def run_knn():
# get 100 rows
cases = 300
start = random.randint(0, len(test_data) - (cases + 1))
examples = dict()
for k, v in list(test_data.items())[start:start+cases]:
examples[k] = v
right = 0
wrong = 0
i = 1
correct_certain = []
incorrect_certain = []
positives = 0
negatives = 0
correct_avg_dist = []
incorrect_avg_dist = []
correct_min_dist = []
incorrect_min_dist = []
for key in examples:
row = examples[key]
actual = row[0]
classification = clsfy.knn(row[1:], train_data, 2, 3)
if classification[0] == 0:
negatives += 1
else:
positives += 1
if classification[0] == actual:
right += 1
print("Correct classification. Certainty: ", classification[1])
correct_certain.append(classification[1])
correct_min_dist.append(classification[2])
correct_avg_dist.append(classification[3])
else:
print("Incorrect classification. Certainty: ", classification[1], ' Actual: ', actual)
incorrect_certain.append(classification[1])
incorrect_min_dist.append(classification[2])
incorrect_avg_dist.append(classification[3])
wrong += 1
i += 1
print("\n\nRight: ", right, '/', cases, ' = ', right/cases)
print("Wrong: ", wrong, '/', cases, ' = ', wrong/cases)
print("\nCorrect Certainty: ", sum(correct_certain)/len(correct_certain))
print("Incorrect Certainty: ", sum(incorrect_certain)/len(incorrect_certain))
print("\nPositives: ", positives, '/', cases)
print("Negatives: ", negatives, '/', cases)
print("\nCorrect Min. Distance Average: ", sum(correct_min_dist)/len(correct_min_dist))
print("Incorrect Min. Distance Average: ", sum(incorrect_min_dist)/len(incorrect_min_dist))
print("\nCorrect Avg Distance Average: ", sum(correct_avg_dist)/len(correct_avg_dist))
print("Incorrect Avg Distance Average: ", sum(incorrect_avg_dist)/len(incorrect_avg_dist))
def main_loop(neighbors, training_data):
while True:
print_options()
choice = input('Select an option: ')
if choice.lower() == 'e':
break
elif choice.lower() == 'n':
x_t = get_iris_data()
classification = clsfy.knn(x_t, training_data, 3, neighbors)[0]
if classification == IRIS_SETOSA[0]:
print('Classification: ', IRIS_SETOSA[1])
elif classification == IRIS_VERSICOLOR[0]:
print('Classification: ', IRIS_VERSICOLOR[1])
elif classification == IRIS_VIRGINICA[0]:
print('Classification: ', IRIS_VIRGINICA[1])
else:
print('Invalid option.')
return
train[k] = v
test = dict()
for k, v in list(data.items())[-100:]:
test[k] = v
right = 0
wrong = 0
correct_avg_dists = []
correct_min_dists = []
incorrect_min_dists = []
incorrect_avg_dists = []
for key in test:
row = test[key]
actual_class = row[0]
class_data = clsfy.knn(row[1:], train, 3, neighbors)
predicted_class = class_data[0]
if actual_class != predicted_class:
wrong += 1
incorrect_min_dists.append(class_data[1])
incorrect_avg_dists.append(class_data[2])
else:
right += 1
correct_min_dists.append(class_data[1])
correct_avg_dists.append(class_data[2])
print('\n\nNeighbors: ', neighbors)
print("Correct: ", right, '/', len(test), ' = ', right / len(test))
print("Wrong: ", wrong, '/', len(test), ' = ', wrong / len(test))
# w: write params to json
if cv2.waitKey(1) & 0xFF == ord('w'):
data = {
'digit_spacing': spacing,
'digit_start': digit_start,
'digit_width': width,
'y_offset_bottom': y_offset_bottom,
'y_offset_top': y_offset_top
}
print(data)
with open('segvision.json', 'w+') as f:
json.dump(data, f, sort_keys=True, indent=4)
# i: identify image on command
if cv2.waitKey(1) & 0xFF == ord('i'):
identified = knn(cropped_digits, 7)
print(identified)
# s: save images
if cv2.waitKey(1) & 0xFF == ord('s'):
for i, digit in enumerate(cropped_digits):
cv2.imwrite("digit" + str(i) + ".jpg", roi)
# esc: quit
if cv2.waitKey(1) & 0xFF == 27:
break
crop = False
except TypeError:
pass
def extract_digits(img):
y_offset_top = 2
y_offset_bottom = 3
window_w = WINDOW_MIN
h, w = img.shape[:2]
vals = []
crop_start = []
# find the coords of the first digit
digit_start = find_digit_start(img)
digits = []
# loop through enitre screen
for x in range(digit_start, w, window_w):
digit = Digit(img)
# digit candidates
y0 = 0 + y_offset_top
y1 = h - y_offset_bottom
x0 = x
x1 = x + window_w
ret, digit_crop, percent_white = crop_digits(img, y0, y1, x0, x1)
if ret == 0:
# digits.append(digit_crop)
digit.add_crop(digit_crop)
digit.add_coords((y0, y1, x0, x1))
digits.append(digit)
# display these digits
sep_digits = []
results = []
for digit in digits:
# account for variance in digit crops
more_digits = digit_variance(digit, 2)
sep_digits.append(more_digits)
ret = knn(digit.crop, k=5)
# print('result:', ret)
results.append(ret)
date = str(datetime.now())
name = "screen" + date[:10] + "_" + date[20:] + ".jpg"
# cv2.imshow('digit', digit.crop)
# cv2.waitKey(0)
# # go through all the digits and run them through knn
# print('digit group size', len(sep_digits))
# for digit_group in sep_digits:
# results = []
# print('size:', len(digit_group))
# for digit in digit_group:
# img = digit.crop
# cv2.imshow('digit', img)
# cv2.waitKey(0)
# ret = knn(cv2.cvtColor(img,cv2.COLOR_BGR2GRAY), k=5)
# # print(ret)
# for val in ret:
# results.append(val)
# # identify the digit
# if len(results) > 0:
# try:
# print('digit is:',statistics.mode(results))
# except statistics.StatisticsError:
# print('digit is:',results[0])
# # cv2.imshow('digit', digit.crop)
# # cv2.waitKey(0)
return results
else:
english_preprocessor = EnglishProcessor()
positional_index = Positional(
preprocessor=english_preprocessor)
positional_index.add_docs(train['Text'])
tf_idf = TF_IDF(positional_index, english_preprocessor)
docvecs = np.transpose(tf_idf.tf_idf_matrix)
inferred_X_test = []
for doc in test['Text']:
_, vec = tf_idf.search(doc, True)
inferred_X_test.append(vec.T)
k = int(input("Enter the k value"))
knn_predict = knn(docvecs, inferred_X_test, y_train, k)
knn_precision = precision_score(y_test,
knn_predict,
average='micro')
knn_recall = recall_score(y_test, knn_predict, average='micro')
knn_f1 = f1_score(y_test, knn_predict, average='micro')
knn_f1_manually = (2 * knn_precision *
knn_recall) / (knn_precision + knn_recall)
knn_accuracy = accuracy_score(y_test, knn_predict)
print('knn result: ', knn_predict)
print('knn precision: ', knn_precision)
print('knn recall: ', knn_recall)
print('knn F1 score: ', knn_f1)
print('knn F1 manually: ', knn_f1_manually)
print('knn accuracy: ', knn_accuracy)
print('metrics:')