def build_model():
model = NeuralNetwork(loss=CrossEntropy(),
n_iterations=800,
learning_rate=0.13)
model.add(Layer(4, 10, activation=Relu()))
model.add(Layer(10, 3, activation=Softmax()))
return model
def test_softmax_calculate_gradient(self):
# Given
pre_activation = np.array([[1, 2, 3, 6], [2, 4, 5, 6], [3, 8, 7, 6]])
target = np.array([[1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0]])
softmax = Softmax()
# When
activation = softmax.apply_activation(pre_activation)
grad = softmax.calculate_gradient(activation, target)
def __init__(self):
self.d1_layer = Dense(784, 100)
self.a1_layer = ReLu()
self.drop1_layer = Dropout(0.5)
self.d2_layer = Dense(100, 50)
self.a2_layer = ReLu()
self.drop2_layer = Dropout(0.25)
self.d3_layer = Dense(50, 10)
self.a3_layer = Softmax()
def test_softmax_apply_activation(self):
# Given
pre_activation = np.array([[1, 2, 3, 6], [2, 4, 5, 6], [3, 8, 7, 6]])
softmax = Softmax()
# When
activation = softmax.apply_activation(pre_activation)
# Then
sum_of_columns = 4.0
self.assertTrue(
np.isclose(sum_of_columns, np.sum(np.sum(activation, axis=0)),
1e-3))
def sequential(self,
network=[1, 1, 1],
activation=[ReLU(), Softmax()],
loss=Cross_entropy(),
regu=Default(),
weight_type='default'):
self.net = network
self.activation1 = activation[0]
self.activation2 = activation[1]
self.loss = loss
self.regu = regu
self.init_weight(weight_type)
val = df.drop(train.index)
yr = train.iloc[:, 0].to_numpy()
X_train, y_train = train.iloc[:, 1:].to_numpy() / 255.0, onehotcode(yr)
X_train = X_train.reshape((X_train.shape[0], X_train.shape[1], 1))
y_train = y_train.reshape((y_train.shape[0], y_train.shape[1], 1))
print(X_train.shape, y_train.shape)
X_val, y_val = val.iloc[:, 1:].to_numpy() / 255.0, val.iloc[:, 0].to_numpy()
X_val = X_val.reshape((X_val.shape[0], X_val.shape[1], 1))
print(X_val.shape, y_val.shape)
nn = NN()
nn.sequential(network=[784, 128, 10],
activation=[Tanh(), Softmax()],
loss=Cross_entropy(),
regu=Ridge(n=X_train.shape[0], lmda=5),
weight_type='glorot_normal')
nn.load_model('tanL2128')
#nn.fit(X_train,y_train,X_val,y_val,32,2)
#nn.save_model('tanL2128')
nn.weight_heatmap()
df2 = pd.read_csv('mnist_test.csv')
X_test, y_test = df2.iloc[:, 1:].to_numpy() / 255.0, df2.iloc[:, 0].to_numpy()
X_test = X_test.reshape((X_test.shape[0], 28, 28))
print(X_test.shape, y_test.shape)
nn.annote_test(X_test[:100], 10, 10)
def __init__(self):
self.layers = []
self.activation = Softmax()
self.init()
