Add files via upload

Author: MuhammedBuyukkinaci

multiclass_classifcation_cpu.py

@@ -0,0 +1,227 @@
+"""
+Created on Sat Mar 24 11:25:41 2018
+
+@author: Muhammed Buyukkinaci
+"""
+
+import tensorflow as tf # ML library for graphs
+
+#Importing libraries
+import cv2 # image processing
+import numpy as np # mathematical operations
+import os # working with directories
+from random import shuffle # mixing up or currently ordered data that might lead our network astray in training.
+from tqdm import tqdm # a nice progress bar for loops
+
+#WORKING DIRECTORIES
+TRAIN_DIR = 'C:\\Users\\Muhammed Buyukkinaci\\python\\images\\multiclass_classification\\training'
+TEST_DIR = 'C:\\Users\\Muhammed Buyukkinaci\\python\\images\\multiclass_classification\\testing'
+IMG_SIZE = 80# our images are 80x80x3
+
+print(os.getcwd())
+
+#Converting the output into one-hot format
+def label_img(img):
+ word_label = img.split('_')[0]
+ if word_label == 'chair': return [1,0,0,0]
+ elif word_label == 'kitchen': return [0,1,0,0]
+ elif word_label == 'knife': return [0,0,1,0]
+ elif word_label == 'saucepan': return [0,0,0,1]
+
+
+#A function to read images from training folder
+def create_train_data():
+ training_data = []
+ for img in tqdm(os.listdir(TRAIN_DIR)):
+ label = label_img(img)
+ path = TRAIN_DIR + '\\' + img
+ img = cv2.imread(path,1)
+ img = cv2.resize(img, (IMG_SIZE,IMG_SIZE))
+ training_data.append([np.array(img),np.array(label)])
+ shuffle(training_data)
+ np.save('train_data_mc.npy', training_data)
+ return training_data
+
+#If you are reading images, make the line below uncomment
+#train_data = create_train_data()
+#If you are reading .npy file, use the line below.
+train_data = np.load('train_data_mc.npy')
+
+
+#A function to read images from testing folder
+def process_test_data():
+ testing_data = []
+ for img in tqdm(os.listdir(TEST_DIR)):
+ label = label_img(img)
+ path = TEST_DIR + '\\' + img
+ img = cv2.imread(path,1)
+ img = cv2.resize(img, (IMG_SIZE,IMG_SIZE))
+ testing_data.append([np.array(img),label]) 
+ shuffle(testing_data)
+ np.save('test_data_mc.npy', testing_data)
+ return testing_data
+
+#If you are reading images, make the line below uncomment.
+
+#test_data = process_test_data()
+
+#If you are reading .npy file, use the line below.
+test_data = np.load('test_data_mc.npy')
+
+#For visualizations
+import matplotlib.pyplot as plt
+
+#A CHAIR
+plt.imshow(np.array(train_data[180][0]))
+
+#KITCHEN
+plt.imshow(np.array(train_data[100][0]))
+
+#KNIFE
+plt.imshow(np.array(train_data[111][0]))
+
+#SAUCEPAN
+plt.imshow(np.array(train_data[154][0]))
+
+#Splitting train and CV data
+train = train_data[:4800]
+cv = train_data[4800:]
+
+X = np.array([i[0] for i in train]).reshape(-1,IMG_SIZE,IMG_SIZE,3)
+Y = np.array([i[1] for i in train])
+
+cv_x = np.array([i[0] for i in cv]).reshape(-1,IMG_SIZE,IMG_SIZE,3)
+cv_y = np.array([i[1] for i in cv])
+
+test_x = np.array([i[0] for i in test_data]).reshape(-1,IMG_SIZE,IMG_SIZE,3)
+test_y = np.array([i[1] for i in test_data])
+
+print(X.shape)
+print(Y[2])
+print(cv_x.shape)
+print(test_x.shape)
+
+#HYPERPARAMETERS
+steps = 4800
+epochs = 60
+step_size = 16
+total_batch = int(steps/step_size)
+LR = 0.00001
+
+#FUNCTIONS TO CODE LAYERS
+def init_weights(shape):
+ init_random_dist = tf.truncated_normal(shape, stddev=0.1)
+ return tf.Variable(init_random_dist)
+
+def init_bias(shape):
+ init_bias_vals = tf.constant(0.1, shape=shape)
+ return tf.Variable(init_bias_vals)
+
+def conv2d(x, W):
+ return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')
+
+def max_pool_2by2(x):
+ return tf.nn.max_pool(x, ksize=[1, 2, 2, 1],
+ strides=[1, 2, 2, 1], padding='SAME')
+
+def convolutional_layer(input_x, shape):
+ W = init_weights(shape)
+ b = init_bias([shape[3]])
+ return tf.nn.relu(conv2d(input_x, W) + b)
+
+def normal_full_layer(input_layer, size):
+ input_size = int(input_layer.get_shape()[1])
+ W = init_weights([input_size, size])
+ b = init_bias([size])
+ return tf.matmul(input_layer, W) + b
+
+print((X[5].shape)
+
+#GRAPH
+tf.reset_default_graph()
+
+#Defining Placeholdera
+x = tf.placeholder(tf.float32,shape=[None,80,80,3])
+y_true = tf.placeholder(tf.float32,shape=[None,4])
+
+#Defining GRAPH
+with tf.name_scope('Model'):
+ convo_1 = convolutional_layer(x,shape=[6,6,3,32])
+ convo_1_pooling = max_pool_2by2(convo_1)
+ convo_2 = convolutional_layer(convo_1_pooling,shape=[6,6,32,64])
+ convo_2_pooling = max_pool_2by2(convo_2)
+ convo_3 = convolutional_layer(convo_2_pooling,shape=[6,6,64,64])
+ convo_3_pooling = max_pool_2by2(convo_3)
+ convo_4 = convolutional_layer(convo_3_pooling,shape=[6,6,64,128])
+ convo_4_pooling = max_pool_2by2(convo_4)
+ convo_2_flat = tf.reshape(convo_4_pooling,[-1,5*5*128])
+ 
+ full_layer_one = tf.nn.relu(normal_full_layer(convo_2_flat,4096))
+ hold_prob1 = tf.placeholder(tf.float32)
+ full_one_dropout = tf.nn.dropout(full_layer_one,keep_prob=hold_prob1)
+ 
+ full_layer_two = tf.nn.relu(normal_full_layer(full_one_dropout,2048))
+ hold_prob2 = tf.placeholder(tf.float32)
+ full_two_dropout = tf.nn.dropout(full_layer_two,keep_prob=hold_prob2)
+ 
+ full_layer_three = tf.nn.relu(normal_full_layer(full_two_dropout,1024))
+ hold_prob3 = tf.placeholder(tf.float32)
+ full_three_dropout = tf.nn.dropout(full_layer_three,keep_prob=hold_prob3)
+ 
+ full_layer_four = tf.nn.relu(normal_full_layer(full_three_dropout,512))
+ hold_prob4 = tf.placeholder(tf.float32)
+ full_four_dropout = tf.nn.dropout(full_layer_four,keep_prob=hold_prob4)
+ 
+ y_pred = normal_full_layer(full_four_dropout,4)
+
+#Writing Loss and Accuracy
+with tf.name_scope('Loss'):
+ cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(labels=y_true,logits=y_pred))
+
+with tf.name_scope('ADAM'):
+ train = tf.train.AdamOptimizer(learning_rate=LR).minimize(cross_entropy)
+
+with tf.name_scope('Accuracy'):
+ matches = tf.equal(tf.argmax(y_pred,1),tf.argmax(y_true,1))
+ acc = tf.reduce_mean(tf.cast(matches,tf.float32))
+
+init = tf.global_variables_initializer()
+
+tf.summary.scalar("loss", cross_entropy)
+tf.summary.scalar("accuracy", acc)
+merged_summary_op = tf.summary.merge_all()
+
+acc_list = []
+logloss_list = []
+acc_train = []
+
+saver = tf.train.Saver()
+
+#If you are using CPU, just use with tf.Session() as sess:
+#Starting Session
+
+with tf.Session() as sess:
+ sess.run(init)
+ summary_writer = tf.summary.FileWriter(TRAIN_DIR, graph=tf.get_default_graph())
+ for i in range(epochs):
+ for j in range(0,steps,step_size): 
+ _ , c , summary,d = sess.run([train,cross_entropy,merged_summary_op,acc],feed_dict={x:X[j:j+step_size] , y_true:Y[j:j+step_size] ,hold_prob1:0.5,hold_prob2:0.5,hold_prob3:0.5,hold_prob4:0.5})
+ summary_writer.add_summary(summary, i * total_batch + j)
+ acc_train.append(d)
+ mean_of_ll = sess.run(cross_entropy,feed_dict={x:cv_x,y_true:cv_y ,hold_prob1:1.0,hold_prob2:1.0,hold_prob3:1.0,hold_prob4:1.0})
+ mean_of_acc = sess.run(acc,feed_dict={x:cv_x ,y_true:cv_y ,hold_prob1:1.0,hold_prob2:1.0,hold_prob3:1.0,hold_prob4:1.0})
+ logloss_list.append(mean_of_ll)
+ acc_list.append(mean_of_acc)
+ print(i,mean_of_ll,mean_of_acc)
+ saver.save(sess, os.getcwd()+"\\CNN_MC.ckpt")
+ print("test accuracy = ",np.mean([sess.run(acc,feed_dict={x:test_x[:400],y_true:test_y[:400] ,hold_prob1:1.0,hold_prob2:1.0,hold_prob3:1.0,hold_prob4:1.0}),sess.run(acc,feed_dict={x:test_x[400:800],y_true:test_y[400:800] ,hold_prob1:1.0,hold_prob2:1.0,hold_prob3:1.0,hold_prob4:1.0}),sess.run(acc,feed_dict={x:test_x[800:],y_true:test_y[800:] ,hold_prob1:1.0,hold_prob2:1.0,hold_prob3:1.0,hold_prob4:1.0})])) 
+ print("cross_entropy loss = ",np.mean([sess.run(cross_entropy,feed_dict={x:test_x[:400],y_true:test_y[:400] ,hold_prob1:1.0,hold_prob2:1.0,hold_prob3:1.0,hold_prob4:1.0}),sess.run(cross_entropy,feed_dict={x:test_x[400:800],y_true:test_y[400:800] ,hold_prob1:1.0,hold_prob2:1.0,hold_prob3:1.0,hold_prob4:1.0}),sess.run(cross_entropy,feed_dict={x:test_x[800:],y_true:test_y[800:] ,hold_prob1:1.0,hold_prob2:1.0,hold_prob3:1.0,hold_prob4:1.0})])) 
+
+import pandas as pd
+
+pd.Series(acc_list).plot(kind='line',title='Accuracy on CV data')
+
+pd.Series(logloss_list).plot(kind='line',title='CV ERROR')
+
+
+