def main():
size_of_learn_sample = int(len(x) * 0.9)
print(size_of_learn_sample)
nn = Network(x, y, 0.5)
nn.train()
nn.printer()
from data_loader import DataLoader
from neural_network import Network
from training_parameters import LAYER_SIZES, MINI_BATCH_SIZE, EPOCHS, LEARNING_RATE
from config import VERBOSE, TRAIN_IMAGE_DIR, TRAIN_LABEL_DIR,\
TRAINING_DATA_SKIP_FOOTER, TEST_IMAGE_DIR, TEST_LABEL_DIR, TEST_DATA_SKIP_FOOTER, TEST_PREDICTIONS_FILENAME
if VERBOSE:
print('loading training data...')
training_data = DataLoader.get_training_data(TRAIN_IMAGE_DIR, TRAIN_LABEL_DIR,
TRAINING_DATA_SKIP_FOOTER)
if VERBOSE:
print('loading test data...')
test_data = DataLoader.get_test_data(TEST_IMAGE_DIR, TEST_LABEL_DIR,
TEST_DATA_SKIP_FOOTER)
if VERBOSE:
print('begin training...')
neural_network = Network(LAYER_SIZES)
neural_network.train(training_data, test_data, MINI_BATCH_SIZE, EPOCHS,
LEARNING_RATE, VERBOSE, TEST_PREDICTIONS_FILENAME)
if __name__ == '__main__':
# Load data
X, y = load_data('./data/expert_q.txt')
# Initialize the model
nn = Network(hidden_layers=(256, 256, 256, 256),
lr=9e-2,
epoch=600,
bias=True,
batch_size=64)
try:
nn.load_weights()
except:
print('WARNING: No pre-trained networks!')
# Training
nn.train(X, y)
nn.save_weights()
# Display accuracy
accuracy(nn, X, y)
print()
# Simulate 1,000 games
N = 1000
environment = pm.PongModel(0.5, 0.5, 0.03, 0.01, 0.4)
window = gfx.GFX()
window.fps = 4e16
avg_score = simulate_game(N, window, environment, nn)
print('Avg score of {0} games:{1:.3f}'.format(N, avg_score))
from neural_network import Network
network = Network(training_iteration=500000,
learning_rate=0.3,
error_threshold=0.0001)
network.add_layer(3, 2)
network.add_layer(1)
network.train([
[[0, 0], [0]],
[[0, 1], [1]],
[[1, 0], [1]],
[[1, 1], [1]],
])
output = network.process([0, 0])
print('0 OR 0 = {}'.format(output))
output = network.process([0, 1])
print('0 OR 1 = {}'.format(output))
output = network.process([1, 0])
print('1 OR 0 = {}'.format(output))
output = network.process([1, 1])
print('1 OR 1 = {}'.format(output))
