All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
- Small tweak to QR decomposition lapack call -> overall runtime and memory usage improvement of ~24% and 19%, respectively. Oops.
test_vars
option to specify the variable(s) for which to calculate PL confidence intervals and p-values.
- Fixed bug where
.summary(xname)
would appendIntercept
toxname
such that repeated calls would break.
- Option to use Wald method for computing p-values and confidence intervals instead of LRT and profile likelihood. Set
wald=True
to use (#11). - Tests for
load_sex2()
andload_endometrial()
(#9). - Test for profile likelihood confidence intervals (#9).
skip_lrt
option is nowskip_pvals
(#11).
.summary()
no longer breaks if skipping confidence interval or p-value calculation (#11).
- Diabetes and sex2 csv files removed from testing dir (#9).
- Added the endometrial cancer dataset analyzed in Heinze and Schemper, 2002. Load using
load_endometrial()
(#8).
- Disabled step-halving by default to follow
logistf
, which disabled it in version 1.24.1 for some reason (#8).
v0.3.0 adds a couple of convenience features.
- Added
.summary()
method to print summary of results (#6). See the readme for a usage example. - Added the sex2 dataset from logistf as
load_sex2()
(#7).
v0.2.0 is the MVP release implemented in Numpy and Scipy.
- Calculate confidence intervals using profile penalized likelihood (#5).
skip_ci
parameter to skip calculation of confidence intervals.alpha
parameter to specify confidence interval.pl_max_iter
,pl_max_halfstep
,pl_max_stepsize
parameters for controlling profile penalized likelihood procedure.
- Predictions calculated during fitting procedure are now clipped to the range
[1e-15, 1-1e-15]
instead of[0, 1]
.
- Added missing
loglik_
attribute to readme.
- Parameters and attributes documentation in readme.
- Penalized likelihood ratio test implemented for calculating p-values (#2).
skip_lrt
parameter for skipping the p-value calculations.
- Calculate coef standard errors in
_bse()
. loglik_
and_bse
attributes for fitted log-likelihood and coef standard errors.
- Call lapack routines to get Q only instead of
np.linalg.qr
.
- Added info for PyPI in pyproject.toml.
- Added install instruction to readme.
Initial release