CNN for Visual Recognition (assignment1_Q1)

  1 import numpy as np
  2 from matplotlib.cbook import todate
  3 
  4 class KNearestNeighbor:
  5   """ a kNN classifier with L2 distance """
  6 
  7   def __init__(self):
  8     pass
  9 
 10   def train(self, X, y):
 11     """
 12     Train the classifier. For k-nearest neighbors this is just 
 13     memorizing the training data.
 14 
 15     Input:
 16     X - A num_train x dimension array where each row is a training point.
 17     y - A vector of length num_train, where y[i] is the label for X[i, :]
 18     """
 19     self.X_train = X
 20     self.y_train = y
 21     
 22   def predict(self, X, k=1, num_loops=0):
 23     """
 24     Predict labels for test data using this classifier.
 25 
 26     Input:
 27     X - A num_test x dimension array where each row is a test point.
 28     k - The number of nearest neighbors that vote for predicted label
 29     num_loops - Determines which method to use to compute distances
 30                 between training points and test points.
 31 
 32     Output:
 33     y - A vector of length num_test, where y[i] is the predicted label for the
 34         test point X[i, :].
 35     """
 36     if num_loops == 0:
 37       dists = self.compute_distances_no_loops(X)
 38     elif num_loops == 1:
 39       dists = self.compute_distances_one_loop(X)
 40     elif num_loops == 2:
 41       dists = self.compute_distances_two_loops(X)
 42     else:
 43       raise ValueError(‘Invalid value %d for num_loops‘ % num_loops)
 44 
 45     return self.predict_labels(dists, k=k)
 46 
 47   def compute_distances_two_loops(self, X):
 48     """
 49     Compute the distance between each test point in X and each training point
 50     in self.X_train using a nested loop over both the training data and the 
 51     test data.
 52 
 53     Input:
 54     X - An num_test x dimension array where each row is a test point.
 55 
 56     Output:
 57     dists - A num_test x num_train array where dists[i, j] is the distance
 58             between the ith test point and the jth training point.
 59     """
 60     num_test = X.shape[0]
 61     num_train = self.X_train.shape[0]
 62     dists = np.zeros((num_test, num_train))
 63     for i in xrange(num_test):
 64       for j in xrange(num_train):
 65         #####################################################################
 66         # TODO:                                                             #
 67         # Compute the l2 distance between the ith test point and the jth    #
 68         # training point, and store the result in dists[i, j]               #
 69         #####################################################################
 70         dists[i,j] = np.sqrt(np.sum(np.square(X[i,:] - self.X_train[j,:])))
 71         #####################################################################
 72         #                       END OF YOUR CODE                            #
 73         #####################################################################
 74     return dists
 75 
 76   def compute_distances_one_loop(self, X):
 77     """
 78     Compute the distance between each test point in X and each training point
 79     in self.X_train using a single loop over the test data.
 80 
 81     Input / Output: Same as compute_distances_two_loops
 82     """
 83     num_test = X.shape[0]
 84     num_train = self.X_train.shape[0]
 85     dists = np.zeros((num_test, num_train))
 86     for i in xrange(num_test):
 87       #######################################################################
 88       # TODO:                                                               #
 89       # Compute the l2 distance between the ith test point and all training #
 90       # points, and store the result in dists[i, :].                        #
 91       #######################################################################
 92       dists[i, :] = np.sqrt(np.sum(np.square(self.X_train - X[i,:]), axis=1))
 93       #######################################################################
 94       #                         END OF YOUR CODE                            #
 95       #######################################################################
 96     return dists
 97 
 98   def compute_distances_no_loops(self, X):
 99     """
100     Compute the distance between each test point in X and each training point
101     in self.X_train using no explicit loops.
102 
103     Input / Output: Same as compute_distances_two_loops
104     """
105     num_test = X.shape[0]
106     num_train = self.X_train.shape[0]
107     dists = np.zeros((num_test, num_train)) 
108     #########################################################################
109     # TODO:                                                                 #
110     # Compute the l2 distance between all test points and all training      #
111     # points without using any explicit loops, and store the result in      #
112     # dists.                                                                #
113     # HINT: Try to formulate the l2 distance using matrix multiplication    #
114     #       and two broadcast sums.                                         #
115     #########################################################################
116     tDot = np.multiply(np.dot(X, self.X_train.T), -2)
117     t1 = np.sum(np.square(X), axis=1, keepdims=True)
118     t2 = np.sum(np.square(self.X_train), axis=1)
119     tDot = np.add(t1, tDot)
120     tDot = np.add(tDot, t2)
121     dists = np.sqrt(tDot)
122     #########################################################################
123     #                         END OF YOUR CODE                              #
124     #########################################################################
125     return dists
126 
127   def predict_labels(self, dists, k=1):
128     """
129     Given a matrix of distances between test points and training points,
130     predict a label for each test point.
131 
132     Input:
133     dists - A num_test x num_train array where dists[i, j] gives the distance
134             between the ith test point and the jth training point.
135 
136     Output:
137     y - A vector of length num_test where y[i] is the predicted label for the
138         ith test point.
139     """
140     num_test = dists.shape[0]
141     y_pred = np.zeros(num_test)
142     for i in xrange(num_test):
143       # A list of length k storing the labels of the k nearest neighbors to
144       # the ith test point.
145       closest_y = []
146       #########################################################################
147       # TODO:                                                                 #
148       # Use the distance matrix to find the k nearest neighbors of the ith    #
149       # training point, and use self.y_train to find the labels of these      #
150       # neighbors. Store these labels in closest_y.                           #
151       # Hint: Look up the function numpy.argsort.                             #
152       #########################################################################
153       # pass
154       closest_y = self.y_train[np.argsort(dists[i, :])[:k]]
155       #########################################################################
156       # TODO:                                                                 #
157       # Now that you have found the labels of the k nearest neighbors, you    #
158       # need to find the most common label in the list closest_y of labels.   #
159       # Store this label in y_pred[i]. Break ties by choosing the smaller     #
160       # label.                                                                #
161       #########################################################################
162       
163       y_pred[i] = np.argmax(np.bincount(closest_y))
164       #########################################################################
165       #                           END OF YOUR CODE                            # 
166       #########################################################################
167 
168     return y_pred