-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
177 lines (141 loc) · 6.27 KB
/
train.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
from cProfile import label
from datetime import datetime
import tensorflow as tf
import numpy as np
import pandas as pd
import sys
from matplotlib import pyplot as plt
#Config needed because otherwise the GPU has no memory for training
config = tf.compat.v1.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.compat.v1.Session(config=config)
#Needed to set precision to float64
tf.keras.backend.set_floatx('float64')
def usage():
print("USAGE: python3 train.py <train.csv> <test.csv> OPTIONAL: <nnet.npz>")
def main(train_csv, test_csv, nnet = None):
train = pd.read_csv(train_csv)
test = pd.read_csv(test_csv)
x_train = train[["roll", "u_x","u_y", "yaw_der", "steering", "throttle"]].to_numpy()
y_train = train[[ "u_x_der", "u_y_der", "yaw_der_der","roll_der"]].to_numpy()
x_test = test[["roll", "u_x","u_y", "yaw_der", "steering", "throttle"]].to_numpy()
y_test = test[["u_x_der", "u_y_der", "yaw_der_der","roll_der"]].to_numpy()
logdir = "logs/" + datetime.now().strftime("%Y%m%d-%H%M%S")
tensorboard_callback = tf.keras.callbacks.TensorBoard(log_dir=logdir)
# THESE ARE THE DESIGN PARAMETERS OF THE NEURAL NETWORK
# input_shape = 6 # number of independent input parameters
# input_timesteps = 1 # number of previous vehicle states to inlcude into neural network input
# output_shape = 4 # number of vehicle states (output of NN); DO NOT CHANGE
learning_rate = 0.001 # should not exceed 0.0005
initializer = "he"
l2regularization = 0.001
l1regularization = 0.0
#CREATE NEURAL NETWORK
if initializer == "he":
kernel_init = tf.keras.initializers.he_uniform(seed=True)
elif initializer == "glorot":
kernel_init = tf.keras.initializers.GlorotUniform(seed=True)
reg_dense = tf.keras.regularizers.l1_l2(l1regularization,
l2regularization)
#input_shape = input_timesteps * output_shape
model = tf.keras.models.Sequential()
model.add(tf.keras.Input(shape=(6,)))
layer_a = tf.keras.layers.Dense(32,
use_bias=True,
bias_initializer='zeros',
activation='tanh',
# kernel_initializer=kernel_init,
kernel_regularizer=reg_dense)
layer_b = tf.keras.layers.Dense(32,
bias_initializer='zeros',
use_bias=True,
activation='tanh',
# kernel_initializer=kernel_init,
kernel_regularizer=reg_dense)
layer_c = tf.keras.layers.Dense(4)
model.add(layer_a)
model.add(layer_b)
model.add(layer_c)
#After adding the layers to the model the weight matrix has values now
if nnet is not None:
old_nnet = np.load("autorally.npz")
layer_a.set_weights(list([np.transpose(old_nnet['dynamics_W1']),np.transpose(old_nnet['dynamics_b1'])]))
layer_b.set_weights(list([np.transpose(old_nnet['dynamics_W2']),np.transpose(old_nnet['dynamics_b2'])]))
layer_c.set_weights(list([np.transpose(old_nnet['dynamics_W3']),np.transpose(old_nnet['dynamics_b3'])]))
optimizer = 'Nesterov-ADAM'
clipnorm = 1.0
loss_function = 'mean_squared_error'
if optimizer == 'ADAM':
optimizer = tf.keras.optimizers.Adam(lr=learning_rate,
clipnorm=clipnorm) # , beta_1=0.9, beta_2=0.999)
if optimizer == 'SGD':
optimizer = tf.keras.optimizers.SGD(lr=learning_rate,
nesterov=True,
clipnorm=clipnorm)
if optimizer == 'RMSPROP':
optimizer = tf.keras.optimizers.RMSprop(lr=learning_rate,
momentum=0.9,
clipnorm=clipnorm)
if optimizer == 'ADADELTA':
optimizer = tf.keras.optimizers.Adadelta(lr=learning_rate) # , rho=0.95)
if optimizer == 'Nesterov-ADAM':
optimizer = tf.keras.optimizers.Nadam(lr=learning_rate,
clipnorm=clipnorm) # , beta_1=0.91, beta_2=0.997)
model.compile(optimizer=optimizer,
loss=loss_function,
metrics=['mae', 'mse'])
model.summary()
#Training the model
epochs = 10
error_epochs = np.empty([epochs, 4], dtype='float64')
for epoch in range(epochs):
print("Epoch %d/%d" % (epoch+1, epochs))
#Run one epoch of the training
model.fit(x_train, y_train, epochs=1,
callbacks=[tensorboard_callback])
# if(epoch%10 == 0):
# model.save('./tmp/model')
#Calculate losses/progress per variable
predic = model.predict(x_train)
diff_epoch = abs(predic - y_train)
error_epochs[epoch,0] = np.average(diff_epoch[:,0])
error_epochs[epoch,1] = np.average(diff_epoch[:,1])
error_epochs[epoch,2] = np.average(diff_epoch[:,2])
error_epochs[epoch,3] = np.average(diff_epoch[:,3])
#Plot the data
x_axis = list(range(epochs))
fig = plt.figure()
ax = fig.add_subplot(1, 1, 1)
ax.set_title('Average of errors per epoch')
ax.plot(x_axis, error_epochs[:,0], color='tab:blue', label = "roll_der" )
ax.plot(x_axis, error_epochs[:,1], color='tab:orange', label = "u_x_der" )
ax.plot(x_axis, error_epochs[:,2], color='tab:red', label = "u_y_der" )
ax.plot(x_axis, error_epochs[:,3], color='tab:gray', label = "yaw_der_der" )
ax.legend()
plt.savefig('Errors.png')
#Evaluating its overall performance
model.evaluate(x_test, y_test)
#Save the weights to a npz file
weight_name = "dynamics_W"
bias_name = "dynamics_b"
it = 1
files = {}
# iterate over each set of weights and biases
for layer in model.layers:
files[bias_name + str(it)] = layer.bias
files[weight_name + str(it)] = tf.Variable(np.transpose(layer.weights[0]))
it +=1
np.savez_compressed('custom.npz', dynamics_b1=files['dynamics_b1'],dynamics_b2=files['dynamics_b2'],dynamics_b3=files['dynamics_b3'], \
dynamics_W3=files['dynamics_W3'],dynamics_W2=files['dynamics_W2'],dynamics_W1=files['dynamics_W1'])
if __name__ == '__main__':
if len(sys.argv) == 3:
train_csv = sys.argv[1]
test_csv = sys.argv[2]
main(train_csv, test_csv)
elif len(sys.argv) == 4:
train_csv = sys.argv[1]
test_csv = sys.argv[2]
nnet = sys.argv[3]
main(train_csv, test_csv, nnet)
else:
usage()