def testCreateNetwork(self):
##Generating Phenotype
inputNode = NodeGene(0,1,1,"SENSOR")
outputNode = NodeGene(1,1,1,"OUTPUT")
conn = ConnectionGene(1,0,1,True,1)
connections = [conn]
nodes = [inputNode,outputNode]
genes = Genome(connections,nodes)
network = Network(genes)
assert(network.fire([1]) == [1])
def analysis(layers,
learningRate,
batchSize,
iteration,
probabilistic,
training_labels,
training_images,
testing_labels,
testing_images,
saveName="LastNN.pkl"):
print("\n" + str(layers) +
" - Batch {} - Rate {} - Probabilitic {}".format(
batchSize, learningRate, probabilistic))
net = Network.NeuralNetwork(layers)
print("Start backpropagation")
net.backpropagation(training_images, training_labels, learningRate,
batchSize, probabilistic, iteration)
numberTrain, outputTrain = test(net, training_images, training_labels)
numberTest, outputTest = test(net, testing_images, testing_labels)
print("Test")
print("On the training set : {} / {}".format(numberTrain,
len(training_labels)))
print("On the testing set : {} / {}".format(numberTest,
len(testing_labels)))
save.save(net, saveName)
return net, outputTrain, outputTest
def buildNetwork(self):
inputs = []
for i in range(len(self.menus) - 3):
if self.menus[i].get() == '':
self.errorLabel.grid(row=len(self.labels) + 3,
column=1,
pady=5,
sticky='w')
return
inputs.append(int(self.menus[i].get()))
# HyperParameters
try:
learnRate = float(self.menus[-1].get())
except:
self.errorLabel.grid(row=len(self.labels) + 3,
column=1,
pady=5,
sticky='w')
self.menus[-1].set('')
activationFunc = self.menus[-3].get()
errorFunc = self.menus[-2].get()
self.errorLabel.grid_remove()
Globals.NN = Network.Network(inputs, learnRate, activationFunc,
errorFunc)
Node.drawNodes(self.canvas)
def main():
parse_args()
seed = torch.seed() % 20
log_dir = cfg.TENSORBOARD_DIR + cfg.MODEL.NAME + "/" + str(seed)
logger = Logger(log_dir)
model_dir = cfg.MODEL.DIR + cfg.MODEL.NAME + "/" + str(seed)
if os.path.isdir(model_dir) != True:
os.makedirs(model_dir)
model = Network(cfg)
dump_input = torch.rand((1, cfg.DATASET.NUM_JOINTS, cfg.DEFAULT_FRAMES))
logger.add_graph(model, (dump_input, )) # Log Model Architecture
#Toggle to get model summary
summary(model, dump_input.shape[1:], batch_size=32, device="cpu")
trainer = ResTCN_trainer(model)
optimizer = trainer.optimizer
print("------STARTING TRAINING-------")
for epoch in range(cfg.EPOCHS):
training_log = trainer.train()
print("-" * 50)
print("Epoch: {} & Loss: {}".format(epoch, training_log["Loss"]))
print("-" * 50)
logger.log_scalars(training_log, logger.step)
logger.step += 1
if epoch % cfg.SAVE_FREQUENCY == 0:
perf_indicator = trainer.cal_accuracy('train')
save_checkpoint(
{
'epoch': epoch + 1,
'model': cfg.MODEL.NAME + str(seed),
'state_dict': model.state_dict(),
'perf': perf_indicator,
'optimizer': optimizer.state_dict(),
},
output_dir=model_dir)
def neural_network(dataset_train, dataset_test):
# Prendere la colonna Target del dataset
classiOutput = set()
for esempio in dataset_train:
classiOutput.add(esempio[len(esempio) - 1])
# Selezionare i Target senza duplicazione
listaClassi = []
for c in classiOutput:
listaClassi.append(c)
# Costruzione dell topologie della Rete
lng = len(dataset_train[0])
numeroInput = lng - 1
# Creo l'oggetto Network
rete = Network(numeroInput, listaClassi, math.tanh)
# Aggiunta degli strati
rete.add_hidden_layer(4)
# Lo strato di Out avrà 3 unitÃ
rete.add_output_layer()
rete.train(dataset_train)
# Classificazione
giuste = 0
sbagliate = 0
lable_learning = list()
for test in dataset_test:
output_class = rete.classify(test[:len(test) - 1]) # Passaggio dell'input alla rete
lable_learning.append(output_class)
if output_class == test[len(test) - 1]:
giuste += 1
else:
sbagliate += 1
rete.close_report()
return lable_learning
def test_XOR(self):
"""
Tests the xor gate on an Neural Network
:return:
"""
inputs = [[0, 0, 1], [0, 1, 1], [1, 1, 1], [1, 0, 1]]
answers = [[0], [1], [0], [1]]
XOR = Network(3, 2, 3, 1)
XOR.training(300000, 0.5, inputs, answers)
self.assertEqual(XOR.forward_propagation(inputs[0])[0] < 0.5, True)
self.assertEqual(XOR.forward_propagation(inputs[1])[0] > 0.5, True)
self.assertEqual(XOR.forward_propagation(inputs[2])[0] > 0.5, True)
self.assertEqual(XOR.forward_propagation(inputs[3])[0] < 0.5, True)
def test_nand(self):
"""
Tests the nand gate on an Neural Network
:return:
"""
inputs = [[0, 0], [0, 1], [1, 1], [1, 0]]
answers = [[1], [0], [0], [0]]
NAND = Network(1, 1, 2, 1)
NAND.training(50000, 0.1, inputs, answers)
self.assertEqual(NAND.forward_propagation(inputs[0])[0] > 0.5, True)
self.assertEqual(NAND.forward_propagation(inputs[1])[0] < 0.5, True)
self.assertEqual(NAND.forward_propagation(inputs[2])[0] < 0.5, True)
self.assertEqual(NAND.forward_propagation(inputs[3])[0] < 0.5, True)
def __init__(self, graphWin, player_id, popName=False, playerSize=50):
playerStartingPoint = [30, 540]
self.player_id = player_id
Rectangle.__init__(
self, Point(playerStartingPoint[0], playerStartingPoint[1]),
Point(playerStartingPoint[0] + playerSize,
playerStartingPoint[1] + playerSize))
self.graphWin = graphWin
self.speed = 0
self.status = 'nothing'
self.alive = True
self.setFill(
color_rgb(random.randint(0, 255), random.randint(0, 255),
random.randint(0, 255)))
self.setOutline(color_rgb(0, 0, 0))
if self.graphWin.renderization:
self.draw(graphWin)
if self.graphWin.gameMode == 'aitrain':
self.brain = Network(popName + str(player_id))
def optimize(network: Network, x, y, epochs=5, batch_size=10, alpha=0.03, validation_split=0.1,
metrics=[Accuracy()]):
error_history = []
metrics_history = []
# split the data into train set and validation set
data = split_data(x, y, validation_split, test_set_name="validation")
data_valid = {
"x": data["validation"]["x"],
"y": data["validation"]["y"]
}
num_train_examples = len(data["train"]["y"])
for epoch in range(epochs):
i = 0
# Shuffle train data
indices = np.arange(num_train_examples)
np.random.shuffle(indices)
data_train = {
"x": data["train"]["x"][indices],
"y": data["train"]["y"][indices]
}
while i < num_train_examples:
data_batch = {
"x": data_train["x"][i: i + batch_size],
"y": data_train["y"][i: i + batch_size]
}
i += batch_size
# forward propagation
tensors = network.forward(data_batch)
# error
error_history.append(tensors[-1][0].elements)
# backward propagation
deltas = network.backprop(tensors)
deltas.reverse()
# parameter update
for layer, tensorlist in zip(network.layers, deltas):
if tensorlist is not None:
for tensor in tensorlist:
tensor.elements = alpha * tensor.elements
layer.update_parameter(tensorlist)
# Evaluate with validation set
epoch_metrics = [None] * len(metrics)
prediction = network.predict(data_valid).elements
for i in range(len(metrics)):
epoch_metrics[i] = metrics[i].score(prediction, data_valid["y"])
metrics_history.append(epoch_metrics)
# Print progress update
sys.stdout.write(
'\r' + "Epoch %i / " % (epoch + 1) + str(epochs) + " Metrics: " + str(epoch_metrics))
sys.stdout.flush()
return error_history, metrics_history
import mnist_loader
from NeuralNetwork import Network
train_data, validate_data, test_data = mnist_loader.load_data_wrapper()
net = Network(configurationFileName='NetConfig')
print('Done {0} from {1}'.format(net.check(test_data), len(test_data)))
from NeuralNetwork import Network
nn = Network('lastmodel')
nn.train(10000)
from NeuralNetwork import Network zero = [0, 1, 1, 0, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0] one = [0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0] two = [0, 1, 1, 0, 1, 0, 0, 1, 0, 0, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1] three = [1, 1, 1, 1, 0, 0, 0, 1, 0, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1] exampleZero = [zero, [0, 0]] exampleOne = [one, [0, 1]] exampleTwo = [two, [1, 0]] exampleThree = [three, [1, 1]] examples = [exampleZero, exampleOne, exampleTwo, exampleThree] network = Network() network.addLayer(10, 20) network.addLayer(2, None) network.train(examples) print network.process(zero) print network.process(one) print network.process(two) print network.process(three)
NUM_TRAIN_BATCHES = 1500
LEN_TEST_TEXT = 2000 # Number of test characters of text to generate after training the network
ckpt_filename = 'model'
midi_filename = 'midiOut.txt'
## Initialize the network
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.InteractiveSession(config=config)
net = Network(in_size=in_size,
lstm_size=lstm_size,
num_layers=num_layers,
out_size=out_size,
session=sess,
learning_rate=0.003,
name="char_rnn_network")
sess.run(tf.global_variables_initializer())
saver = tf.train.Saver(tf.global_variables())
## 1) TRAIN THE NETWORK
if ckpt_file == "":
last_time = time.time()
batch = np.zeros((batch_size, time_steps, in_size))
batch_y = np.zeros((batch_size, time_steps, in_size))
def addWeakClassifier(self, layers): """ Adds a neural network in the Adaboost """ self.weakClassifiers.append(Network.NeuralNetwork(layers))
def main():
topology = []
topology.append(2)
topology.append(3)
topology.append(3)
topology.append(1)
net = Network(topology)
neuron.eta = 0.09
neuron.alpha = 0.015
inputs = [[0, 0], [0, 1], [1, 0], [1, 1]]
outputs = [[0], [1], [1], [0]]
while True:
err = 0
for i in range(len(inputs)):
net.setInput(inputs[i])
net.feedForward()
net.backPropagate(outputs[i])
err = err + net.getError(outputs[i])
print("Error: ", err)
if err < 1.5:
break
while True:
a = int(input("First input: "))
b = int(input("Second input: "))
net.setInput([a, b])
net.feedForward()
print(net.getThResults())
def neural_network_factory(filename: str) -> Tuple:
file = open(filename)
config = json.load(file)
""" Dataset """
dataset_name: str = config["dataset"]["name"]
case_fraction: float = config["dataset"]["case_fraction"]
validation_fraction: float = config["dataset"]["validation_fraction"]
test_fraction: float = config["dataset"]["test_fraction"]
cases = dataset_factory(dataset_name)
""" Arcitechture """
input_size: int = config["arcitechture"]["input_size"]
layer_specification: List = config["arcitechture"]["layer_specification"]
weight_range = (config["arcitechture"]["weight_range"]["from"],
config["arcitechture"]["weight_range"]["to"])
bias_range = (config["arcitechture"]["bias_range"]["from"],
config["arcitechture"]["bias_range"]["to"])
activation_functions = list(
map(lambda x: activation_factory(x),
config["arcitechture"]["activation_functions"]))
""" Training """
optimizer: Callable = optimizer_factory(config["training"]["optimizer"])
epochs = config["training"]["epochs"]
minibatch_size = config["training"]["minibatch_size"]
loss_function = loss_factory(config["training"]["loss_function"])
learning_rate = config["training"]["learning_rate"]
test_frequency = config["training"]["test_frequency"]
""" Visualization """
visualization_on = config["visualization"]["on"]
display_weights = config["visualization"]["display_weights"]
display_biases = config["visualization"]["display_biases"]
display_layers = config["visualization"]["display_layers"]
map_size = config["visualization"]["map_batch_size"]
dendrogram_layers = config["visualization"]["dendrogram_layers"]
# Create network
network = Network(input_size=input_size,
dimensions=layer_specification,
activations=activation_functions,
loss_function=loss_function,
optimizer=optimizer,
learning_rate=learning_rate,
minibatch_size=minibatch_size,
epochs=epochs,
weight_range=weight_range,
bias_range=bias_range,
test_frequency=test_frequency,
display_weights=display_weights,
display_layers=display_layers,
display_biases=display_biases,
visualization_on=visualization_on,
dendrogram_layers=dendrogram_layers,
map_size=map_size)
# Create casemanager
cases = CaseManager(
cases=cases,
validation_fraction=validation_fraction,
test_fraction=test_fraction,
case_fraction=case_fraction,
)
return network, cases
f.close()
print("Loading test label data")
f = open("../test/t10k-labels.idx1-ubyte", "rb")
Y_test, nlabels_test = loadLabelFile(f)
f.close()
X = np.array(X)
y = np.array(y)
nn = Network(layers = [ FullConnectedLayer(
nNodes = 784,
keep_prob = 0.5),
SoftmaxLayer(10)],
mini_batch_size = 1024,
num_iterations = 50,
learning_rate = 0.01,
momentum_rate = 0.9,
rmsprop_rate = 0.999,
l2_regularization = 0.7,
verbose = False)
nn.fit(X, y)
X_test = np.array(X_test)
Y_test = np.array(Y_test)
test_error = nn.validate(X_test, Y_test) * 1 / nImages_test
print("Test error : ", 1 - test_error)
print("Test acc. : %.02f"%(test_error * 100))
print("--------------------------------")
THIRD_LAYER_INPUT = 10
FOURTH_LAYER_INPUT = 15
LABELS_NUM = 2
NUM_OF_SAMPLES = 100
x = np.random.random((NUM_OF_SAMPLES, FIRST_LAYER_INPUT)).T
#x = np.array([[1, 0], [2, 1], [4, 1], [5, 2], [1, 4], [2, 3], [0.9, 1],
# [2, 5], [1, 5], [6, 3]]).T
#NUM_OF_SAMPLES = 100
y_arr = []
for i in range(NUM_OF_SAMPLES):
if i % 2 == 0:
y_arr.append(np.array([1, 0]))
else:
y_arr.append(np.array([0, 1]))
y = np.array(y_arr).T
n = Network()
n.add_layer(Layer(FIRST_LAYER_INPUT, SECOND_LAYER_INPUT, TANH_ACTIVATION))
n.add_layer(Layer(SECOND_LAYER_INPUT, THIRD_LAYER_INPUT, TANH_ACTIVATION))
n.add_layer(Layer(THIRD_LAYER_INPUT, FOURTH_LAYER_INPUT, TANH_ACTIVATION))
n.add_layer(Layer(FOURTH_LAYER_INPUT, LABELS_NUM, None, softmax_layer=True))
curr_batch = np.random.permutation(range(NUM_OF_SAMPLES))
batch_x = np.array([x.T[ind] for ind in curr_batch]).T
batch_y = np.array([y.T[ind] for ind in curr_batch]).T
l = n.get_layer(0)
# Softmax test functions
l_sm = Layer(FIRST_LAYER_INPUT, LABELS_NUM, None, softmax_layer=True)
W_EXAMPLE = np.array([[2, 0], [0, 1]]).T
B_EXAMPLE = (np.array([1, 1]))
]
plt.scatter(coord_y_pos, coord_x_pos, alpha=0.2)
plt.scatter(coord_y_neg, coord_x_neg, alpha=0.2)
plt.savefig('spirals/plot_{}.png'.format(ind))
mat = scipy.io.loadmat('SwissRollData.mat')
labels = mat['Ct']
training = mat['Yt']
labels_validation = mat['Cv']
samples_validation = mat['Yv']
x = training
y = labels
x_validation = samples_validation
y_validation = labels_validation
n = Network()
n.add_layer(Layer(2, 10, TANH_ACTIVATION))
#n.add_layer(Layer(10, 10, TANH_ACTIVATION))
#n.add_layer(Layer(10, 10, TANH_ACTIVATION))
n.add_layer(Layer(10, 10, TANH_ACTIVATION))
n.add_layer(Layer(10, 2, None, softmax_layer=True))
epochs = [20]
batch_sizes = [100]
learning_rates = [0.5]
for epoch, batch_size, learning_rate in product(epochs, batch_sizes,
learning_rates):
print('epochs {} batch_size {} learning_rate {}'.format(
epoch, batch_size, learning_rate))
n, obj = stochastic_gradient_descent(n,
x,
y,
import numpy as np
import random
import math
from NeuralNetwork import Network
n_inputs, hidden_layers_length, n_neurons, n_outputs = 2, 5, 4, 2
network = Network(n_inputs, hidden_layers_length, n_neurons, n_outputs)
print("\ninputlayer: \n" + str(network.inputLayer.weights))
for i in range(hidden_layers_length):
print("\nhiddenLayers" + str(i+1) +":\n" + str(network.hiddenLayers[i].weights))
print("\noutputLayer: \n" + str(network.outputLayer.weights))
def main():
result_list = [[None], [None]]
topology = []
topology.append(2)
topology.append(3)
topology.append(3)
topology.append(1)
net1 = Network(topology)
net2 = Network(topology)
neuron.eta = 0.09
neuron.alpha = 0.015
inputs1 = [[0, 0], [0, 1]]
outputs1 = [[0], [1]]
inputs2 = [[1, 0], [1, 1]]
outputs2 = [[1], [0]]
threads = []
nets = [net1, net2]
outputs = [outputs1, outputs2]
inputs = [inputs1, inputs2]
for i in range(2):
t = threading.Thread(target=TrainNet,
args=(inputs[i], [nets[i]], outputs[i], i,
result_list))
threads.append(t)
t.start()
for i in range(2):
threads[i].join()
net1 = result_list[0]
net2 = result_list[1]
while True:
a = int(input("First input: "))
b = int(input("Second input: "))
if (a > 0):
net2.setInput([a, b])
net2.feedForward()
print(net1.getThResults())
else:
net1.setInput([a, b])
net1.feedForward()
print(net1.getThResults())
import mnist_loader
from NeuralNetwork import Network
training_data, validation_data, test_data = mnist_loader.load_data_wrapper()
net = Network([784, 30, 10])
net.train(training_data, 30, 10, 3.0, test_data=test_data)
net.save_configuration()
print("Done")
data_part = 5 # only one fifth of the whole dataset to speed up training
for i in range(len(labels_full) // batch_size // data_part):
images.append(images_full[i * batch_size:(i + 1) * batch_size])
labels.append(labels_full[i * batch_size:(i + 1) * batch_size])
y = []
for batch in labels:
y.append([])
for label in batch:
y[-1].append([1.0 if i == label else 0.0 for i in range(10)])
y = np.array(y)
network = Network(SIZES)
network.epoch(images, y, batch_training_size) # from dynamic parameters
'''
for i, el in enumerate(l3):
print(labels[0][i], "=", np.argmax(el), " predictions: ", el)
'''
testing_images, testing_labels = mndata.load_testing()
correct = 0.0
for i, (image, label) in enumerate(zip(testing_images, testing_labels)):
prediction = network.run(image)
if label == prediction:
correct += 1.0
correct_rate = correct / (i + 1.0)
print("{} = {} (correct {}%)".format(label, prediction,