/
main.py
215 lines (169 loc) · 10.8 KB
/
main.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
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
import pathlib
import matplotlib.pyplot as plt
import pandas as pd
import seaborn as sns
import numpy as np
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers
import sklearn.model_selection as sk
from sklearn.metrics import confusion_matrix
import os
import time
os.environ['PYTHONHASHSEED'] = '0'
np.random.seed(42)
tf.random.set_random_seed(88)
#try to cut down on randomness
start = time.perf_counter()
def main():
# data comes in as 65 chars which is 64 for the image pixels and 1 for the label
# turn this into a 8x8x1
testing_df = pd.read_csv('data/test/optdigits.tes', header=None)
X_testing, y_testing = testing_df.loc[:, 0:63], testing_df.loc[:, 64]
training_df = pd.read_csv('data/train/optdigits.tra', header=None)
X_training, y_training = training_df.loc[:, 0:63], training_df.loc[:, 64]
X_training, X_validation, y_training, y_val = sk.train_test_split(X_training, # validation data
y_training,
test_size=0.20,
random_state=42)#this random shuffle witll cause inconsistency in data
# shapping data so it can be used for normal NN and conv NN
X_train = X_training.to_numpy().reshape(-1, 8, 8, 1)
X_test = X_testing.to_numpy().reshape(-1, 8, 8, 1)
X_validation = X_validation.to_numpy().reshape(-1, 8, 8, 1)
# this one hot encodes the data. You shouldn't use this for the confusion matrices
y_validation = keras.utils.to_categorical(y_val, 10)
y_train = keras.utils.to_categorical(y_training, 10)
y_test = keras.utils.to_categorical(y_testing, 10)
x_evalData = X_test # X_validation for 20% of training data used as validation data, X_test for testing data
y_evalData = y_test # y_validation for 20% of training data used as validation data, y_test for testing data
labels = y_testing # used for confusion matrix. use y_val when using X_validation and y_validation otherwise use y_testing
loss = 'categorical_crossentropy'
# categorical_crossentropy experiments
trainModels(1, 25, 64, 64, 0, .001, 0.0, X_train, y_train, x_evalData, y_evalData, loss, 'relu', labels)
trainModels(1, 25, 128, 64, 0.01, .05, 0.0, X_train, y_train, x_evalData, y_evalData, loss,
'relu', labels)
trainModels(5, 50, 64, 128, 0.001, .02, 0.0, X_train, y_train, x_evalData, y_evalData, loss,
'relu', labels)
trainModels(5, 50, 128, 128, 0.1, .005, 0.0, X_train, y_train, x_evalData, y_evalData, loss,
'relu', labels)
trainModels(5, 75, 10, 64, 0.001, .005, 0.0, X_train, y_train, x_evalData, y_evalData, loss,
'relu', labels)
trainModels(20, 100, 120, 256, 0.01, .01, 0.0, X_train, y_train, x_evalData, y_evalData, loss,
'relu', labels)
trainModels(20, 100, 256, 64, 0.001, .01, 0.0, X_train, y_train, x_evalData, y_evalData, loss,
'relu', labels)
loss = 'mean_squared_error'
# mean_squared_error experiments
trainModels(1, 25, 64, 64, .006, .01, 0.0, X_train, y_train, x_evalData, y_evalData, loss, 'relu', labels)
trainModels(1, 25, 64, 64, 0, .01, 0.0, X_train, y_train, x_evalData, y_evalData, loss, 'relu', labels)
trainModels(1, 25, 128, 64, 0.01, .05, 0.0, X_train, y_train, x_evalData, y_evalData, loss,
'relu', labels)
trainModels(5, 50, 64, 128, 0.001, .02, 0.0, X_train, y_train, x_evalData, y_evalData, loss,
'relu', labels)
trainModels(5, 50, 128, 128, 0.1, .05, 0.0, X_train, y_train, x_evalData, y_evalData, loss,
'relu', labels)
trainModels(10, 100, 120, 256, 0.01, .5, 0.0, X_train, y_train, x_evalData, y_evalData, loss,
'relu', labels)
trainModels(10, 100, 256, 64, 0.01, .09, 0.0, X_train, y_train, x_evalData, y_evalData, loss,
'relu', labels)
# categorical crossentropy tanh experiments
# loss = 'categorical_crossentropy'
trainModels(1, 25, 100, 64, .02, .01, 0.0, X_train, y_train, x_evalData, y_evalData, loss, 'tanh', labels)
trainModels(2, 30, 50, 32, 0.002, .003, 0.0, X_train, y_train, x_evalData, y_evalData, loss, 'tanh', labels)
trainModels(3, 60, 90, 64, 0.03, .02, 0.0, X_train, y_train, x_evalData, y_evalData, loss, 'tanh', labels)
trainModels(5, 50, 50, 128, 0.06, .001, 0.0, X_train, y_train, x_evalData, y_evalData, loss, 'tanh', labels)
trainModels(10, 75, 30, 80, 0.05, .02, 0.0, X_train, y_train, x_evalData, y_evalData, loss, 'tanh', labels)
trainModels(20, 90, 30, 256, 0.2, .01, 0.0, X_train, y_train, x_evalData, y_evalData, loss, 'tanh', labels)
trainModels(30, 100, 30, 64, 0.08, .02, 0.0, X_train, y_train, x_evalData, y_evalData, loss, 'tanh', labels)
loss = 'categorical_crossentropy'
# categorical crossentropy relu experiments2
trainModels(1, 25, 100, 64, .02, .01, 0.0, X_train, y_train, x_evalData, y_evalData, loss, 'relu', labels)
trainModels(2, 30, 50, 32, 0.002, .003, 0.0, X_train, y_train, x_evalData, y_evalData, loss, 'relu', labels)
trainModels(3, 60, 90, 64, 0.01, .03, 0.0, X_train, y_train, x_evalData, y_evalData, loss, 'relu', labels)
trainModels(5, 50, 50, 128, 0.06, .001, 0.0, X_train, y_train, x_evalData, y_evalData, loss, 'relu', labels)
trainModels(10, 75, 30, 80, 0.05, .02, 0.0, X_train, y_train, x_evalData, y_evalData, loss, 'relu', labels)
trainModels(20, 90, 30, 256, 0.02, .001, 0.0, X_train, y_train, x_evalData, y_evalData, loss, 'relu', labels)
trainModels(30, 100, 30, 64, 0.02, .01, 0.0, X_train, y_train, x_evalData, y_evalData, loss, 'relu', labels)
loss = 'categorical_crossentropy'
# Convolutional experiments relu experiments2
trainConvModels(1, 30, 10, 128, 0.02, .01, 0.0, X_train, y_train, x_evalData, y_evalData, loss, 'relu', labels, 10,
400, 5, .3)
trainConvModels(2, 30, 15, 128, 0.02, .05, 0.0, X_train, y_train, x_evalData, y_evalData, loss, 'relu', labels, 5,
200, 3, .3)
trainConvModels(3, 90, 20, 128, 0.02, .01, 0.0, X_train, y_train, x_evalData, y_evalData, loss, 'relu', labels, 10,
400, 5, .2)
trainConvModels(4, 120, 25, 128, 0.02, .02, 0.0, X_train, y_train, x_evalData, y_evalData, loss, 'relu', labels, 10,
200, 6, .6)
trainConvModels(5, 150, 30, 128, 0.02, .01, 0.0, X_train, y_train, x_evalData, y_evalData, loss, 'relu', labels, 10,
400, 5, .5)
elapsed = time.perf_counter() - start
print('Elapsed %.3f seconds.' % elapsed)
def trainModels(numHiddenLayers, numEpochs, numHiddenUnitsPerLayer, batchSize, momentum, learningRate, decay, X_train,
y_train, x_evalData, y_evalData, lossFunction, hiddenUnitActivation, y_val):
model = tf.keras.Sequential()
# Adds a densely-connected layer with 64 units to the model:
model.add(layers.Dense(64, activation='relu', input_shape=(8, 8, 1)))
model.add(layers.Flatten())
for x in range(numHiddenLayers):
# Add another:
model.add(layers.Dense(numHiddenUnitsPerLayer, activation=hiddenUnitActivation))
# output with softmax activation function
model.add(layers.Dense(10, activation='softmax'))
sgd = tf.keras.optimizers.SGD(
lr=learningRate, momentum=momentum, decay=decay)
model.compile(optimizer=sgd,
loss=lossFunction,
metrics=['accuracy'])
callbacks = [tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=2, verbose=1)]
# This employs the early stopping technique. Since the patience is 2 the model will stop training if there is
# no improvement in 2 epochs
# ,tf.keras.callbacks.ModelCheckpoint(filepath='best_model.h5', monitor='val_loss', save_best_only=True)] #saving checkpoint
hist = model.fit(X_train, y_train, batch_size=batchSize, epochs=numEpochs, verbose=0)
scores = model.evaluate(x_evalData, y_evalData, verbose=1)
print("hidden layers, = ", numHiddenLayers, " hidden units = ", numHiddenUnitsPerLayer,
" epochs = ", numEpochs, "Batch Size = ", batchSize, " learning rate = ", learningRate, " momentum rate = ",
momentum, "Loss = "
, scores[0], "Accuracy = ", scores[1])
labels = (tf.argmax(y_val, axis=0))
prediction = model.predict(x_evalData)
printConfusionMatrix(y_val, prediction)
def trainConvModels(numHiddenLayers, numEpochs, numHiddenUnitsPerLayer, batchSize, momentum, learningRate, decay,
X_train, y_train, x_evalData, y_evalData, lossFunction, hiddenUnitActivation, y_val,
kernalSize, filters, poolSize, dropout):
model = tf.keras.Sequential()
# Adds a densely-connected layer with 64 units to the model:
model.add(tf.keras.layers.Convolution2D(filters, kernalSize, input_shape=(8, 8, 1), data_format="channels_first",
activation='relu',
padding='same'))
model.add(tf.keras.layers.MaxPooling2D(pool_size=(poolSize, poolSize),
data_format="channels_last")) # channel dimension/depth is dim 1 to match 32 x 3 x 3
model.add(layers.Dropout(dropout))
model.add(tf.keras.layers.Flatten())
for x in range(numHiddenLayers):
# Add another:
model.add(layers.Dense(numHiddenUnitsPerLayer, activation=hiddenUnitActivation))
# output with softmax activation function
model.add(layers.Dense(10, activation='softmax'))
sgd = tf.keras.optimizers.SGD(
lr=learningRate, momentum=momentum, decay=decay)
model.compile(optimizer=sgd,
loss=lossFunction,
metrics=['accuracy'])
callbacks = [tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=2, verbose=0)]
# This employs the early stopping technique. Since the patience is 2 the model will stop training if there is
# no improvement in 2 epochs
# ,tf.keras.callbacks.ModelCheckpoint(filepath='best_model.h5', monitor='val_loss', save_best_only=True)] #saving checkpoint
hist = model.fit(X_train, y_train, batch_size=batchSize, epochs=numEpochs, verbose=0)
scores = model.evaluate(x_evalData, y_evalData, verbose=1)
print("hidden layers, = ", numHiddenLayers, " hidden units = ", numHiddenUnitsPerLayer,
" epochs = ", numEpochs, "Batch Size = ", batchSize, " learning rate = ", learningRate, " momentum rate = ",
momentum, "Loss = "
, scores[0], "Accuracy = ", scores[1])
# labels = (tf.argmax(y_val, axis=0))
prediction = model.predict(x_evalData)
printConfusionMatrix(y_val, prediction)
def printConfusionMatrix(y_val, prediction):
confusion = confusion_matrix(y_val, np.argmax(prediction, axis=1))
print(confusion)
if __name__ == "__main__":
main()