Function test_BatchNorm2D

numpy_ml/tests/test_nn.py:1011–1064

(N=15)

def test_BatchNorm2D(N=15):
    from numpy_ml.neural_nets.layers import BatchNorm2D

    N = np.inf if N is None else N

    np.random.seed(12345)

    i = 1
    while i < N + 1:
        n_ex = np.random.randint(2, 10)
        in_rows = np.random.randint(1, 10)
        in_cols = np.random.randint(1, 10)
        n_in = np.random.randint(1, 3)

        # initialize BatchNorm2D layer
        X = random_tensor((n_ex, in_rows, in_cols, n_in), standardize=True)
        L1 = BatchNorm2D()

        # forward prop
        y_pred = L1.forward(X)

        # standard sum loss
        dLdy = np.ones_like(X)
        dLdX = L1.backward(dLdy)

        # get gold standard gradients
        gold_mod = TorchBatchNormLayer(
            n_in, L1.parameters, mode="2D", epsilon=L1.epsilon, momentum=L1.momentum
        )
        golds = gold_mod.extract_grads(X, Y_true=None)

        params = [
            (L1.X[0], "X"),
            (L1.hyperparameters["momentum"], "momentum"),
            (L1.hyperparameters["epsilon"], "epsilon"),
            (L1.parameters["scaler"].T, "scaler"),
            (L1.parameters["intercept"], "intercept"),
            (L1.parameters["running_mean"], "running_mean"),
            #  (L1.parameters["running_var"], "running_var"),
            (y_pred, "y"),
            (L1.gradients["scaler"], "dLdScaler"),
            (L1.gradients["intercept"], "dLdIntercept"),
            (dLdX, "dLdX"),
        ]

        print("Trial {}".format(i))
        for ix, (mine, label) in enumerate(params):
            assert_almost_equal(
                mine, golds[label], err_msg=err_fmt(params, golds, ix), decimal=3
            )

            print("\tPASSED {}".format(label))

        i += 1


def test_RNNCell(N=15):