proc hpgenselect lasso output interpretation

[chester2018]

Hi everyone! I'm new to LASSO regression, so I'm unsure how to interpret the output from this procedure. Here is the code I ran below:

proc hpgenselect data=data;
	class var1 (ref="0") var2 (ref="0") 
	var3 (ref="0") var4 (ref="0") 
  var5(ref="0") var6 (ref="0") var7 (ref="0") 
  var8 (ref="0") var9 (ref="0");
   model varx (ref="0")= var1 var2 var3 var4 var5 var6 var7 var8 var9
   / dist=mult link=glogit;
   selection method=lasso(choose=aicc) details=all;
   performance details;
run;

I'm confused on Selection Details and Parameter Estimates. I am under the impression I would use Proc Hpgenselect for the Lasso regression since the dependent variable is multinominal (rather than continuous). And I am under the impression that Lasso is useful to determine which independent variables would influence the dependent variable the most.

Any help will be greatly appreciated! *Var 3 for some reason doesn't come up in the output*

Step	Description	Effects In Model	Lambda	AIC	AICC	BIC
0	Initial Model	1	1	9524.699	9524.706	9543.186
1	var1	3	.8	9515.842	9515.883	9565.142
	var7	3	.8	9515.842	9515.883	9565.142
2	var6	4	.64	9507.382	9507.458	9575.170
3	var2	5	0.512	9492.006	9492.224	9609.094
4	var5	7	0.4096	9484.026	9484.526	9662.739
	var9	7	0.4096	9484.026	9484.526	9662.739
5	var8	8	0.3277	9451.726	9452.536	9679.739
6	var4	9	0.2621	9422.729	9423.822	9687.717
7		9	0.2097	9391.901	9392.994	9656.890
8		9	0.1678	9374.869	9376.118	9658.345
9		9	0.1342	9363.952	9365.427	9672.077
10		9	0.1074	9351.942	9353.600	9678.556
11		9	0.0859	9345.010	9346.927	9696.274
12		9	0.0687	9333.868	9335.785	9685.132
13		9	0.055	9329.730	9331.784	9693.319
14		9	0.044	9323.719	9325.773	9687.307
15		9	0.0352	9319.290	9321.344	9682.878
16		9	0.0281	9315.928	9317.982	9679.517
17		9	0.0225	9313.381	9315.435	9676.970
18		9	0.018	9311.495	9313.549	9675.084
19		9	0.0144	9312.130	9314.254	9681.881
20		9	0.0115	9311.125	9313.249*	9680.876

Parameter Estimates
Parameter	Varx Group	DF	Estimate
Intercept	1	1	-0.923848
Intercept	3	1	-0.090371
Intercept	2	1	-0.475460
var1 1	1	1	0.208376
var1 1	3	1	0.085181
var1 1	2	1	-0.033519
var1 2	1	1	0.564515
var1 2	3	1	0.614004
var1 2	2	1	0.208413
var1 3	1	1	0.775994
var1 3	3	1	0.824564
var1 3	2	1	0.306114
var2 1	1	1	-0.258637
var2 1	3	1	-0.344585
var2 1	2	1	-0.420374
var2 2	1	1	-0.183059
var2 2	3	1	-0.006305
var2 2	2	1	-0.034611
var2 3	1	1	-0.656491
var2 3	3	1	-0.643286
var2 3	2	1	-0.339439
var41	1	1	0.216545
var41	3	1	0.176734
var41	2	1	-0.017829
var5 1	1	1	0.049015
var5 1	3	1	-0.122792
var51	2	1	0.374545
var5 2	1	1	-0.018575
var5 2	3	1	-0.220216
var5 2	2	1	0.117048
var6 1	1	1	0.576625
var6 1	3	1	0.338929
var6 1	2	1	0.416339
var6 2	1	1	0.998560
var6 2	3	1	0.743957
var6 2	2	1	0.502328
var6 3	1	1	1.339983
var6 3	3	1	0.982288
var6 3	2	1	0.665514
var7 1	1	1	-0.287458
var7 1	3	1	-0.587046
var7 1	2	1	-0.274003
var7 2	1	1	-0.346579
var7 2	3	1	-0.053197
var7 2	2	1	0.017640
var8 1	1	1	-0.059514
var8 1	3	1	-0.033493
var8 1	2	1	0.072684
var8 2	1	1	0.006447
var8 2	3	1	0.132097
var8 2	2	1	0.037160
var9 1	1	1	0.051825
var9 1	3	1	0.129466
var9 1	2	1	0.359189
var9 2	1	1	0.322085
var9 2	3	1	0.201435
var9 2	2	1	0.571104
var9 3	1	1	-0.118308
var9 3	3	1	-0.155053
var9 3	2	1	0.346671

[sbxkoenk]

---

@chester2018 wrote:

Any help will be greatly appreciated! *Var 3 for some reason doesn't come up in the output*

---

lasso is used as variable selection method. Your Var_3 is not relevant for improving multi-class classification.

See also here (there is also coverage of overall "variable importance") :

• Home  >>  SAS Communities Library  >> Multinomial Classification in SAS Model Studio

https://communities.sas.com/t5/SAS-Communities-Library/Multinomial-Classification-in-SAS-Model-Studi...

Last update: ‎01-17-2024
Updated by: SAS Employee AndyRavenna

Koen