classification_tf.py

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

def create_gaussian_data(mean1, mean2, variance1, variance2,  size_t):
    """ 
    Generate random gaussian data
    with specified mean and variance
    """

    data1 = np.random.normal(mean1, variance1, size = size_t)
    data2 = np.random.normal(mean2, variance2, size = size_t)
    data = np.hstack([data1, data2])
    labels = [ 0 if i < size_t else 1 for i in range(0, 2*size_t)]
    data_train, data_test, labels_train, lables_test = train_test_split(data, 
                                                                        labels,
                                                                        train_size = 0.7)
    return data_train, data_test, labels_train, lables_test

def creat_network(x, x_labels, test_x, test_y):
    """
        Create 2 layer Neural Network
    """
    # Define the Network
    data = tf.placeholder(tf.float32, shape = [None, 1]) # since only one dimensional
    labels = tf.placeholder(tf.float32, shape = [None, 1]) # since only one dimensional
    weight1 = tf.Variable(tf.random_normal([1, 10]), dtype = tf.float32)
    bias1 = tf.Variable([10], dtype = tf.float32)
    weight2 = tf.Variable(tf.random_normal([10, 1]), dtype = tf.float32)
    bias2 = tf.Variable([2], dtype = tf.float32)
    # Define the Operations
    hidden = tf.nn.sigmoid(tf.multiply(data, weight1) + bias1)
    y_ = tf.nn.sigmoid(tf.matmul(hidden, weight2) + bias2)
    y_ = tf.cast(tf.greater_equal(y_, 0.5), tf.float32)
    loss = tf.reduce_mean(tf.squared_difference(y_, labels)) 
    opt = tf.train.GradientDescentOptimizer(0.5).minimize(loss)
    sess = tf.Session()
    sess.run(tf.global_variables_initializer())
    # Run for 1000 epochs
    for i in range(0, 1000):
        _, l, weight = sess.run([opt, loss, weight2], feed_dict = {data: x, labels: x_labels})
        # Get Training Loss at every 100 iterations
        if(i % 100 == 0):
            print l, weight
    # Run on test set and get the classification accuracy
    correct_pred = (tf.equal(y_, test_y[0:10]))
    accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))
    print test_x
    ypred = sess.run([y_], feed_dict={data: test_x[0:10]})
    print ypred



def main():
    train_x, test_x, train_y, test_y = create_gaussian_data(0, 5, 1, 3, 10000)
    train_x = train_x.reshape(1, train_x.shape[0])
    train_y = np.array(train_y, dtype = np.float32)
    train_y = train_y.reshape(1, train_y.shape[0])
    test_x = test_x.reshape(1, test_x.shape[0])
    test_y = np.array(test_y, dtype = np.float32)
    test_y = test_y.reshape(1, test_y.shape[0])
    
    print train_y.T.shape
    creat_network(train_x.T, train_y.T, test_x.T, test_y.T)

if __name__ == '__main__':
    main()