nonLinearRegression.py

import tensorflow as tf
import numpy as np

def sigmoid(x):
	return 1 / (1 + np.exp(-x))

x_learn = np.array([[1.000,2.000],
					[10.000,2.0000],
					[9.000,3.0000],
					[5.000,8.000],
					[2.00,8.0000],
					[1.00,8.0000],
					[3.00,8.0000],
					[11.00,3.0000],
					[12.00,1.0000],
					[5.00,30.0000],
					[-10.00,8.0000],
					[8.00,8.0000],
					[2.00,11.0000],
				    [10.000,2.0000],
					[9.000,3.0000],
					[5.000,8.000],
					[2.00,8.0000],
					[1.00,100.0000],
					[1.00,8.0000],
					[11.00,3.0000],
					[12.00,1.0000],
					[6.00,30.0000],
					[3.00,7.5000],
					[8.00,9.0000],
					[10.00,5.0000]
				   ],dtype=np.float32)

x_learn2 = 10 * np.random.random((2000,2)) - 1
y_learn2 = np.power(x_learn2.T[0],2) + x_learn2.T[1]
y_learn2 = y_learn2.reshape(2000,1)
y_learn2 = y_learn2

y_learn = np.array([[3.000],
					[102.000],
					[84.000],
					[33.000],
					[12.00],
					[9.00],
					[17.00],
					[124.00],
					[145.00],
					[55.00],
					[108.00],
					[72.00],
					[15.00],
					[102.000],
					[84.000],
					[33.000],
					[12.00],
					[101.00],
					[9.00],
					[124.00],
					[145.00],
					[66.00],
					[16.50],
					[73.00],
					[105.00]
				   ],dtype=np.float32)


x_test  = np.array([[1,3],[8,10],[9,3],[5,8],[0,8]],dtype=np.float32)
y_test =    np.array([[4],[74],[84],[33],[8]],dtype=np.float32)

slearning_rate = .79966432

x = tf.placeholder(tf.float32,[None,2])
y = tf.placeholder(tf.float32,[None,1])

weights = {
	'h1': tf.Variable(10 * tf.random_normal([2,2],dtype= tf.float32)),
	'h2': tf.Variable(2 * tf.random_normal([2,4], dtype = tf.float32)),
	'h3': tf.Variable(tf.random_normal([4,4], dtype = tf.float32)),
	'h4': tf.Variable(tf.random_normal([4,1], dtype = tf.float32))
}

biases = {
	'b1': tf.Variable(10 * tf.random_normal([1,2],dtype= tf.float32)),
	'b2': tf.Variable(2*tf.random_normal([1,4],dtype= tf.float32)),
	'b3': tf.Variable(tf.random_normal([1,4],dtype= tf.float32)),
	'b4': tf.Variable(tf.random_normal([1],dtype= tf.float32))
}

def passForward(_X,_weights,_biases):
	layer_1 = tf.nn.tanh(tf.add(tf.matmul(_X,_weights['h1']),_biases['b1']))
	layer_2 = tf.nn.tanh(tf.add(tf.matmul(layer_1,_weights['h2']),_biases['b2']))
	layer_3 = tf.nn.sigmoid(tf.add(tf.matmul(layer_2,_weights['h3']),_biases['b3']))
	layer_4 = tf.matmul(layer_3,_weights['h4']) + _biases['b4']
	return layer_4

global_step = tf.Variable(0, trainable=False)

lossFunc = tf.nn.l2_loss(passForward(x,weights,biases) - y)

learning_rate = tf.train.exponential_decay(slearning_rate, global_step,10000, 0.92, staircase=True)

train_step = tf.train.AdagradOptimizer(learning_rate).minimize(lossFunc,global_step)

init = tf.initialize_all_variables()

sess = tf.InteractiveSession()
sess.run(init)
for j in xrange(200000):
	 sess.run(train_step, feed_dict={x: x_learn2,y: y_learn2})
		
	 if j%2000 == 0:
		print lossFunc.eval({x: x_learn2,y: y_learn2})/2000
		print learning_rate.eval()
		print passForward(x,weights,biases).eval({x: x_test})
		#print y_learn2
		

print passForward(x_test,weights,biases).eval()