latqcdtools.statistics.jackknife
=============

`_pareAxis(data, axis, nblocks)`
 
    In order to ensure that all jackknife blocks have the same length, we pare the data along
    the conf_axis, so that nblocks divides the length of data along conf_axis.
    
`_pseudobins(jackknifeBins, avg)`
 
    Calculate the 'pseudovalue' from the ith jackknife estimator. The pseudovalue is unbiased
    up to O(1/N), where N is the number of data. See e.g. eq. (1.1) of Miller, Biometrika 1974. 
    
`jackknife(f, data, numb_blocks=20, conf_axis=1, nproc=1, return_sample=False, args=())`
 
    Carry out a jackknife of an arbitrary function f of some data.

    Args:
        f (func)
        data (array-like)
        numb_blocks (int, optional): Number of jackknife blocks. Defaults to 20.
        conf_axis (int, optional): The axis that represents the configuration axis, i.e., measurements
          are along this axis. Defaults to 1 when data has dimension of at least 2; else defaults to 0.
        nproc (int, optional): Number of threads to use. Defaults to 1. Increasing nproc is likely to
          have a deleterious impact on performance when f is a fast function. When f is slower, for
          example if you are doing curve fits or integrating in the block, increasing nproc may
          improve performance. Feel free to benchmark it for your use-case. 
        return_sample (bool, optional): Return pseudovalues? Defaults to False.
        args (tuple, optional)

    Returns:
        jackknife mean and error (optionally pseudovalues)