def test_gradient_descent_correctness_nonsep_data(non_linear_separable_data):
x_test, y_test, wrong_idx = non_linear_separable_data
w_ans = GradientDescent(lr=0.01).execute(x_test, y_test, epochs=10)
y_pred = sign(add_cons(x_test) @ w_ans)
assert len(np.where(y_test != y_pred)[0].tolist()) == 1
assert wrong_idx[0] == np.where(y_test != y_pred)[0][0]
def linear_separable_data():
"""generate linear Seperable data"""
np.random.seed(123)
x_test = np.random.rand(30, 2)
w_test = np.random.rand(2)
b_test = np.random.rand(1)
y_test = sign(x_test @ w_test - b_test)
return x_test, y_test
def test_pocket_correctness_nonsep_data(non_linear_separable_data):
x_test, y_test, wrong_idx = non_linear_separable_data
w_ans = Pocket().execute(x_test, y_test, updates=100)
y_pred = sign(add_cons(x_test) @ w_ans)
assert len(np.where(y_test != y_pred)[0].tolist()) == 1
assert wrong_idx[0] == np.where(y_test != y_pred)[0][0]
def test_adaboost_stump():
x_test, y_test = make_moons(20, random_state=123)
y_test = sign(y_test)
model = AdaBoostStump(n_estimators=10)
model.fit(x_test, y_test)
y_pred = model.predict(x_test)
accuracy = model.evaluate(x_test, y_test)[0]
# test predict()
assert np.array_equal(y_pred, y_test)
# test evaluate()
assert accuracy == 1.0
def non_linear_separable_data():
"""generate non-separable data"""
np.random.seed(123)
x_test = np.random.rand(40, 2)
w_test = np.random.rand(2)
b_test = np.random.rand(1)
margin = np.abs((x_test @ w_test - b_test) /
np.sqrt(np.sum(w_test**2) + b_test**2)) > 0.03
x_test = x_test[margin, :]
y_test = sign(x_test @ w_test - b_test)
wrong_idx = np.random.randint(0, len(x_test), 1) # One mistake
y_test[wrong_idx] = -1 * y_test[wrong_idx]
return x_test, y_test, wrong_idx
def test_bagging_pocket():
x_test, y_test = make_moons(20, noise=0.05, random_state=4, shuffle=False)
x_test, y_test = np.delete(x_test, [0, 10],
axis=0), np.delete(y_test, [0, 10])
y_test = sign(y_test)
model = BaggingPerceptron(n_estimators=15, optimizer=Pocket(updates=2000))
model.fit(x_test, y_test)
y_pred = model.predict(x_test)
# test predict()
assert np.array_equal(y_pred, y_test)
# test evaluate()
assert model.evaluate(x_test, y_test)[0] == 1.0 # Accuracy == 1.0
def test_gradient_descent_correctness_linsep_data(linear_separable_data):
x_test, y_test = linear_separable_data
w_ans = GradientDescent(lr=0.01).execute(x_test, y_test, epochs=np.Inf)
y_pred = sign(add_cons(x_test) @ w_ans)
assert np.array_equal(y_pred, y_test)
def test_pocket_correctness_linsep_data(linear_separable_data):
x_test, y_test = linear_separable_data
w_ans = Pocket().execute(x_test, y_test, updates=np.Inf)
y_pred = sign(add_cons(x_test) @ w_ans)
assert np.array_equal(y_pred, y_test)
def test_linear_seperable_correctness(linear_separable_data):
x_test, y_test = linear_separable_data
w_ans = LinearSeparable().execute(x_test, y_test)
y_pred = sign(add_cons(x_test) @ w_ans)
assert np.array_equal(y_pred, y_test)
def test_data():
x_test, y_test = make_classification(random_state=1)
y_test = sign(y_test, zero=-1)
return x_test, y_test