import knn import pygame import sys import time trainingSet = knn.load("../data/optdigitsTrain.csv") testSet = knn.load("../data/optdigitsTest.csv") avg = knn.load("../data/averages.csv") def runKNNTest(): correct = 0. percentage = 0 for i in range(len(testSet)): if i * 100 / len(testSet) > percentage: percentage = i * 100 / len(testSet) print '{0}\r'.format("Progress: " + str(percentage) + "%"), prediction = knn.predictLabel(trainingSet, testSet[i][:-1], 1) correct += prediction == testSet[i][-1] print str(correct * 100 / len(testSet)) + "% correct" def runKNNTestWeighted(): w = [[.5, .5, .5, .5, .5, .5, .5, .5], [.5, .5, .5, 1, 1, .5, .5, .5], [.5, .5, 1, 1, 1, 1, .5, .5], [.5, 1, 1, 2, 2, 1, 1, .5], [.5, 1, 1, 2, 2, 1, 1, .5], [.5, .5, 1, 1, 1, 1, .5, .5], [.5, .5, .5, 1, 1, .5, .5, .5], [.5, .5, .5, .5, .5, .5, .5, .5]] w = [x for i in w for x in i] correct = 0. percentage = 0 for i in range(len(testSet)): if i * 100 / len(testSet) > percentage: percentage = i * 100 / len(testSet) print '{0}\r'.format("Progress: " + str(percentage) + "%"), prediction = knn.predictLabelWeighted(trainingSet, testSet[i][:-1], 1, w) correct += prediction == testSet[i][-1] print str(correct * 100 / len(testSet)) + "% correct" def convertImageToData(n): results = [] for x in range(8): for y in range(8): results.append(0) for i in range(4): results[-1] += n[y * 4 + i][x * 4:x * 4 + 4].count(1) return results def calculateAverages(): n = [[0 for j in range(64)]for i in range(10)] total = [0 for i in range(10)] for i in trainingSet + testSet: for j in range(64): n[i[-1]][j] += i[j] total[i[-1]] += 1 for i in range(10): n[i] = [int(round(float(n[i][a]) / total[i])) for a in range(64)] + [i] return n def vectorToBitmap(n): pass def main(): pygame.init() white = 255, 255, 255 black = 0, 0, 0 red = 255, 0, 0 green = 0, 255, 0 myfont = pygame.font.SysFont("Arial", 200) textsurface = myfont.render('', False, black) board = [[0 for i in range(32)] for j in range(32)] heatmap = convertImageToData(board) averageHeatmap = convertImageToData(board) closestHeatmap = convertImageToData(board) screenSize = 10 screen = pygame.display.set_mode((screenSize * 96, screenSize * 64)) currentTime = 0. accumulator = 0. fps = .5 strokes = [[]] adjustedStrokes = [[]] pixelated = pygame.Surface((32 * screenSize, 32 * screenSize)) penDown = False prediction = 0 minY = maxY = minX = maxX = 0 while 1: for event in pygame.event.get(): if event.type == pygame.QUIT: sys.exit() elif event.type == pygame.MOUSEMOTION: if pygame.mouse.get_pressed()[0]: if not penDown: strokes.append([]) penDown = True p = pygame.mouse.get_pos() strokes[-1].append((max(min(p[0], screenSize * 32), 0), max(min(p[1], screenSize * 32), 0))) minY = min([min(i, key=lambda n:n[1]) for i in strokes if len(i) > 0], key=lambda n: n[1])[1] maxY = max([max(i, key=lambda n:n[1]) for i in strokes if len(i) > 0], key=lambda n: n[1])[1] minX = min([min(i, key=lambda n:n[0]) for i in strokes if len(i) > 0], key=lambda n: n[0])[0] maxX = max([max(i, key=lambda n:n[0]) for i in strokes if len(i) > 0], key=lambda n: n[0])[0] totalSize = (maxX - minX + 1, maxY - minY + 1) center = ((minX + maxX) / 2, (minY + maxY) / 2) origin = (minX, minY) zoom = max(totalSize) adjustedStrokes = [[((i[0] - center[0]) * 32 * screenSize / zoom + 16 * screenSize, (i[1] - center[1]) * 32 * screenSize / zoom + 16 * screenSize) for i in j] for j in strokes] else: penDown = False elif event.type == pygame.KEYDOWN: if event.key == pygame.K_SPACE: # print knn.predictLabel(trainingSet + testSet, heatmap, 1) minY = maxY = minX = maxX = 0 strokes = [[]] adjustedStrokes = [[]] board = [[0 for i in range(32)] for j in range(32)] elif event.key == pygame.K_ESCAPE: sys.exit() elif event.key == pygame.K_RETURN: # record correct key and save to current set print "test" accumulator += time.clock() - currentTime currentTime = time.clock() while accumulator > fps: accumulator -= fps board = [[0 for i in range(32)] for j in range(32)] for i in range(32): for j in range(32): if 1 in [pixelated.get_at((screenSize * j + n, screenSize * i + m)) != white for n in range(screenSize) for m in range(screenSize)]: for y in range(-1, 2): for x in range(-1, 2): if x == 0 or y == 0: board[max(min(j + x, 31), 0)][max(min(i + y, 31), 0)] = 1 heatmap = convertImageToData(board) if not pygame.mouse.get_pressed()[0]: prediction = knn.getNeighborData(trainingSet + testSet, heatmap) closestHeatmap = prediction averageHeatmap = avg[prediction[-1]] textsurface = myfont.render(str(prediction[-1]), False, black) screen.fill((100, 100, 100)) pygame.draw.rect(screen, white, (0, 0, screenSize * 96, screenSize * 64)) # vectorized if maxX != 0 or maxY != 0: pygame.draw.line(screen, green, (minX, minY), (minX, maxY)) pygame.draw.line(screen, green, (maxX, minY), (maxX, maxY)) pygame.draw.line(screen, green, (minX, maxY), (maxX, maxY)) pygame.draw.line(screen, green, (minX, minY), (maxX, minY)) for i in strokes: for j in i: screen.set_at(j, red) # vectorized, normalized, and scaled pygame.draw.rect(pixelated, white, (0, 0, screenSize * 32, screenSize * 32)) for i in adjustedStrokes: if len(i) > 1: pygame.draw.aalines(pixelated, black, 0, i, 1) screen.blit(pixelated, (32 * screenSize, 0)) # 32x32 pixelated for j in range(32): for i in range(32): pygame.draw.rect(screen, black if board[i][j] else white, (64 * screenSize + i * screenSize, j * screenSize, screenSize, screenSize)) # 8x8 gradient for i in range(64): c = 255 - heatmap[i] * (255 / 16) pygame.draw.rect(screen, (c, c, c), ((i % 8) * screenSize * 4, 32 * screenSize + (i / 8) * screenSize * 4, screenSize * 4, screenSize * 4)) # closest 8x8 for i in range(64): c = 255 - closestHeatmap[i] * (255 / 16) pygame.draw.rect(screen, (c, c, c), (32 * screenSize + (i % 8) * screenSize * 4, 32 * screenSize + (i / 8) * screenSize * 4, screenSize * 4, screenSize * 4)) # average 8x8 for i in range(64): c = 255 - averageHeatmap[i] * (255 / 16) pygame.draw.rect(screen, (c, c, c), (32 * screenSize + (i % 8) * screenSize * 4, 64 * screenSize + (i / 8) * screenSize * 4, screenSize * 4, screenSize * 4)) # finalCharacter font draw/print screen.blit(textsurface, (76 * screenSize, 32 * screenSize)) # dividing lines pygame.draw.line(screen, red, (32 * screenSize, 0), (32 * screenSize, 64 * screenSize)) pygame.draw.line(screen, red, (64 * screenSize, 0), (64 * screenSize, 64 * screenSize)) pygame.draw.line(screen, red, (0, 32 * screenSize), (96 * screenSize, 32 * screenSize)) pygame.display.flip() if __name__ == "__main__": runKNNTestWeighted() # main() # vector input or image raw input scaled 32x32 # 8x8 closest average or example raw # pull in the vector dataset from UCI once it can translate # local set builds over time with option to correct wrong answers # press digit to correct or verify it and add current to database