Sampling for randomization statistics (spynal.randstats.sampling)

Random sample generators for randomization, permutation (shuffle), and bootstrap statistics

Function reference

  • permutations : Generate random permutations (resampling w/o replacement)

  • bootstraps : Generate random bootstrap samples (resampling w/ replacement)

  • signs : Generate random binary variables (eg for sign tests)

  • jackknifes : Generate jackknife samples (exclude each observation in turn)

  • subsets : Generate random length-k subsets (resampling w/o replacement)

Function reference

permutations(n, n_resamples=9999, seed=None)

Yield generator with a set of n_resamples random permutations of integers 0:n-1, as would be needed for permutation/randomization tests

Parameters:

  • n (int) – Number of items to randomly resample from. Will usually correspond to number of observations/trials

  • n_resamples (int, default: 9999 (appropriate number for test w/ 10,000 samples)) – Number of independent resamples to generate.

  • seed (int, default: None) – Random generator seed for repeatable results. Set=None for unseeded random numbers.

Yields:

resamples (generator, shape=(n_resamples,) of [ndarray, shape=(n,), dtype=int]) – Generator to iterate over for permutation test. Each iteration contains a random permutation of integers 0:n-1.

bootstraps(n, n_resamples=9999, seed=None)

Yield generator with a set of n_resamples random resamplings with replacement of integers 0:n-1, as would be needed for bootstrap tests or confidence intervals

Parameters:

  • n (int) – Number of items to randomly resample from. Will usually correspond to number of observations/trials

  • n_resamples (int, default: 9999 (appropriate number for test w/ 10,000 samples)) – Number of independent resamples to generate.

  • seed (int, default: None) – Random generator seed for repeatable results. Set=None for unseeded random numbers.

Yields:

resamples (generator, shape=(n_resamples,) of [ndarray, shape=(n,), dtype=int]) – Generator to iterate over for boostrap test or confidence interval computation. Each iteration contains a random resampling with replacement from integers 0:n-1.

signs(n, n_resamples=9999, seed=None)

Yield generator with a set of n_resamples random Bernoulli(p=0.5) variables (ie binary 0/1 w/ probability of 0.5), each of length <n>, as would be needed to set the signs of stats in a sign test.

Parameters:

  • n (int) – Number of items to randomly resample from. Will usually correspond to number of observations/trials

  • n_resamples (int, default: 9999 (appropriate number for test w/ 10,000 samples)) – Number of independent resamples to generate.

  • seed (int, default: None) – Random generator seed for repeatable results. Set=None for unseeded random numbers.

Yields:

resamples (generator, shape=(n_resamples,) of [ndarray, shape=(n,), dtype=bool]) – Generator to iterate over for random sign test. Each iteration contains a random resampling of n Bernoulli random variables.

jackknifes(n, n_resamples=None, seed=None)

Yield generator with a set of n_resamples = n boolean variables, each of length n, and each of which excludes one observation/trial in turn, as would be needed for a jackknife or leave-one-out test.

Parameters:

  • n (int) – Number of items to randomly resample from. Will usually correspond to number of observations/trials

  • n_resamples (int) – Automatically set=n here. Only included for consistent interface.

  • seed (None) – Not used. Only included for consistent interface with other functions.

Yields:

resamples (generator, shape=(n,) of [ndarray, shape=(n,), dtype=bool]) – Generator to iterate over for jackknife test. Each iteration is all 1’s except for a single 0, the observation (trial) excluded in that iteration. For the ith resample, the ith trial is excluded.

subsets(n, k, n_resamples=9999, seed=None)

Yield generator with a set of n_resamples random length-k subsets of integers 0:n-1.

Random sampling version of “n-choose-k” function.

Parameters:

  • n (int) – Number of items to randomly resample from. Will usually correspond to number of observations/trials.

  • k (int) – Length of subset to select. For example, k=2 implies random sampling from “n-choose-2”.

  • n_resamples (int, default: 9999 (appropriate number for test w/ 10,000 samples)) – Number of independent resamples to generate.

  • seed (int, default: None) – Random generator seed for repeatable results. Set=None for unseeded random numbers.

Yields:

resamples (generator, shape=(n_resamples,) of [ndarray, shape=(k,), dtype=int]) – Generator to iterate over for a random-subset test. Each iteration contains a random subset of integers 0:n-1.