Scribd
Upload
4 views

Assignment 2

Uploaded by

ykl.zjut
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
4 views

Assignment 2

Uploaded by

ykl.zjut
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Assignment2.

py Wed Sep 19 09:18:07 2018 1


#Camille Chow
#ECE 471 Assignment 2

import numpy as np
import tensorflow as tf
import matplotlib.pyplot as plt
from tqdm import tqdm

BATCH_SIZE = 32
NUM_BATCHES = 10000
NUM_SAMP = 400

class Data():
def __init__(self):
sigma = 0.1
np.random.seed()

#training data for label 0


self.index = np.arange(NUM_SAMP)
self.t0 = np.random.uniform(.1, 1.8, NUM_SAMP)
self.x0 = np.atleast_2d(6 * self.t0 * np.cos(2*np.pi*self.t0) + np.random.normal(
0, sigma, NUM_SAMP)).T
self.y0 = np.atleast_2d(6 * self.t0 * np.sin(2*np.pi*self.t0) + np.random.normal(
0, sigma, NUM_SAMP)).T
self.class0 = np.hstack((self.x0, self.y0))
#training data for label 1
self.t1 = np.random.uniform(.1, 1.8, NUM_SAMP)
self.x1 = np.atleast_2d(6 * self.t1 * np.cos(2*np.pi*self.t1 + np.pi) + np.random
.normal(0, sigma, NUM_SAMP)).T
self.y1 = np.atleast_2d(6 * self.t1 * np.sin(2*np.pi*self.t1 + np.pi) + np.random
.normal(0, sigma, NUM_SAMP)).T
self.class1 = np.hstack((self.x1, self.y1))

self.labels = np.hstack((np.zeros(NUM_SAMP), np.ones(NUM_SAMP)))


self.coords = np.vstack((self.class0, self.class1))

def get_batch(self):
index = np.arange(2 * NUM_SAMP)
choices = np.random.choice(index, size=BATCH_SIZE)

return self.coords[choices], self.labels[choices]

def f(x): #takes x in R2 and outputs logits and sigmoid


layer1_size = 50
layer2_size = 50

#trainable variables
w1 = tf.get_variable(’w1’, [2, layer1_size], tf.float32, tf.random_normal_initializer
())
b1 = tf.get_variable(’b1’, [layer1_size], tf.float32, tf.zeros_initializer())
w2 = tf.get_variable(’w2’, [layer1_size, layer2_size], tf.float32, tf.random_normal_i
nitializer())
b2 = tf.get_variable(’b2’, [layer2_size], tf.float32, tf.zeros_initializer())
w3 = tf.get_variable(’w3’, [layer2_size, 1], tf.float32, tf.random_normal_initializer
())
b3 = tf.get_variable(’b3’, [], tf.float32, tf.zeros_initializer())

#perceptron
layer1 = tf.add(tf.matmul(x, w1), b1)
layer2 = tf.add(tf.matmul(tf.nn.elu(layer1), w2), b2)
out = tf.add(tf.matmul(tf.nn.elu(layer2), w3), b3)

return tf.squeeze(out), tf.squeeze(tf.sigmoid(out))

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


t = tf.placeholder(tf.float32, [None])
logits, t_hat = f(x)

#optimization functions
lam = .001
Assignment2.py Wed Sep 19 09:18:07 2018 2
loss = tf.losses.sigmoid_cross_entropy(t, logits) + tf.add_n([lam * tf.nn.l2_loss(v) for
v in tf.trainable_variables()])
optim = tf.train.MomentumOptimizer(learning_rate=.01, momentum=.95).minimize(loss)

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

data = Data()

#training
for _ in tqdm(range(0, NUM_BATCHES)):
x_np, t_np = data.get_batch()
loss_np, _ = sess.run([loss, optim], feed_dict={x: x_np, t: t_np})

xaxis = np.linspace(-13, 13, 100)


yaxis = np.linspace(-13, 13, 100)
xx, yy = np.meshgrid(xaxis,yaxis)
coords = np.array(list(zip(xx.flatten(), yy.flatten())))

z = sess.run(t_hat, feed_dict={x: coords})


z = np.reshape(z, [100, 100])

plt.figure(1, figsize=[13,13])

plt.contourf(xx, yy, z)

plt.scatter(data.x0,data.y0, edgecolor=’k’)
plt.scatter(data.x1,data.y1, edgecolor=’k’)

plt.xlabel(’x’)
plt.ylabel(’y’)
plt.title(’Spirals’)

plt.show()
Spirals

10

0
y

10

10 5 0 5 10
x

You might also like