def main(_):
config = nn_config.NNConfig(FLAGS.conf).get_config()
model = nn_model.NNModel(config)
if not tf.gfile.Exists(config.dump_path):
tf.gfile.MakeDirs(config.dump_path)
with tf.Session() as sess:
model.set_session(sess)
if config.train:
model.train()
else:
model.inference()
def main():
train_x, train_y = ut.read_data('train_x'), ut.read_data("train_y")
train_x = ut.norm_data(train_x)
train_y = ut.norm_data(train_y)
# leaving 20% of the data for validation
dev_size = int(len(train_x) * 0.2)
dev_x = train_x[-dev_size:, :]
dev_y = train_y[-dev_size:]
train_x, train_y = train_x[:-dev_size, :], train_y[:-dev_size]
model = nn.NNModel(784,50,10)
model.train_nn(train_x,train_y,dev_x,dev_y,20,0.1)
print("hey")
print '****** saving population...'
count = 0
for m in population:
save_model(m, '../saved_models/ga200' + str(count) + '.mdl')
count += 1
if __name__ == '__main__':
models = []
population_size = 50
if len(sys.argv) > 1:
for i in range(population_size):
model = load_model('../saved_models/ga200' + str(i) + '.mdl')
models.append(model)
else:
for i in range(population_size):
model = nn_mdl.NNModel()
model.initialize(28 * 28, 256, 128, 10)
models.append(model)
train_set, dev_set = load_mnist('../mnist_data')
roulette_wheel = []
for i in range(population_size):
for k in range(int((population_size - i)**0.5)):
roulette_wheel.append(i)
for i in range(7):
rand.shuffle(roulette_wheel)
train_classifier(train_set, dev_set, 30, models, roulette_wheel)
# rows = param.shape[0]
# mean = np.sum(param) / (rows * cols)
# std = np.sqrt(np.sum(np.abs(param - mean)) / (rows * cols))
# print 'param mean:', mean
# print 'param std:', std
# print 'param max', np.max(param)
# print 'param min', np.min(param)
# plt.imshow(np.concatenate((np.concatenate((train_set[0][0:-1].reshape((28,28)), model.feed_forward(train_set[0][0:-1]).reshape((28,28))), axis=1),
# np.concatenate((train_set[1][0:-1].reshape((28, 28)), model.feed_forward(train_set[1][0:-1]).reshape((28, 28))), axis=1),
# np.concatenate((train_set[2][0:-1].reshape((28, 28)), model.feed_forward(train_set[2][0:-1]).reshape((28, 28))), axis=1),
# np.concatenate((train_set[3][0:-1].reshape((28, 28)), model.feed_forward(train_set[3][0:-1]).reshape((28, 28))), axis=1),
# np.concatenate((train_set[4][0:-1].reshape((28, 28)), model.feed_forward(train_set[4][0:-1]).reshape((28, 28))), axis=1),
# np.concatenate((train_set[5][0:-1].reshape((28, 28)), model.feed_forward(train_set[5][0:-1]).reshape((28, 28))), axis=1),),
# axis=0))
# plt.show()
# model_file = '../saved_models/nn2classifier.mdl.pickle'
# model = nn_mdl.NNModel([256, 256, 10])
model = nn_mdl.NNModel([28 * 28, 400, 200, 10])
# model = load_model(model_file)
train_classifier(train_set,
dev_set,
num_iterations=20,
learning_rate=0.01,
model=model,
regularization=1e-7,
model_file=model_file)
# train_auto_encoder(train_set, dev_set, num_iterations=20, learning_rate=0.001, model=model,