相空间重构、时间序列重构、状态空间重构、phase space reconstruction、state space reconstruction

闲话几句

相空间重构听起来高大上，但是一旦确定了嵌入维数和延迟时间，之后的操作就十分简单了。

下边给个代码示例一下“之后的操作”，至于嵌入维数和延迟时间的确定，还是需要查查论文的，后期看情况再写。

python程序

def embed_vectors_1d(self, lag, embed):
    """Embeds vectors from a one dimensional time series in m-dimensional
    space.

    Parameters
    ----------
    X : 1d array
        Training or testing set.
    lag : int
        Lag value as calculated from the first minimum of the mutual info.
    embed : int
        Embedding dimension. How many lag values to take.
    predict : int
        Distance to forecast (see example).

    Returns
    -------
    features : 2d array
        Contains all of the embedded vectors. Shape (num_vectors,embed).

    Example
    -------
    >>> X = [0,1,2,3,4,5,6,7,8,9,10]
    em = 3
    lag = 2
    predict=3

    >>> embed_vectors_1d
    features = [[0,2,4], [1,3,5], [2,4,6], [3,5,7]]
    """

    tsize = self.X.shape[0]
    t_iter = tsize-(lag*(embed-1))

    features = np.zeros((t_iter,embed))

    for ii in range(t_iter):

        end_val = ii+lag*(embed-1)+1

        part = self.X[ii : end_val]

        features[ii,:] = part[::lag]

    return features

这个程序是从skccm里边摘抄出来的，程序中用了for循环，这个在效率上不太高。

朴素的解释

相空间重构是要解决什么问题呢？最近看的两篇网文^{1 2}很好的解释了这个问题，先把图片粘贴到这里，有人想看的话我再加上文字解释。

see also

1. skccm：https://github.com/NickC1/skccm（GitHub可能无法打开）

---

1. Visualization of EDM Prediction with a Chaotic Time Series ↩︎

2. Explaining simplex projection ↩︎