knn example with digit recognition

10_knn/README.md

@@ -0,0 +1,11 @@
+## Digit recognition with knn
+
+Here's some example Matlab code that shows KNN in action to guess handwritten digits. Here's what the output looks like:
+
+![](/images/16_correct_500_comparisons.tif)
+
+Each row represents a guess. The last column contains the image that we're trying to guess the digit for. The first three columns show the 3 nearest neighbors for that last image.
+
+Images with boxes around them represent images that we did not guess correctly. You can see KNN works pretty well -- with just 500 comparison images we are getting 80% accuracy. This jumps to 90% with 5000 comparison images (see the images/ directory).
+
+[Uses the MNIST dataset](http://yann.lecun.com/exdb/mnist/)

10_knn/main.m

@@ -0,0 +1,79 @@
+function foo = main()
+    addpath('mnistHelper');
+    train_images = loadMNISTImages('train-images-idx3-ubyte');
+    train_labels = loadMNISTLabels('train-labels-idx1-ubyte');
+
+    test_images = loadMNISTImages('t10k-images-idx3-ubyte');
+    test_labels = loadMNISTLabels('t10k-labels-idx1-ubyte');
+
+    % showData(train_images, 100, 100);
+    guesses(20, 3, train_images, train_labels, test_images, test_labels);
+end
+
+function foo = showData(images, rows, cols)
+    grid = [];
+
+    i = 0;
+    for x = 1:rows
+        imgs = [];
+        for y = 1:cols
+           i = i + 1;
+           imgs = [imgs reshape(images(:, i), 28, 28)];
+        end
+        grid = [grid; imgs];
+    end
+    imshow(grid);
+end
+
+function d = distance(train_image, test_image)
+    v = train_image - test_image;
+    v = double(v);
+    d = sqrt(v * v');
+end
+
+function result = border(image, value)
+  image = reshape(image, 28, 28);
+  result = zeros(28, 28);
+  result(:, :) = value;
+  result(2:27, 2:27) = image(2:27, 2:27);
+  result = reshape(result, 784, 1);
+end
+
+function foo = guesses(count, k, train_images, train_labels, test_images, test_labels)
+    [foo num_train_images] = size(train_images);
+    [foo num_test_images] = size(test_images);
+
+    correct = 0;
+
+    grid = [];
+    for x_ = 1:count
+        x = floor(rand() * num_test_images);
+        test_image = test_images(:, x);        
+        correct_label = test_labels(x);
+        dist = [];
+        num_train_images = 50000;
+        for i = 1:num_train_images
+
+            train_image = train_images(:, i);
+            d = distance(train_image', test_image');
+            dist = [dist; [d i]];
+        end
+        sorted_ = (sortrows(dist, 1));
+        sorted = sorted_(1:k, :);
+        labels = [];
+        for i = 1:k
+          grid = [grid train_images(:, sorted(i, 2))];
+          labels = [labels train_labels(sorted(i, 2))];
+        end
+        guess_label = mode(labels);
+        if guess_label == correct_label
+            correct = correct + 1;
+            grid = [grid test_image];
+        else
+            grid = [grid border(test_image, 255)];
+        end
+    end
+    correct
+    
+    showData(grid, count, k+1);
+end

10_knn/mnistHelper/loadMNISTImages.m

@@ -0,0 +1,26 @@
+function images = loadMNISTImages(filename)
+%loadMNISTImages returns a 28x28x[number of MNIST images] matrix containing
+%the raw MNIST images
+
+fp = fopen(filename, 'rb');
+assert(fp ~= -1, ['Could not open ', filename, '']);
+
+magic = fread(fp, 1, 'int32', 0, 'ieee-be');
+assert(magic == 2051, ['Bad magic number in ', filename, '']);
+
+numImages = fread(fp, 1, 'int32', 0, 'ieee-be');
+numRows = fread(fp, 1, 'int32', 0, 'ieee-be');
+numCols = fread(fp, 1, 'int32', 0, 'ieee-be');
+
+images = fread(fp, inf, 'unsigned char');
+images = reshape(images, numCols, numRows, numImages);
+images = permute(images,[2 1 3]);
+
+fclose(fp);
+
+% Reshape to #pixels x #examples
+images = reshape(images, size(images, 1) * size(images, 2), size(images, 3));
+% Convert to double and rescale to [0,1]
+images = double(images) / 255;
+
+end

10_knn/mnistHelper/loadMNISTLabels.m

@@ -0,0 +1,19 @@
+function labels = loadMNISTLabels(filename)
+%loadMNISTLabels returns a [number of MNIST images]x1 matrix containing
+%the labels for the MNIST images
+
+fp = fopen(filename, 'rb');
+assert(fp ~= -1, ['Could not open ', filename, '']);
+
+magic = fread(fp, 1, 'int32', 0, 'ieee-be');
+assert(magic == 2049, ['Bad magic number in ', filename, '']);
+
+numLabels = fread(fp, 1, 'int32', 0, 'ieee-be');
+
+labels = fread(fp, inf, 'unsigned char');
+
+assert(size(labels,1) == numLabels, 'Mismatch in label count');
+
+fclose(fp);
+
+end