def main(root, pattern='*'):
    bins = np.logspace(-4, 2, 127)

    for t in lmj.cubes.Experiment(root).trials_matching(pattern):
        stats, jac = t.jacobian(10, markers=MARKERS)

        plt.colorbar(plt.imshow(jac.values[500:800].T,
                                vmin=-1, vmax=1, cmap='coolwarm'))
        plt.show()

        hist = np.zeros((len(jac), len(bins)), float)
        for t, j in enumerate(jac.values):
            for i in bins.searchsorted(abs(j)):
                hist[t, i-1] += 1
        plt.colorbar(plt.imshow(hist[500:800, :-1].T, cmap='coolwarm'))
        plt.show()
Ejemplo n.º 2
0
def main(root, pattern='*'):
    bins = np.logspace(-4, 2, 127)

    for t in lmj.cubes.Experiment(root).trials_matching(pattern):
        stats, jac = t.jacobian(10, markers=MARKERS)

        plt.colorbar(
            plt.imshow(jac.values[500:800].T, vmin=-1, vmax=1,
                       cmap='coolwarm'))
        plt.show()

        hist = np.zeros((len(jac), len(bins)), float)
        for t, j in enumerate(jac.values):
            for i in bins.searchsorted(abs(j)):
                hist[t, i - 1] += 1
        plt.colorbar(plt.imshow(hist[500:800, :-1].T, cmap='coolwarm'))
        plt.show()
Ejemplo n.º 3
0
def main(args):
    data = np.load(args.dataset, mmap_mode='r')
    N = data.shape[1]
    T = args.frames
    K = int(N * T * args.overcomplete)

    def batches():
        batch = np.zeros((args.batch_size, N * T), 'f')
        for b in range(args.batch_size):
            o = np.random.randint(len(data) - T - 1)
            batch[b] = data[o:o+T].ravel()
        return [batch]

    net = theanets.Autoencoder([N * T, (K, 'linear'), (N * T, 'tied')])
    net.train(batches,
              monitors={'hid1:out': (-0.1, -0.01, 0.01, 0.1)},
              **models.kwargs_from_args(args))

    D = net.find('hid1', 'w').get_value().T

    R = 6
    C = 18
    dimg = np.zeros((R * (N + 1) - 1, C * (T + 1) - 1), float)
    bimg = np.zeros((R * (N + 1) - 1, C * (T + 1) - 1), float)
    idx = abs(D).max(axis=1).argsort()[::-1]
    for r in range(R):
        for c in range(C):
            o = np.random.randint(len(data) - T - 1)
            dimg[r*(N+1):r*(N+1)+N, c*(T+1):c*(T+1)+T] = data[o:o+T].T
            bimg[r*(N+1):r*(N+1)+N, c*(T+1):c*(T+1)+T] = D[idx[r * C + c]].reshape((T, N)).T

    _, (dax, bax) = plt.subplots(1, 2)
    dax.imshow(dimg, cmap='coolwarm')
    sns.despine(ax=dax, left=True, bottom=True)
    dax.set_xticks([])
    dax.set_yticks([])
    bax.imshow(bimg, cmap='coolwarm')
    sns.despine(ax=bax, left=True, bottom=True)
    bax.set_xticks([])
    bax.set_yticks([])
    plt.show()

    logging.info('%s: saving %s %s', args.codebook, D.shape, D.dtype)
    np.save(args.codebook, D)