switching.py 文件源码

def count_matrices(data, start_state=None, threshold=None, display=False):
    num_clusters = 2
    if threshold is None:
        clust = clusterer(data)
        state = clust.fit_predict(data.reshape(-1, 1)).reshape(data.shape)
    else:
        logger.debug("Cluster data based on threshold = {}".format(threshold))
        state = data > threshold

    init_state  = state[:,:,0]
    final_state = state[:,:,1]
    switched    = np.logical_xor(init_state, final_state)

    init_state_frac = [np.mean(init_state == ct) for ct in range(num_clusters)]
    for ct, fraction in enumerate(init_state_frac):
        logger.debug("Initial fraction of state %d: %f" %(ct, fraction))

    if start_state is not None and start_state in range(num_clusters):
        start_stt = start_state
    else:
        start_stt = np.argmax(init_state_frac)
    logger.debug("Start state set to state: {}".format(start_stt))
    logger.debug("Switched state is state: {}".format(1-start_stt))

    # This array contains a 2x2 count_matrix for each coordinate tuple
    count_mat = np.zeros((init_state.shape[0], 2, 2))

    # count_mat      = np.zeros((2,2), dtype=np.int)
    count_mat[:,0,0] = np.logical_and(init_state == 0, np.logical_not(switched)).sum(axis=-1)
    count_mat[:,0,1] = np.logical_and(init_state == 0, switched).sum(axis=-1)
    count_mat[:,1,0] = np.logical_and(init_state == 1, switched).sum(axis=-1)
    count_mat[:,1,1] = np.logical_and(init_state == 1, np.logical_not(switched)).sum(axis=-1)

    return count_mat, start_stt