def test_sigmoid(self):
X = array([[1, 1, 1], [1, 1, 1]])
theta = array([[1, 1, 1]]).transpose()
z = X.dot(theta)
hypothesis = sigmoid(z)
expected = array([[0.95257], [0.95257]])
numpy.testing.assert_almost_equal(expected, hypothesis, 5)
def activation_cost_function(X, Y, weights):
W, b = utils.vector_to_weights(weights, n, 1) # expand weights
z = core.calculate_z(X, W, b)
a = core.sigmoid(z)
j, dW, db = core.logistic_cost_function(X, a, Y)
"""
Flatten gradients
Transposing them because I changed the minimization functions to work with NNs
"""
gradients = utils.weights_to_vector(dW.T, db.T)
return j, gradients
def forward_propagate(self, X, weights):
"""
:param X: Input features (n x m matrix)
:param weights: A list of (W, b) tuples, one for each layer
:return: List of activations, list of zs
"""
activations = [X] # The inputs are the 0th activation
zs = []
for i in range(
1, len(self.layers)
): # Skip the 0th layer, since these are just the inputs
W = weights[i - 1]["W"]
b = weights[i - 1]["b"]
z = core.calculate_z(activations[i - 1], W, b)
a = core.sigmoid(z)
zs.append(z)
activations.append(a)
return activations, zs
def activation_cost_function(X, Y, W, b):
z = core.calculate_z(X, W, b)
a = core.sigmoid(z)
j, dW, db = core.logistic_cost_function(X, a, Y)
return j, dW.T, db.T # Transposing the gradients because I changed the minimization functions to work with NNs
def activate(X, W, b):
z = core.calculate_z(X, W, b)
a = core.sigmoid(z)
return a
def test_cost(self):
z = calculate_z(self.X, self.W, self.b)
a = sigmoid(z)
j, dW, db = logistic_cost_function(self.X, a, self.Y)
expected = 0.048587
numpy.testing.assert_almost_equal(j, expected, 5)