PCA model added

Author: NikhilGupta1997

Principal Component Analysis/PCA.py

@@ -0,0 +1,136 @@
+'''
+============================
+Principal Component Analysis
+============================
+
+'''
+
+from sklearn import svm
+import numpy as np
+import matplotlib.pyplot as plt
+import math
+from matplotlib import cm
+from mpl_toolkits.mplot3d import Axes3D
+from matplotlib.ticker import LinearLocator, FormatStrFormatter
+import matplotlib.animation as animation
+import random
+import glob
+import os
+import sys, getopt
+from time import clock
+from scipy import ndimage
+from scipy import misc
+from PIL import Image
+from scipy.misc import toimage
+from scipy.interpolate import interp1d
+from sklearn.model_selection import cross_val_score
+
+width = 0
+hight = 0
+
+def readX(file):
+	global width
+	global height
+	inputs = []
+	labels = []
+	class_label = -1;
+	for dirName, subdirList, fileList in os.walk(file):
+		print dirName, subdirList, fileList
+		class_label += 1
+		for filename in fileList:
+			if dirName != file and filename != ".DS_Store":
+				image = Image.open(os.path.join(dirName, filename)).convert('L')
+				width, height = image.size
+				inputs.append(np.asarray(image).flatten().tolist())
+				labels.append(class_label)
+	return np.matrix(inputs), labels
+
+def display_face(face, name):
+	data = face.copy()
+	data.resize(height, width)
+	toimage(data).save('{}.png'.format(name))
+
+def map_gray(eigenface):
+	interpolate = interp1d([eigenface.min(), eigenface.max()],[0,255])
+	return interpolate(eigenface)
+
+def convert_to_list(inputs):
+	train_data = []
+	for i in xrange(inputs.shape[0]):
+		param = []
+		for j in xrange(inputs.shape[1]):
+			param.append(inputs.item(i,j))
+		train_data.append(param)
+	return train_data
+
+inputs, labels = readX(str(sys.argv[1]))
+print labels
+
+# Part A
+average_face = np.mean(inputs, axis=0)
+mean_faces = np.subtract(inputs, average_face)
+normalized_faces = np.divide(mean_faces, np.std(mean_faces, axis=0)).T
+print normalized_faces.shape
+display_face(average_face, "average_face")
+
+# Part B
+U, s, V = np.linalg.svd(normalized_faces, full_matrices=False)
+best_50 = np.matrix(s[:50])
+with open('principal_components.txt', 'wb') as f:
+	for line in best_50:
+		np.savetxt(f, line, fmt='%.2f')
+
+# Part C
+for i in range(0,5):
+	eigenface = U.T[i]
+	display_face(map_gray(eigenface), 'eigenface{}'.format(i))
+
+# Part D
+eigenfaces = []
+for i in range(0,50):
+	eigenfaces.append(U.T[i][0].tolist()[0])
+with open('eigenfaces.txt', 'wb') as f:
+	count = 0
+	for line in eigenfaces:
+		count += 1
+		f.write("eigenvector {}\n".format(count))
+		np.savetxt(f, line, fmt='%.6f')
+
+new_proj = np.matrix(eigenfaces) * normalized_faces
+proj = new_proj / np.linalg.norm(new_proj, axis=0)
+mean_proj = np.subtract(proj, np.mean(proj, axis=1))
+normalized_proj = np.divide(mean_proj, np.std(mean_proj, axis=1))
+
+reduced_images = []
+for row in proj.T:
+	image = map_gray(np.dot(row,eigenfaces))
+	image += average_face
+	image = map_gray(image)
+	reduced_images.append(image.tolist()[0])
+reduced_images = np.matrix(reduced_images)
+with open('reduced_images.txt', 'wb') as f:
+	count = 0
+	for line in reduced_images:
+		count += 1
+		f.write("image {}\n".format(count))
+		np.savetxt(f, line, fmt='%.2f')
+
+# Part E
+clf = svm.SVC(decision_function_shape='ovo')
+start = clock()
+scores = np.mean(cross_val_score(clf, convert_to_list(normalized_proj.T), labels, cv=10))
+end = clock()
+print "score1 = ", scores, 'in {} secs'.format(end - start)
+
+clf = svm.SVC(decision_function_shape='ovo')
+start = clock()
+scores = np.mean(cross_val_score(clf, convert_to_list(normalized_faces.T), labels, cv=10))
+end = clock()
+print "score2 = ", scores, 'in {} secs'.format(end - start)
+
+# Part F
+for i in range(1,4):
+	j = random.randint(1, inputs.shape[0])
+	display_face(inputs[j], "input{}".format(i))
+	display_face(reduced_images[j], "output{}".format(i))
+