MGCX

class hyppo.time_series.MGCX(compute_distance='euclidean', max_lag=0, **kwargs)

Cross Multiscale Graph Correlation (MGCX) test statistic and p-value.

MGCX is an independence test between two (paired) time series of not necessarily equal dimensions. The population parameter is 0 if and only if the time series are independent. It is based upon energy distance between distributions.

Parameters

  • compute_distance (str, callable, or None, default: "euclidean") -- A function that computes the distance among the samples within each data matrix. Valid strings for compute_distance are, as defined in sklearn.metrics.pairwise_distances,

    • From scikit-learn: ["euclidean", "cityblock", "cosine", "l1", "l2", "manhattan"] See the documentation for scipy.spatial.distance for details on these metrics.

    • From scipy.spatial.distance: ["braycurtis", "canberra", "chebyshev", "correlation", "dice", "hamming", "jaccard", "kulsinski", "mahalanobis", "minkowski", "rogerstanimoto", "russellrao", "seuclidean", "sokalmichener", "sokalsneath", "sqeuclidean", "yule"] See the documentation for scipy.spatial.distance for details on these metrics.

    Set to None or "precomputed" if x and y are already distance matrices. To call a custom function, either create the distance matrix before-hand or create a function of the form metric(x, **kwargs) where x is the data matrix for which pairwise distances are calculated and **kwargs are extra arguements to send to your custom function.

  • max_lag (int, default: 0) -- The maximum number of lags in the past to check dependence between x and the shifted y. If None, then max_lag=np.ceil(np.log(n)). Also the M hyperparmeter below.

  • **kwargs -- Arbitrary keyword arguments for compute_distance.

Notes

The statistic can be derived as follows 1:

Let \(x\) and \(y\) be \((n, p)\) and \((n, q)\) series respectively, which each contain \(y\) observations of the series \((X_t)\) and \((Y_t)\). Similarly, let \(x[j:n]\) be the \((n-j, p)\) last \(n-j\) observations of \(x\). Let \(y[0:(n-j)]\) be the \((n-j, p)\) first \(n-j\) observations of \(y\). Let \(M\) be the maximum lag hyperparameter. The cross distance correlation is,

\[\mathrm{MGCX}_n (x, y) = \sum_{j=0}^M \frac{n-j}{n} MGC_n (x[j:n], y[0:(n-j)])\]

References

1

Cencheng Shen, Jaewon Chung, Ronak Mehta, Ting Xu, and Joshua T Vogelstein. Independence testing for temporal data. Transactions on Machine Learning Research, 2024.

Methods Summary

MGCX.statistic(x, y)

Helper function that calculates the MGCX test statistic.

MGCX.test(x, y[, reps, workers, random_state])

Calculates the MGCX test statistic and p-value.
MGCX.statistic(x, y)

Helper function that calculates the MGCX test statistic.

Parameters

x,y (ndarray of float) -- Input data matrices. x and y must have the same number of samples. That is, the shapes must be (n, p) and (n, q) where n is the number of samples and p and q are the number of dimensions. Alternatively, x and y can be distance matrices, where the shapes must both be (n, n).

Returns

  • stat (float) -- The computed MGCX statistic.

  • opt_lag (int) -- The computed optimal lag.

  • opt_scale ((int, int)) -- The computed optimal scale as a pair of two elements.

MGCX.test(x, y, reps=1000, workers=1, random_state=None)

Calculates the MGCX test statistic and p-value.

Parameters

  • x,y (ndarray of float) -- Input data matrices. x and y must have the same number of samples. That is, the shapes must be (n, p) and (n, q) where n is the number of samples and p and q are the number of dimensions. Alternatively, x and y can be distance matrices, where the shapes must both be (n, n).

  • reps (int, default: 1000) -- The number of replications used to estimate the null distribution when using the permutation test used to calculate the p-value.

  • workers (int, default: 1) -- The number of cores to parallelize the p-value computation over. Supply -1 to use all cores available to the Process.

  • auto (bool, default: True) -- Automatically uses fast approximation when n and size of array is greater than 20. If True, and sample size is greater than 20, then hyppo.tools.chi2_approx will be run. Parameters reps and workers are irrelevant in this case. Otherwise, hyppo.tools.perm_test will be run.

Returns

  • stat (float) -- The computed MGCX statistic.

  • pvalue (float) -- The computed MGCX p-value.

  • mgcx_dict (dict) --

    Contains additional useful returns containing the following keys:

    • opt_lagint

      The optimal lag that maximizes the strength of the relationship with respect to lag.

    • opt_scale(int, int)

      The optimal scale that maximizes the strength of the relationship with respect to scale.

Examples

The optimal scale should be global (n, n) for cases of linear correlation.

>>> import numpy as np
>>> from hyppo.time_series import MGCX
>>> np.random.seed(456)
>>> x = np.arange(7)
>>> y = x
>>> stat, pvalue, mgcx_dict = MGCX().test(x, y, reps = 100)
>>> '%.1f, %.2f, [%d, %d]' % (stat, pvalue, mgcx_dict['opt_scale'][0],
... mgcx_dict['opt_scale'][1])
'1.0, 0.06, [7, 7]'

The increasing the max_lag can increase the ability to identify dependence.

>>> import numpy as np
>>> from hyppo.time_series import MGCX
>>> np.random.seed(1234)
>>> x = np.random.permutation(10)
>>> y = np.roll(x, -1)
>>> stat, pvalue, mgcx_dict = MGCX(max_lag=1).test(x, y, reps=1000)
>>> '%.1f, %.2f, %d' % (stat, pvalue, mgcx_dict['opt_lag'])
'1.1, 0.00, 1'

Examples using hyppo.time_series.MGCX

Time Series Testing

Time Series Testing

previous

KSampleHHG

next

DcorrX