df3e7eaf84e65be7e521c6de0379fb588de533ba,pynndescent/pynndescent_.py,,init_from_random,#Any#Any#Any#Any#Any#Any#Any#,43

Before Change

def init_from_random(n_neighbors, data, query_points, heap, dist, dist_args, rng_state):
    for i in range(query_points.shape[0]):
        if heap[0, i, 0] == -1:
            for j in range(np.sum(heap[0, i] == -1)):
                idx = np.abs(tau_rand_int(rng_state)) % data.shape[0]
                d = dist(data[idx], query_points[i], *dist_args)
                heap_push(heap, i, d, idx, 1)
    return


@numba.njit(fastmath=True)

After Change

        //         idx = np.abs(tau_rand_int(rng_state)) % data.shape[0]
        //         d = dist(data[idx], query_points[i], *dist_args)
        //         heap_push(heap, i, d, idx, 1)
        while heap[0, i, 0] == -1:
            idx = np.abs(tau_rand_int(rng_state)) % data.shape[0]
            d = dist(data[idx], query_points[i], *dist_args)
            // potentially we can get away with this
            unchecked_heap_push(heap, i, d, idx, 1)

    return


@numba.njit(fastmath=True)

Instances