Number of permutations

[Daniele]

Hi,

I compute the TE between two different time series with these settings.
They are all quite clear (your documentation is excellent!), but I did not get what is the number of permutations exactly.
Any suggestions?

    #(3) construct a IDTxl Data object
    data=Data(np.hstack((x,y)),
          dim_order='sp',
          normalise=False)

    #(4) estimate TE
    settings = {
        'cmi_estimator': 'JidtKraskovCMI',
        'n_perm_max_stat': 100,
        'n_perm_min_stat': 100,
        'n_perm_max_seq': 100,
        'n_perm_omnibus': 100,
        'max_lag_sources': 5,
        'min_lag_sources': 0,
        'max_lag_target': 1}
    nw_bivar = BivariateTE()
    results = nw_bivar.analyse_single_target(
                   settings, data, target=1, sources=0)      #indices refer to the order in Data stacking
    te_bivar = results.get_single_target(1, fdr=False)#['te']#[0]  #first arg is the index of target in Data

    print('te_bivar:',te_bivar)

[p.wol...@gmail.com]

Hi Daniele,

glad to hear, the documentation helps. The number of permutations is used in IDTxl's non-parametric statistical tests and determines how often the test statistic (the quantity that is tested) is estimated from shuffled data. Thus, the number of permutations determines the size of the test distribution against which the original estimate is compared. The larger the number, the more precise the calculated p-value is (the number of permutations also determines the minimal possible p-value, which is 1/n_perm if the test statistic is larger/smaller than all estimates from permuted values). For example, for the maximum statistic, the number of permutations defines how often steps 1 and 2 of the test as described in the theoretical introduction are repeated.

Does this answer your question?

Best, Patricia