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE)
run_metadata = tf.RunMetadata()
def train_network(num_epochs, num_steps, state_size=4, verbose=True):
    sess.run(tf.initialize_all_variables())
    training_losses = []

    test_epoch = genBatch(genData(num_data_points, num_steps, batch_size, indices), batch_size, num_steps)

    for idx, epoch in enumerate(genEpochs(num_epochs, num_data_points, num_steps, batch_size, indices)):
        training_loss = 0
        train_acc = 0
        train_num_steps = 0
        training_state = [np.zeros((batch_size, state_size)) for i in range(num_stacked)]
        if verbose:
            print("EPOCH %d" % idx)
        for step, (X, Y) in enumerate(epoch):
            train_num_steps += 1

            if trace:
                (training_loss_, _ , train_accuracy_, train_summaries_) = \
                    sess.run([ total_loss,
                              train_step,
                              accuracy,
                              train_summaries],
                                  feed_dict={x:X, y:Y, lr:learning_rate}, run_metadata=run_metadata, options=run_options)
            else:
                (training_loss_, _ , train_accuracy_, train_summaries_) = \
                    sess.run([ total_loss,
                              train_step,
                              accuracy,
                              train_summaries],
                                  feed_dict={x:X, y:Y, lr:learning_rate}, run_metadata=run_metadata)

            train_acc += train_accuracy_
            training_loss += training_loss_

            train_writer.add_summary(train_summaries_, idx)

        test_loss = 0
        test_acc = 0
        test_num_steps = 0
        for batch_num, (X_test, Y_test) in enumerate(test_epoch):
            if trace:
                (test_loss_, test_accuracy_, test_summaries_) = sess.run([total_loss, accuracy, test_summaries],
                    feed_dict={x:X_test, y:Y_test, lr:learning_rate}, run_metadata=run_metadata, options=run_options)
            else:
                (test_loss_, test_accuracy_, test_summaries_) = sess.run([total_loss, accuracy, test_summaries],
                    feed_dict={x:X_test, y:Y_test, lr:learning_rate}, run_metadata=run_metadata)
            test_loss += test_loss_
            test_acc += test_accuracy_
            train_writer.add_summary(test_summaries_, idx)
            test_num_steps += 1

        train_acc = train_acc/train_num_steps
        training_loss = training_loss/num_steps
        test_acc = test_acc/test_num_steps
        test_loss = test_loss/test_num_steps
        if verbose:
            print("train loss: %f train acc: %f, test loss %f, test acc %f"
                    % (training_loss, train_acc, test_loss, test_acc))
        training_losses.append(training_loss)
        training_loss = 0


    print("trace",trace)
    if trace:
        tl = timeline.Timeline(run_metadata.step_stats)
        ctf = tl.generate_chrome_trace_format()
        with open('profile_dizzy3.json', 'w') as f:
            f.write(ctf)
    return training_losses

training_losses = train_network(num_epochs,num_steps, state_size)