Function test_LSTMCell

numpy_ml/tests/test_nn.py:1482–1556

(N=15)

def test_LSTMCell(N=15):
    from numpy_ml.neural_nets.layers import LSTMCell

    N = np.inf if N is None else N

    np.random.seed(12345)

    i = 1
    while i < N + 1:
        n_ex = np.random.randint(1, 10)
        n_in = np.random.randint(1, 10)
        n_out = np.random.randint(1, 10)
        n_t = np.random.randint(1, 10)
        X = random_tensor((n_ex, n_in, n_t), standardize=True)

        # initialize LSTM layer
        L1 = LSTMCell(n_out=n_out)

        # forward prop
        Cs = []
        y_preds = []
        for t in range(n_t):
            y_pred, Ct = L1.forward(X[:, :, t])
            y_preds.append(y_pred)
            Cs.append(Ct)

        # backprop
        dLdX = []
        dLdAt = np.ones_like(y_preds[t])
        for t in reversed(range(n_t)):
            dLdXt = L1.backward(dLdAt)
            dLdX.insert(0, dLdXt)
        dLdX = np.dstack(dLdX)
        y_preds = np.dstack(y_preds)
        Cs = np.array(Cs)

        # get gold standard gradients
        gold_mod = TorchLSTMCell(n_in, n_out, L1.parameters)
        golds = gold_mod.extract_grads(X)

        params = [
            (X, "X"),
            (np.array(Cs), "C"),
            (y_preds, "y"),
            (L1.parameters["bo"].T, "bo"),
            (L1.parameters["bu"].T, "bu"),
            (L1.parameters["bf"].T, "bf"),
            (L1.parameters["bc"].T, "bc"),
            (L1.parameters["Wo"], "Wo"),
            (L1.parameters["Wu"], "Wu"),
            (L1.parameters["Wf"], "Wf"),
            (L1.parameters["Wc"], "Wc"),
            (L1.gradients["bo"].T, "dLdBo"),
            (L1.gradients["bu"].T, "dLdBu"),
            (L1.gradients["bf"].T, "dLdBf"),
            (L1.gradients["bc"].T, "dLdBc"),
            (L1.gradients["Wo"], "dLdWo"),
            (L1.gradients["Wu"], "dLdWu"),