Two new apps glmnetUI and mgcvUI

[RCA]

Like earthUI, these two similar user interfaces operate different regression methods, which are very popular.

All of these programs will download a raw intermediate sales grid that has feature contributions, adjustments, net and gross acts and adjusted sale price.

You can use glmnetU to verify, by comparison, the results of earthUI. glmnet() is linear, but even though it is only linear, it can still do a good job. In fact if sometimes gets R2 and CVR2 values that are nearly as good as earthUI. HOWEVER, its downside is that it has gross % totals that are 2-3 times larger than earthUI or mgcvUI.

Now mgcvUI that is. a real beast. It uses tensors extensively and involves a lot of math. You can feed the Earth output into mccv, allowing it to use Earth to define knots. Then it does its thing - it will convert the segmented linear into an smooted curvilinear. In general, it also needs larger adjustments, but for the few comps you get in the sales grid, it will be about the same as Earth. Only your variable graphs will look like this:

 

[RCA]

Like earthUI, these two similar user interfaces operate different regression methods, which are very popular.

All of these programs will download a raw intermediate sales grid that has feature contributions, adjustments, net and gross acts and adjusted sale price.

You can use glmnetU to verify, by comparison, the results of earthUI. glmnet() is linear, but even though it is only linear, it can still do a good job. In fact if sometimes gets R2 and CVR2 values that are nearly as good as earthUI. HOWEVER, its downside is that it has gross % totals that are 2-3 times larger than earthUI or mgcvUI.

Now mgcvUI that is. a real beast. It uses tensors extensively and involves a lot of math. You can feed the Earth output into mccv, allowing it to use Earth to define knots. Then it does its thing - it will convert the segmented linear into an smooted curvilinear. In general, it also needs larger adjustments, but for the few comps you get in the sales grid, it will be about the same as Earth. Only your variable graphs will look like this:

...

glmetUI is easy enough to use. But I still need to add instructions.

mgcvUI - well, that can get a bit hairy. The best reference is Simon N. Woods "Generalized Additive Models: An introduction with R, Second Edition, pages 94,182, 191.

Although the best and easiest way to get good R2s, CVR2s, and low gross adjustment totals is still MARS - or earth().

-- And definitely don't bother with Bayesian Inference for residential appraisal - it has a lot of flaws - it depends for its existence on numerous assumptions and circular reasoning (e.g., Empirical Bayesian Inference).

===

I will have all of these producing pretty, - what I call - "Intermediate Sales Grids" in another week or so. - "Intermediate, because the residuals aren't broken down - some appraiser has to do that. But you may find comparing the output to your old appraisals interesting. Maybe even fun, especially if they are not your old appraisals. ...