pru.py

#https://github.com/delip/blog-stuff/blob/master/tensorflow_ufp.ipynb

#https://github.com/delip/blog-stuff/blob/master/tensorflow_ufp.ipynb

import tensorflow as tf
import numpy as np
import math, random


np.random.seed(100)
funtion_to_learn = lambda x: x * 2

NODOS = 10
EXAMPLES = 1000
TRAIN_SPLIT = .8
BATCH_SIZE = 1000
EPOCS = 1000

todos = np.random.uniform(-1000, 1000, 1000)
# todos = np.arange(1000)
trainSIZE = int(EXAMPLES * TRAIN_SPLIT)

a = np.split(todos, [trainSIZE])
trainX = a[0].reshape(trainSIZE, 1);
validX = a[1].reshape(EXAMPLES - trainSIZE, 1)

trainY = funtion_to_learn(trainX)
validY = funtion_to_learn(validX)


X = tf.placeholder(tf.float32, [None, 1], name = "X")
Y = tf.placeholder(tf.float32, [None, 1], name = "Y")

w = tf.Variable(tf.zeros([1, NODOS], dtype=tf.float32))
b = tf.Variable(tf.zeros([1, NODOS], dtype=tf.float32))
h = tf.nn.sigmoid(tf.matmul(X , w ) + b)


wo = tf.Variable(tf.zeros([NODOS, 1], dtype=tf.float32))
bo = tf.Variable(tf.zeros([1, 1], dtype=tf.float32))

y_ = tf.matmul(h , wo ) + bo

train_op = tf.train.AdamOptimizer().minimize(tf.nn.l2_loss(y_ - Y))

sess = tf.Session()
sess.run(tf.global_variables_initializer())

errors = []
for i in range(EPOCS):
    for start, end in zip(range(0, trainX.size, BATCH_SIZE),
                          range(BATCH_SIZE, trainX.size, BATCH_SIZE)):
        print (start, end)
        sess.run(train_op, feed_dict={X: trainX, Y: trainY})

    mse = sess.run(tf.nn.l2_loss(y_ - validY), feed_dict={X: validX})
    errors.append(mse)
    if i%100 == 0: print ("epoch %d, validation MSE %s" % (i, mse))