L-Regression.py

#!/usr/bin/env python
# coding: utf-8

# In[1]:


import numpy as np
import matplotlib.pyplot as plt


# In[2]:


N = 100
x_zeros = np.random.multivariate_normal(mean=np.array(
    (-1, -1)), cov=0.1 * np.eye(2), size=(N // 2,))
y_zeros = np.zeros((N // 2,), dtype=np.float32)


# In[3]:


print(f"X0 shape : {x_zeros.shape}, Y0 shape : {y_zeros.shape}")


# In[4]:


x_ones = np.random.multivariate_normal(
    mean=np.array((1, 1)), cov=0.1 * np.eye(2), size=(N // 2,))
y_ones = np.ones((N // 2,), dtype=np.float32)


# In[5]:


print(f"X1 shape : {x_ones.shape}, Y1 shape : {y_ones.shape}")


# In[6]:


# we are keeping y of x_zeros is zero and y of x_ones one.


# In[7]:


X = np.concatenate((x_zeros, x_ones))
print(f"X shape : {X.shape}")


# In[8]:


Y = np.concatenate((y_zeros, y_ones)).reshape((100, 1))
print(f"Y shape : {Y.shape}")


# In[9]:


plt.scatter(x_zeros[:, 0], x_zeros[:, 1], color="blue")
plt.scatter(x_ones[:, 0], x_ones[:, 1], color="red")
plt.show()


# In[10]:


import tensorflow as tf


# In[11]:


with tf.name_scope("placeholders"):
    x = tf.placeholder(tf.float32, (100, 2))
    y = tf.placeholder(tf.float32, (100, 1))

with tf.name_scope("Wights"):
    W = tf.Variable(tf.random_normal((1, 2)))
    b = tf.Variable(tf.random_normal((1, 1)))

with tf.name_scope("Prediction"):
    Z = tf.transpose(tf.add(tf.matmul(W, tf.transpose(x)), b))

with tf.name_scope("loss"):
    L = tf.nn.sigmoid_cross_entropy_with_logits(labels=y, logits=Z)
    L = tf.reduce_mean(L)

with tf.name_scope("Optimizer"):
    train_op = tf.train.AdamOptimizer(0.001).minimize(L)

with tf.name_scope("summaries"):
    tf.summary.scalar("loss", L)
    merged = tf.summary.merge_all()

train_writer = tf.summary.FileWriter(
    r'C:\Users\Shambu\Google Drive\Deep Learning\Notebooks', tf.get_default_graph())


# In[12]:


with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    # training
    for i in range(2001):
        feed_dict = {x: X, y: Y}
        _, loss = sess.run([train_op, L], feed_dict=feed_dict)
        summary = sess.run(merged, feed_dict=feed_dict)
        if i % 100 is 0:
            print(f"Epoch : {i}, Loss : {loss}")
        train_writer.add_summary(summary, i)
    w_out = sess.run(W)
    logits = sess.run(Z, feed_dict={x: X})
    y_hat = sess.run(tf.math.sigmoid(logits))
    y_hat[y_hat >= 0.5] = 1
    y_hat[y_hat < 0.5] = 0
    b_out = sess.run(b)


# In[18]:


accuracy = 1 - (np.sum(abs(Y - y_hat))) / Y.shape[0]
print(f"Accuracy of the model : {accuracy*100}%")


# In[14]:


def decision_points(b, w1, w2):
    max_x1 = np.max(X[:, 0])
    max_x2 = np.max(X[:, 1])
    x1 = -b / w1 - (max_x2 * w2) / w1
    x2 = -b / w2 - (max_x1 * w1) / w2
    return [max_x1, x2], [x1, max_x2]


# In[15]:


plt.scatter(x_zeros[:, 0], x_zeros[:, 1], color="blue")
plt.scatter(x_ones[:, 0], x_ones[:, 1], color="red")
w = np.squeeze(w_out)
x, y = decision_points(b_out, w[0], w[1])
plt.plot(x, y, color="black")
plt.show()