def test_nbr_neuron(list_test):
color_list=['r','g','b','k','m','c','y']
color_list *= 3
k=0
for i in list_test :
my_layer1 = Layer.Linear(6,i)
my_layer2 = ActivationFunctions.Tanh()
my_layer5 = Layer.Linear(i,i)
my_layer6 = ActivationFunctions.Tanh()
my_layer3 = Layer.Linear(i,1)
my_layer4 = ActivationFunctions.Sigmoid()
my_NN = Neural_network.NeuralNet([my_layer1, my_layer2, my_layer5, my_layer6, my_layer3, my_layer4])
chi2_list, error_list = User.train(my_NN, data_train_input, data_train_target, num_epochs = num_epoch_max, optimizer = Optimizer.SGD(lr = my_lr), batch_size=my_batch_size)
data_test_prediction = User.prediction(my_NN,data_test_input)
error_final = Error_round.error_round(data_test_prediction, data_test_target)
plt.plot(range(num_epoch_max), error_list, label= str(i), c=color_list[k])
plt.plot([num_epoch_max],[error_final], marker='o', c=color_list[k])
plt.xlabel('Epoch')
plt.ylabel('Training round error')
k+=1
plt.legend(title='Neurons')
plt.title('Optimisation of the number of neurons')
plt.show()
def test_nbr_layer(list_test, n_neuron):
color_list=['r','g','b','k','m','c','y']
color_list *= 3
k=0
my_layerini1 = Layer.Linear(6,n_neuron)
my_layerini2 = ActivationFunctions.Tanh()
my_layerfini1 = Layer.Linear(n_neuron,1)
my_layerfini2 = ActivationFunctions.Sigmoid()
for i in list_test :
layers_new = [my_layerini1, my_layerini2]
for j in range(i) :
layers_new += [Layer.Linear(n_neuron,n_neuron),ActivationFunctions.Tanh()]
layers_new += [my_layerfini1, my_layerfini2]
my_NN = Neural_network.NeuralNet(layers_new)
chi2_list, error_list = User.train(my_NN, data_train_input, data_train_target, num_epochs = num_epoch_max,optimizer = Optimizer.SGD(lr = my_lr), batch_size=my_batch_size)
data_test_prediction = User.prediction(my_NN,data_test_input)
error_final = Error_round.error_round(data_test_prediction, data_test_target)
plt.plot(range(num_epoch_max), error_list, label= str(i),c=color_list[k])
plt.plot([num_epoch_max],[error_final], marker='o', c=color_list[k])
plt.xlabel('Epoch')
plt.ylabel('Training round error')
k+=1
plt.legend(title='Hidden layers')
plt.title('Optimisation of the number of hidden layers')
plt.show()
def test_NeuralNet() :
my_layer1 = Layer.Linear(3,2)
my_layer2 = ActivationFunctions.Tanh()
my_NN = Neural_network.NeuralNet([my_layer1,my_layer2])
input = np.array([[1,2,3],[4,5,6]])
grad = np.array([[0.5,0.2],[0.1,0.3]])
print('forward', my_NN.forward(input))
print('backward', my_NN.backward(grad))
'''OK'''
def _create_activations(self):
funs = []
for n in self.n_neurons_per_layer:
F = []
for i in range(n):
f = af.Tanh()
if i % 2 == 0:
f = af.Sigmoid()
F.append(f)
F = np.array(F)
funs.append(F)
self.activation_funs = np.array(funs)
def test_train_prediction() :
my_layer1 = Layer.Linear(3,2)
my_layer2 = ActivationFunctions.Tanh()
my_NN = Neural_network.NeuralNet([my_layer1,my_layer2])
input = np.array([[1,2,3],[4,5,6]])
target = np.array([[0.5,0.2],[0.1,0.3]])
User.train(my_NN, input, target, batch_size = 1)
#By careful, we must have size_training = number of rows in our data
input_predict = np.array([[1,1,4],[0.5,2,4]])
print(User.prediction(my_NN,input_predict))
''' OK '''
def gate_transform(self, affine_gates):
"""
apply gate Non-Linearity
"""
h = self.hl_size
affine_gates[0:h, :] = activation_functions.Sigmoid().transform(
self.i(affine_gates))
affine_gates[h:2 * h, :] = activation_functions.Sigmoid().transform(
self.f(affine_gates))
affine_gates[2 * h:3 *
h, :] = activation_functions.Sigmoid().transform(
self.o(affine_gates))
affine_gates[3 * h:, :] = activation_functions.Tanh().transform(
self.g(affine_gates))
transformed_gates = affine_gates
return transformed_gates
def test_XOR() :
my_layer1 = Layer.Linear(2,3)
my_layer2 = ActivationFunctions.Tanh()
my_layer3 = Layer.Linear(3,1)
my_layer4 = ActivationFunctions.Sigmoid()
#my_layer3 = lib2.Arondi()
my_NN = Neural_network.NeuralNet([my_layer1,my_layer2,my_layer3,my_layer4])
input =np.array([[0, 0], [1, 0], [0, 1], [1, 1]])
target = np.array([[0], [1], [1], [0]])
User.train(my_NN, input, target, batch_size = 1,,num_epochs= 1000)
# By careful, we must have size_training = number of rows in our data
input_predict = np.array([[0, 0], [1, 0], [0, 1], [1, 1]])
print(User.prediction(my_NN,input_predict))
''' OK '''
import fcnetwork
import cost_functions
import activation_functions
import learning_methods
import regularization_functions as reg
import mnist as mnist_loader
import numpy as np
import sys
import batch_norm as bn
network_topology = [784, 30, 10]
#activation_functions.PReLU(learning_methods.Momentum(.01,.7),20),
#activation_functions.PReLU(learning_methods.Momentum(.01,.7),30)
network_activations = [activation_functions.Tanh(), \
activation_functions.Softmax()]
def reduceL(t):
for index, v in enumerate(t):
x, y = v
t[index] = x, np.argmax(y)
return t
eta = 3
lmbda = 0
epochs = 64
mini_batch = 10
net=fcnetwork.FCNetwork(network_topology, \
network_activations, \
cost_functions.CrossEntropy(),\
def train_simultaneousNN(
inputs_train: Tensor,
targets_train: Tensor,
loss: Loss.Loss = Loss.MeanSquareError(),
optimizer: OptimizerClass.Optimizer = OptimizerClass.SGD(),
num_epochs: int = 5000,
batch_size: int = 32) -> tuple:
size_training = inputs_train.shape[0]
Result_chi2 = [[], [], [], [], [], [], [], [], []]
list_epoch = np.array(range(10, 50, 5)) / 100 * num_epochs
'''initialisation des 9 NN''' #verifier question seed()
list_net = []
for i in range(9):
layers = []
layers.append(Layer.Linear(6, 4))
layers.append(ActivationFunctions.Tanh())
layers.append(Layer.Linear(4, 2))
layers.append(ActivationFunctions.Tanh())
layers.append(Layer.Linear(2, 1))
layers.append(ActivationFunctions.Sigmoid())
list_net.append(Neural_network.NeuralNet(layers))
destroyed_NN = []
nbr_batch = size_training // batch_size
''' training des 9 NN'''
for epoch in range(num_epochs):
for k in range(9):
if k not in destroyed_NN:
Chi2_train = 0
for i in range(0, size_training, batch_size):
# 1) feed forward
y_actual = list_net[k].forward(inputs_train[i:i +
batch_size])
# 2) compute the loss and the gradients
Chi2_train += loss.loss(targets_train[i:i + batch_size],
y_actual)
grad_ini = loss.grad(targets_train[i:i + batch_size],
y_actual)
# 3)feed backwards
grad_fini = list_net[k].backward(grad_ini)
# 4) update the net
optimizer.step(list_net[k], n_epoch=epoch)
Chi2_train = Chi2_train / nbr_batch
Result_chi2[k].append(Chi2_train)
'''Supression du NN le moins efficace '''
if epoch in list_epoch:
Comparaison = [[], []]
for k in range(9):
if k not in destroyed_NN:
ErrorSlope = np.polyfit(np.array(range(epoch - 49, epoch)),
Result_chi2[k][-50:-1], 1)[0]
MixedError = Result_chi2[k][-1] * (1 -
np.arctan(ErrorSlope) /
(np.pi / 2))
Comparaison[0].append(k)
Comparaison[1].append(MixedError)
k = Comparaison[0][Comparaison[1].index(max(Comparaison[1]))]
destroyed_NN.append(k)
if epoch % 100 == 0:
print('epoch : ' + str(epoch) + "/" + str(num_epochs) + "\r",
end="")
for k in range(9):
if k not in destroyed_NN:
my_NN = list_net[k]
return my_NN, Result_chi2
np.random.seed(1)
'''size of the training set and the testing set '''
train_size = 3000
test_size = 1500
'''Maximal number of epochs '''
my_num_epochs = 500
'''size of the batch'''
my_batch_size = 100
''' learning rate'''
my_lr = 0.001
my_initial_lr = 0.1
my_decay_coeff = 1 / 200
'''Construction of the neural network '''
my_layer1 = Layer.Linear(6, 5)
my_layer2 = ActivationFunctions.Tanh()
my_layer3 = Layer.Linear(5, 4)
my_layer4 = ActivationFunctions.Tanh()
my_layer5 = Layer.Linear(4, 3)
my_layer6 = ActivationFunctions.Tanh()
my_layer7 = Layer.Linear(3, 2)
my_layer8 = ActivationFunctions.Tanh()
my_layer9 = Layer.Linear(2, 1)
my_layer10 = ActivationFunctions.Sigmoid()
my_NN = Neural_network.NeuralNet([
my_layer1, my_layer2, my_layer3, my_layer4, my_layer5, my_layer6,
my_layer7, my_layer8, my_layer9, my_layer10
])
## Importation of the training and testing data
os.chdir(path_ini[:-4])
weights = []
functions = []
# Loop for creating random weights between in range (-2, 2), and activation
# functions alternating between sigmoid and tanh
for i in range(n_layers):
n = n_neurons_per_layer[i]
ins = inputs_per_layer[i]
layer_w = []
layer_f = []
for j in range(n):
layer_w.append(np.random.uniform(-2, 2, ins))
if i % 2 == 0:
layer_f.append(af.Sigmoid())
else:
layer_f.append(af.Tanh())
functions.append(np.array(layer_f))
weights.append(np.array(layer_w))
weights = np.array(weights)
functions = np.array(functions)
#%%
network = nn.NeuralNetwork(n_layers, n_neurons_per_layer, n_in, n_out)
# Set network parameters
network.set_iterations(n_iter)
network.set_learning_rate(0.01)
network.set_weights(weights)