import tensorflow as tf
import numpy as np
from sklearn.utils import shuffle
import matplotlib.pyplot as plt



trainsamples = 200
testsamples = 60

#Here we will represent the model, a simple imput, a hidden layer of sigmoid activation
def model(X, hidden_weights1, hidden_bias1, ow):
hidden_layer = tf.nn.sigmoid(tf.matmul(X, hidden_weights1)+ b)
return tf.matmul(hidden_layer, ow)

dsX = np.linspace(-1, 1, trainsamples + testsamples).transpose()
dsY = 0.4* pow(dsX,2) +2 * dsX + np.random.randn(*dsX.shape) * 0.22 + 0.8

plt.figure() # Create a new figure
plt.title('Original data')
plt.scatter(dsX,dsY) #Plot a scatter draw of the datapoints
plt.show()

X = tf.placeholder("float")
Y = tf.placeholder("float")

hw1 = tf.Variable(tf.random_normal([1, 10], stddev=0.01)) # Create first hidden layer
ow = tf.Variable(tf.random_normal([10, 1], stddev=0.01)) # Create output connection
b = tf.Variable(tf.random_normal([10], stddev=0.01)) # Create bias

model_y = model(X, hw1, b, ow) #

cost = tf.pow(model_y-Y, 2)/(2) # Cost function

train_op = tf.train.AdamOptimizer(0.0001).minimize(cost) # construct an optimizer

# Launch the graph in a session
with tf.Session() as sess:
tf.initialize_all_variables().run() #Initialize all variables
for i in range(1,10):
trainX, trainY =dsX[0:trainsamples], dsY[0:trainsamples]
for x1,y1 in zip (trainX, trainY):
sess.run(train_op, feed_dict={X: [[x1]], Y: y1})
testX, testY = dsX[trainsamples:trainsamples + testsamples], dsY[0:trainsamples:trainsamples+testsamples]
cost1=0.
for x1,y1 in zip (testX, testY):
cost1 += sess.run(cost, feed_dict={X: [[x1]], Y: y1}) / testsamples
print("Average cost for epoch " + str (i) + ":" + str(cost1))
dsX, dsY = shuffle (dsX, dsY) #We randomize the samples to implement a better training