def main(train=False):
if train:
digit_arrays = {
str(digit): [
load_image(f"{BASE_PATH}{digit}_{index}.png")
for index in range(HOW_MANY)
]
for digit in range(DIGITS)
}
flat_arrays = {
digit: [image_to_bipolar_array(image) for image in images]
for digit, images in digit_arrays.items()
}
network = Perceptron(labels=list(digit_arrays.keys()),
learning_rate=LEARNING_RATE,
activation_function=activation_function)
for label, sources in flat_arrays.items():
for source in sources:
network.add_sources(label, source)
network.train(random_starting_weights=True)
network.save_neurons(NEURONS_PICKLE_PATH)
print(f"Cycles: {network.cycles}")
print(f"Adjustments: {network.adjustments}\n")
else:
network = Perceptron()
network.load_neurons(NEURONS_PICKLE_PATH)
draw = Drawing()
while True:
test_image = draw.get_character()
# test_image = load_image(f"{BASE_PATH}0_0.png")
if test_image is None:
break
flat = image_to_bipolar_array(test_image)
out = network.output(flat)
predict = [
digit for digit, output in out.items() if output == ACTIVATED
]
if len(predict) == 0:
out_string = "The network doesn't know what digit that is."
elif len(predict) == 1:
out_string = f"The network thinks that's a {predict[0]}."
else:
out_string = f"The network thinks that might be one of these digits: {predict}."
print(out_string)
def main():
x_letter = np.full(GRID_SHAPE, NOT_ACTIVATED, dtype=np.int8)
np.fill_diagonal(x_letter, ACTIVATED)
x_letter = np.fliplr(x_letter)
np.fill_diagonal(x_letter, ACTIVATED)
x_flat = x_letter.flatten()
t_letter = np.full(GRID_SHAPE, NOT_ACTIVATED, dtype=np.int8)
t_letter[0, :] = [ACTIVATED] * GRID_WIDTH
mid_col = GRID_SHAPE[1] // 2
t_letter[:, mid_col] = [ACTIVATED] * GRID_HEIGHT
t_flat = t_letter.flatten()
targets = {"x": ACTIVATED, "t": NOT_ACTIVATED}
network = Perceptron(LEARNING_RATE, activation_function)
network.add_sources(x_flat, targets["x"])
network.add_sources(t_flat, targets["t"])
network.train(random_starting_weights=True)
print(
f'\n>>Perceptron results:\n\n'
f'Starting weights:'
f'\n{np.array2string(np.array(network.starting_weights), precision=5, max_line_width=10*GRID_WIDTH)}\n'
f'Starting bias: {network.starting_bias:.5f}\n\n'
f'Final weights:'
f'\n{np.array2string(np.array(network.weights), precision=5, max_line_width=10*GRID_WIDTH)}\n'
f'Final bias: {network.bias:.5f}\n\n'
f'Total cycles: {network.cycles}\n')
x_output = network.output(x_flat)
t_output = network.output(t_flat)
x_activation = "Activated" if x_output == ACTIVATED else "Not activated"
t_activation = "Activated" if t_output == ACTIVATED else "Not activated"
print(f'Output for letter X: {x_activation}\n'
f'Output for letter T: {t_activation}\n')
