-
Notifications
You must be signed in to change notification settings - Fork 0
/
feedforward Neural Network.py
178 lines (146 loc) · 7.22 KB
/
feedforward Neural Network.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
176
177
178
'''
Kaggle Competition:
Digit Recognizer - Learn Computer Vision Fundamentals with the famous MNIST Data
Jingyi Luo
'''
import os
os.environ['TF_CPP_MIN_LOG_LEVEL']='2'
#import time
import math
import numpy as np
import pandas as pd
#import matplotlib.pyplot as plt
import tensorflow as tf
from sklearn.model_selection import train_test_split
os.chdir("/Users/ljyi/Desktop/SYS6016/homework/homework02")
tf.reset_default_graph()
n_inputs = 28*28 # MNIST 784
n_hidden1 = 300 # 300
n_hidden2 = 300 # 100
n_hidden3 = 200
n_hidden4 = 100
n_hidden5 = 100
n_outputs = 10
learning_rate = 0.010 # 0.01
n_epochs = 40 # 40
batch_size = 70 #50
X = tf.placeholder(tf.float32, shape=(None, n_inputs), name="X")
y = tf.placeholder(tf.int32, shape=(None), name="y")
training = tf.placeholder_with_default(False, shape=[], name='training')
#tf.set_random_seed(888)
# ---------------------- Computation Graphs Definition ------------------------
scale = 0.01
keep_prob = 0.8
#kernel_regularizer = tf.contrib.layers.l2_regularizer(scale)
#kernel_regularizer = tf.contrib.layers.l1_regularizer(scale)
with tf.name_scope("dnn"): # a context manager. "dnn" the name argument
hidden1 = tf.layers.dense(X, n_hidden1, name="hidden1", activation=tf.nn.relu) # , kernel_regularizer = kernel_regularizer
# hidden1_drop = tf.layers.dropout(hidden1, keep_prob, training=training)
hidden2 = tf.layers.dense(hidden1, n_hidden2, name="hidden2", activation=tf.nn.relu)
# hidden2_drop = tf.layers.dropout(hidden1_drop, dropout_rate)
hidden3 = tf.layers.dense(hidden2, n_hidden3, name="hidden3", activation=tf.nn.relu)
# hidden3_drop = tf.layers.dropout(hidden3, dropout_rate)
hidden4 = tf.layers.dense(hidden3, n_hidden4, name="hidden4", activation=tf.nn.relu)
# hidden4_drop = tf.layers.dropout(hidden4, dropout_rate)
# hidden5 = tf.layers.dense(hidden4, n_hidden5, name="hidden5", activation=tf.nn.relu)
# hidden5_drop = tf.layers.dropout(hidden5, dropout_rate)
logits = tf.layers.dense(hidden4, n_outputs, name="logits") # , kernel_regularizer = kernel_regularizer
with tf.name_scope("loss"):
xentropy = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=y, logits=logits) # y here has one value, it's true label
loss = tf.reduce_mean(xentropy, name="loss")
loss_summary = tf.summary.scalar('log_loss', loss) # output a single scalar value.
# loss context for l1,l2 regularization
#with tf.name_scope("loss"):
# xentropy = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=y, logits=logits) # y here has one value, it's true label
# base_loss = tf.reduce_mean(xentropy, name="base_loss")
# reg_loss = tf.get_collection(tf.GraphKeys.REGULARIZATION_LOSSES)
# loss = tf.add_n([base_loss]+reg_loss, name='loss')
# loss_summary = tf.summary.scalar('log_loss', loss)
with tf.name_scope("train"):
optimizer = tf.train.GradientDescentOptimizer(learning_rate)
training_op = optimizer.minimize(loss)
with tf.name_scope("eval"):
correct = tf.nn.in_top_k(logits, y, 1) #logits: predictions, y:targets, k=1. Batch sized.
accuracy = tf.reduce_mean(tf.cast(correct, tf.float32))
accuracy_summary = tf.summary.scalar('accuracy', accuracy)
# initialize variables and save all variables from training
init = tf.global_variables_initializer()
saver = tf.train.Saver() # save and restore variables.
# ---------------------- Read in data, batch function -------------------------
# read in data
train_r = pd.read_csv('data/train.csv') # 42000*785 # train_r: train_raw
# train_r.info()
# train_r.isnull().sum().sum()
y_train_r = np.array(train_r['label'])
x_train_r = np.array(train_r.drop(columns=['label'])) # 4200*784
test = pd.read_csv('data/test.csv') # 2800*784
# split and scale data
x_train, x_val, y_train, y_val = train_test_split(x_train_r, y_train_r, test_size = 0.1, random_state=8)
x_train = x_train/255 # 37800*784
x_val = x_val/255 # 4200*784
test = test/255 # 28000*784
# function to get bactches
def random_batches(x, y, batch_size, seed=8):
m = x.shape[0]
batches = []
np.random.seed(seed)
permutation = list(np.random.permutation(m)) # shuffling all the rows
shuffled_x = x[permutation, :]
shuffled_y = y[permutation]
num_batches = math.floor(m/batch_size)
for k in range(0, num_batches):
batch_x = shuffled_x[k*batch_size:k*batch_size+batch_size, :]
batch_y = shuffled_y[k*batch_size:k*batch_size+batch_size]
mini_batch = (batch_x, batch_y)
batches.append(mini_batch)
# the left records to form a incomplete batch by their own
# if m% batch_size != 0:
# batch_x = shuffled_x[num_batches*batch_size:m, :]
# batch_y = shuffled_y[num_batches*batch_size:m, :]
# mini_batch = (batch_x, batch_y)
# batches.append(mini_batch)
return batches
# ----------combine summary nodes ------------
# combine all summary nodes to a single op to generates all the summary data
merged = tf.summary.merge_all()
train_writer = tf.summary.FileWriter('./graphs/train_optimize', tf.get_default_graph())
test_writer = tf.summary.FileWriter('./graphs/test_optimize', tf.get_default_graph())
# ------------------------- Model training ------------------------------------
with tf.Session() as sess:
init.run()
# i=0
for epoch in range(n_epochs):
all_batches = random_batches(x_train, y_train, batch_size)
for a_batch in all_batches:
(X_batch, y_batch) = a_batch
sess.run(training_op, feed_dict = {X:X_batch, y:y_batch}) # , training: True
acc_train = accuracy.eval(feed_dict = {X:X_batch, y:y_batch})
acc_val = accuracy.eval(feed_dict= {X: x_val, y: y_val}) # ,
print(epoch, "Train accuracy:", acc_train, "Val accuracy:", acc_val)
# measure validation accuracy, and write validate summaries to FileWriters
test_summary, acc = sess.run([merged, accuracy], feed_dict={X: x_val, y: y_val})
test_writer.add_summary(test_summary, epoch)
print('Accuracy at step %s: %s' % (epoch, acc))
# run training_op on training data, and add training summaries to FileWriters
train_summary, _ = sess.run([merged, training_op], feed_dict={X:X_batch, y:y_batch})
train_writer.add_summary(train_summary, epoch)
train_writer.close()
test_writer.close()
save_path = saver.save(sess, "./model_optimization/my_model_final.ckpt") #save the whole session
# write computation graph to tensorboard
writer = tf.summary.FileWriter('./graphs/MNIST', tf.get_default_graph())
writer.add_graph(sess.graph)
# ------------------------ Test set prediction --------------------------------
with tf.Session() as sess:
saver.restore(sess, "./model_optimization/my_model_final.ckpt")
Z = logits.eval(feed_dict = {X: test})
y_pred = np.argmax(Z, axis = 1)
print("Predicted classes:", y_pred)
# write to a dataframe to upload to kaggle
output = pd.DataFrame(y_pred, columns=['Label'])
output['ImageId'] = range(0, len(test))
# swap two columns
columnsTitles=["ImageId","Label"]
output=output.reindex(columns=columnsTitles)
# write to csv
output.to_csv("prediction_kaggle.csv", index=False)