def init_test():
net = network.bp_classifer([3, 2, 3, 1], 0.1, 0.01)
assert net.num_layers == 4
assert net.num_inputs == 3
assert net.num_outputs == 1
weight_shapes = [w.shape for w in net.weights]
assert weight_shapes == [(2, 3), (3, 2), (1, 3)]
bias_shapes = [b.shape for b in net.biases]
assert bias_shapes == [(2, 1), (3, 1), (1, 1)]
print("init_test passed!\n")
pass
def backprop_test():
net = network.bp_classifer([3, 2, 3, 3], 0.1, 0.01)
x = np.array([1, 3, 0]).reshape(1, 3)
#y = np.array([2]).reshape(1, 1)
y = net.one_hot(2)
dw, db = net.backpropogate(x, y)
weight_shapes = [w.shape for w in dw]
assert weight_shapes == [(2, 3), (3, 2), (3, 3)]
bias_shapes = [b.shape for b in db]
assert bias_shapes == [(2, 1), (3, 1), (3, 1)]
print("backprop_test passed!\n")
pass
def update_test():
net = network.bp_classifer([3, 2, 3], 0.1, 0.01)
x = np.array([1, 3, 0]).reshape(1, 3)
#y = np.array([0]).reshape(1, 1)
y = net.one_hot(0)
w = net.weights.copy()
b = net.biases.copy()
net.update_SGD(x, y)
for i in range(len(w)):
assert net.weights[i].shape == w[i].shape
assert net.biases[i].shape == b[i].shape
print("update_test passed!\n")
pass
def validation_test():
net = network.bp_classifer([3, 2, 3], 0.1, 0.01)
train_X = np.random.randint(10, size=(10, 3))
train_D = np.random.randint(3, size=(10, 1))
val_X = np.random.randint(10, size=(5, 3))
val_D = np.random.randint(3, size=(5, 1))
n_epochs = 5
acc = net.train(train_X, train_D, val_X, val_D, num_epochs=n_epochs)
for epoch in range(n_epochs):
print(f'epoch {epoch + 1}: {acc[epoch]}')
print("validation_test passed!\n")
pass
def train_test():
net = network.bp_classifer([3, 2, 3], 0.1, 0.01)
print("BEFORE TRAINING\n")
for w, b in zip(net.weights, net.biases):
print(w)
print(b)
train_X = np.random.randint(10, size=(10, 3))
train_D = np.random.randint(3, size=(10, 1))
net.train(train_X, train_D)
print("AFTER TRAINING\n")
for w, b in zip(net.weights, net.biases):
print(w)
print(b)
print("train_test passed!\n")
pass