def nielson_3layer_300rows():
labeled_data = mnist.MnistDataset.from_pickled_file(30)
config = network_config.NetworkConfig()
config.neuron_counts = [784, 30, 10]
net = nn.Network(config)
net.train(labeled_data.train, labeled_data.validate)
def invgrad_3layer():
labeled_data = mnist.MnistDataset.from_pickled_file()
config = network_config.NetworkConfig()
config.param_update_c = param_update_fns.InvGradParamUpdate()
config.neuron_counts = [784, 30, 10]
net = nn.Network(config)
net.train(labeled_data.train, labeled_data.validate)
def neilson_3layer_200rows_100epochs():
labeled_data = mnist.MnistDataset.from_pickled_file(20)
config = network_config.NetworkConfig()
config.neuron_counts = [784, 30, 10]
config.epochs = 100
net = nn.Network(config)
net.train(labeled_data.train, labeled_data.validate)
def test_mnist(self):
from loader import mnist_loader
training_data, validation_data, test_data = \
mnist_loader.load_data_wrapper()
net = network.Network([784, 30, 10])
net.SGD(training_data, 30, 10, 3.0)
self.assertTrue(net.evaluate(test_data) * 1.0 / len(test_data) >= 0.9)
def delta_boosted_3layer_300rows():
labeled_data = mnist.MnistDataset.from_pickled_file(30)
config = network_config.NetworkConfig()
config.neuron_counts = [784, 30, 10]
config.delta_boost = 4
# config.eta = 1
net = nn.Network(config)
net.train(labeled_data.train, labeled_data.validate)
def neilson_5layer_200rows_100epochs():
labeled_data = mnist.MnistDataset.from_pickled_file(20)
config = network_config.NetworkConfig()
config.init_c = init_fns.LeCunNormalInit()
# config.activation_c = activation_fns.ReLUActivation()
config.neuron_counts = [784, 30, 20, 15, 10]
net = nn.Network(config)
net.train(labeled_data.train, labeled_data.validate)
def sq_grad_3layer_300rows():
labeled_data = mnist.MnistDataset.from_pickled_file(30, True)
config = network_config.NetworkConfig()
config.neuron_counts = [784, 30, 10]
config.param_update_c = param_update_fns.SquaredParamUpdate()
config.eta = 18
config.epochs = 100
net = nn.Network(config)
net.train(labeled_data.train, labeled_data.test)
def delta_unboosted_5layer_300rows():
labeled_data = mnist.MnistDataset.from_pickled_file(30)
config = network_config.NetworkConfig()
#config.init_c = init_fns.LeCunNormalInit()
# config.activation_c = activation_fns.ReLUActivation()
config.neuron_counts = [784, 30, 20, 15, 10]
config.eta = 0.1
config.delta_boost = 1
net = nn.Network(config)
net.train(labeled_data.train, labeled_data.validate)
def nielson_6layer_full_data():
labeled_data = mnist.MnistDataset.from_pickled_file(30)
config = network_config.NetworkConfig()
config.init_c = init_fns.LeCunNormalInit()
#config.loss_c = loss_fns.QuadraticLoss()
#config.activation_c = activation_fns.SigmoidActivation()
config.neuron_counts = [784, 30, 30, 30, 30, 10]
config.eta = 0.001
config.lmbda = 0.025
config.epochs = 500
net = nn.Network(config)
net.train(labeled_data.train, labeled_data.validate)
def linear_data_classical():
def get_data(count):
inputs = [
np.reshape(k, (1, 1)) for k in np.random.uniform(-5, 5, count)
]
outputs = [[[0], [1]] if k[0] < 0 else [[1], [0]] for k in inputs]
data = list(zip(inputs, outputs, range(len(inputs))))
return data
dataset = ld.LabeledData()
dataset.train = get_data(10)
dataset.test = get_data(100)
dataset.validate = get_data(100)
config = network_config.NetworkConfig()
#config.param_update_c = param_update_fns.InvGradParamUpdate()
config.neuron_counts = [1, 3, 5, 2]
config.epochs = 50
net = nn.Network(config)
net.train(dataset)
def __init__(self, config: nn.network_config.NetworkConfig):
self.config = config
self.networks = [nn.Network(config), nn.Network(config)]
import network.network as network
import network.mnist_loader as mnist
import pickle
# Open the mnist dataset
training_data, validation_data, test_data = mnist.load_data()
# Open the adversarial dataset that we created using the "adversarial_dataset" file
with open('adversarial_samples_training_set.pkl', 'rb') as f:
adversarial_dataset = pickle.load(f, encoding="latin1")
# Train FNN using the adversarial samples along with the normal training dataset.
net2 = network.Network([784,30,10])
net2.SGD(adversarial_dataset + training_data, 100, 5, 0.1)
filename = 'FNN_adversarial.pkl'
pickle.dump(net2, open(filename, 'wb'))
if __name__ == "__main__":
if (len(sys.argv) != 4):
exit()
else:
try:
ip = sys.argv[1]
port = int(sys.argv[2])
name = sys.argv[3]
except e:
exit()
moves = ['N', 'S', 'E', 'O']
i = 0
net = network.Network(ip, port, name)
# print("Numero de joueur : "+str(net.getNumPlayer()))
mouleFixed = None
while 1:
# print("RECEPTION DU PLATEAU ")
game = net.getBoardState()
if not game:
break
myself = game.players[net.getNumPlayer()]
# print("COMPUTING DE L'IA")
depart = game.getPlayerCase(myself)
if mouleFixed and type(game.getCase(mouleFixed.y,
def train_model():
training_data, validation_data, test_data = \
mnist_loader.load_data_wrapper()
net = network.Network([784, 30, 10])
net.SGD(training_data, 30, 10, 3.0, test_data=test_data)
return net