Function test_pad1D

numpy_ml/tests/test_nn.py:2174–2270

(N=15)

def test_pad1D(N=15):
    from numpy_ml.neural_nets.layers import Conv1D
    from .nn_torch_models import TorchCausalConv1d, torchify

    np.random.seed(12345)

    N = np.inf if N is None else N

    i = 1
    while i < N + 1:
        p = np.random.choice(["same", "causal"])
        n_ex = np.random.randint(1, 10)
        l_in = np.random.randint(1, 10)
        n_in, n_out = np.random.randint(1, 3), np.random.randint(1, 3)
        f_width = min(l_in, np.random.randint(1, 5))
        s = np.random.randint(1, 3)
        d = np.random.randint(0, 5)

        X = random_tensor((n_ex, l_in, n_in), standardize=True)
        X_pad, _ = pad1D(X, p, kernel_width=f_width, stride=s, dilation=d)

        # initialize Conv2D layer
        L1 = Conv1D(out_ch=n_out, kernel_width=f_width, pad=0, stride=s, dilation=d)

        # forward prop
        try:
            y_pred = L1.forward(X_pad)
        except ValueError:
            continue

        # ignore n. output channels
        print("Trial {}".format(i))
        print("p={} d={} s={} l_in={} f_width={}".format(p, d, s, l_in, f_width))
        print("n_ex={} n_in={} n_out={}".format(n_ex, n_in, n_out))
        assert y_pred.shape[:2] == X.shape[:2], print(
            "y_pred.shape={} X.shape={}".format(y_pred.shape, X.shape)
        )

        if p == "causal":
            gold = TorchCausalConv1d(
                in_channels=n_in,
                out_channels=n_out,
                kernel_size=f_width,
                stride=s,
                dilation=d + 1,
                bias=True,
            )
            if s != 1:
                print(
                    "TorchCausalConv1D does not do `same` padding for stride > 1. Skipping"
                )
                continue

            XT = torchify(np.moveaxis(X, [0, 1, 2], [0, -1, -2]))
        else:
            gold = nn.Conv1d(
                in_channels=n_in,
                out_channels=n_out,