This version of the CHHK21 code is cleaned and simplified from the full version in the private repository. Namely, we simplify in the following ways:
- Assume sampling distribution of the estimator (i.e. output from sample-aggregate) is sub-Gaussian
- Assume the analyst wants only confidence intervals and not the full covariance matrix. To this end, the algorithm estimates only the variances (rather than the full covariance matrix).
coinpress_generalized.py
: code for CoinPress mean estimationalgorithm.py
: overall algorithm from CHHK21ols_demo.py
: demonstration of how to run the algorithm for an OLS estimatorsample_and_aggregate.py
: code for sample-aggregate and bag of little bootstrapsutils.py
: a few utilities used in other files