def test_gradient_descent_test():
model = TestModel()
optimiser = v.GradientDescentOptimiser(0.1, batch_size=1)
error = v.SumSquared()
xs = np.zeros((10, 2))
ys = np.zeros((10, 1))
optimiser.train(model, xs, ys, error)
assert model.called_predict == 10
assert model.called_backward == 10
assert model.called_cache_reset == 10
def __init__(self, error_function=v.SumSquared()):
super(ErrorCostCategorical, self).__init__(error_function)
self.outputs = []
def __init__(self, error_function=v.SumSquared()):
self.errors = []
self.hits = []
assert callable(error_function)
self.error_function = error_function
def test_sum_squared_single():
xs = np.array(10)
ys = np.array(12)
assert v.SumSquared()(xs, ys) == 2
def test_sum_squared_prime():
xs = np.zeros((4))
ys = np.repeat(2, 4)
expected = np.repeat(-2, 4)
out = v.SumSquared().prime(xs, ys)
assert np.allclose(out, expected)