agmuth / skinny-glms

skinny-glms

Bare bones IRLS solver designed for speed, rcomposibility, and extensibility.

Installation

    pip install git+https://github.com/agmuth/skinny-glms.git

Examples

    examples/

Speed Comparison

Family	Link	n	p	skinnyglms_micro_secs	statsmodels_micro_secs
Gaussian	Identity	10	0	22	519
Binomial	Logit	10	0	130	510
Gamma	Log	10	0	139	1183
Poisson	Log	10	0	142	701
InverseGaussian	Log	10	0	137	1247
Gaussian	Identity	10	1	22	530
Binomial	Logit	10	1	167	704
Gamma	Log	10	1	214	2115
Poisson	Log	10	1	169	880
InverseGaussian	Log	10	1	212	3736
Gaussian	Identity	100	0	21	534
Binomial	Logit	100	0	148	861
Gamma	Log	100	0	154	1254
Poisson	Log	100	0	159	720
InverseGaussian	Log	100	0	160	1310
Gaussian	Identity	100	1	24	557
Binomial	Logit	100	1	152	895
Gamma	Log	100	1	164	1791
Poisson	Log	100	1	166	907
InverseGaussian	Log	100	1	198	4427
Gaussian	Identity	100	10	32	704
Binomial	Logit	100	10	255	1162
Gamma	Log	100	10	1700	12393
Poisson	Log	100	10	218	1103
InverseGaussian	Log	100	10	3688	17262
Gaussian	Identity	1000	0	29	792
Binomial	Logit	1000	0	335	1080
Gamma	Log	1000	0	326	1981
Poisson	Log	1000	0	333	1057
InverseGaussian	Log	1000	0	396	2103
Gaussian	Identity	1000	1	35	914
Binomial	Logit	1000	1	348	1178
Gamma	Log	1000	1	343	2661
Poisson	Log	1000	1	369	1192
InverseGaussian	Log	1000	1	418	3050
Gaussian	Identity	1000	10	87	1949
Binomial	Logit	1000	10	930	3036
Gamma	Log	1000	10	1141	11031
Poisson	Log	1000	10	878	3509
InverseGaussian	Log	1000	10	1327	14423
Gaussian	Identity	1000	100	752	19381
Binomial	Logit	1000	100	5122	24817
Gamma	Log	1000	100	12195	285360
Poisson	Log	1000	100	5062	27484
InverseGaussian	Log	1000	100	87981	-1