# Option: skip the training of the neural network (debugging)
SKIP_TRAINING = True
# Code starts
if not LOAD_FROM_CACHE:
# Load training / testing data
print("Loading training and testing data...")
training, testing = [], [] # Array of tuples: (image, answer)
category_count = len(fUtil.TRAINING_DATA_NAMES)
for category in fUtil.TRAINING_DATA_NAMES:
# Print the progress
print(">> Loading " + fUtil.get_name(fUtil.get_index(category)) +
"... (" + str(fUtil.get_index(category) + 1) + "/" +
str(category_count) + ")")
# Normalize the data for more efficiency
data = fUtil.load_data(category, normalize=True)
# Split the data into training data and testing data
train_limit = int(len(data) * TRAIN_TEST_RATIO)
index = fUtil.get_index(category)
# Append the current data to master data list
training += [(image_data,
[1 if a == index else 0 for a in range(category_count)])
for image_data in data[:train_limit]]
testing += [(image_data,
[1 if a == index else 0 for a in range(category_count)])
for image_data in data[train_limit:]]
# Shuffle training / testing data
print("Shuffling training and testing data...")
random.shuffle(training)
random.shuffle(testing)
def clear(event):
nonlocal draw_data
event.widget.delete("all")
draw_data = [[0] * 28 for _ in range(28)]
def done(_):
master.destroy()
master = tk.Tk()
canvas = tk.Canvas(master, width=560, height=560)
canvas.pack()
canvas.bind('<ButtonPress-1>', draw)
canvas.bind('<B1-Motion>', draw)
canvas.bind('<Double-1>', clear)
canvas.bind('<ButtonPress-2>', done)
master.mainloop()
return np.array([a for b in draw_data for a in b])
if __name__ == "__main__":
dataset_index = -3
data = fUtil.load_data(fUtil.TRAINING_DATA_NAMES[dataset_index])
name = fUtil.TRAINING_DATA_NAMES[dataset_index][:1].upper() + fUtil.TRAINING_DATA_NAMES[dataset_index][1:-4]
show_image(data[0])
show_image(data, bulk_size=49, name=name)
data = show_drawable_canvas()
print(data)
show_image(data)