def test_np_shallow_neural_classifier_gradients(hidden_activation, d_hidden_activation):
model = ShallowNeuralClassifier(
max_iter=10,
hidden_activation=hidden_activation,
d_hidden_activation=d_hidden_activation)
# A tiny dataset so that we can run `fit` and set all the model
# parameters:
X = utils.randmatrix(5, 2)
y = np.random.choice((0,1), 5)
model.fit(X, y)
# Use the first example for the check:
ex = X[0]
label = model._onehot_encode([y[0]])[0]
# Forward and backward to get the gradients:
hidden, pred = model.forward_propagation(ex)
d_W_hy, d_b_hy, d_W_xh, d_b_xh = model.backward_propagation(
hidden, pred, ex, label)
# Model parameters to check:
param_pairs = (
('W_hy', d_W_hy),
('b_hy', d_b_hy),
('W_xh', d_W_xh),
('b_xh', d_b_xh)
)
gradient_check(param_pairs, model, ex, label)
def test_np_model(XOR):
"""Just makes sure that this code will run; it doesn't check that
it is creating good models.
"""
X, y = XOR
model = ShallowNeuralClassifier(hidden_dim=4,
hidden_activation=np.tanh,
d_hidden_activation=utils.d_tanh,
eta=0.05,
tol=1.5e-8,
display_progress=True,
max_iter=100)
model.fit(X, y)
model.predict(X)
def test_np_parameter_setting(param, value):
mod = ShallowNeuralClassifier()
mod.set_params(**{param: value})
assert getattr(mod, param) == value