Turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

- JMP User Community
- :
- Discussions
- :
- Discussions
- :
- How do I use Neural Net model to predict data?

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

How do I use Neural Net model to predict data?

Dec 8, 2009 2:20 PM
(1502 views)

I have got the Neural net modeling estimated parameters:

X1, X2, X3, X4 are input

Y is output:

H1:intercept

H1:X1

H1:X2

H1:X3

H1:X4

H2:intercept

H2:X1

H2:X2

H2:X3

H2:X4

My first question: when I calculate the H1 as H1=1/(1+exp(-(intercept + Sum(H1:Xi*Xi)) is not the H1 value JMP provided.

My second question; I assume the H1 and H2 values from JMP calculation are correct and using the linear combination of them get the correct value YHat1.

However, how to use the YHat1 to calculate the Y. JMP's guide said: there is a

S_y(X) function as "The Identity function". From math, I assume it is just one since I only test one Output Y. The result I can not get right comparing with JMP prediced value.

I feel that part of Guide is too brief without many math definition and seem to me there is no easy way to implement the Neural model to predict without writing JSL.

I do want to write an independent scripts (out of JMP) to predict.

Appreciat someone can offer any help! Thanks in Advance! Ming

X1, X2, X3, X4 are input

Y is output:

H1:intercept

H1:X1

H1:X2

H1:X3

H1:X4

H2:intercept

H2:X1

H2:X2

H2:X3

H2:X4

My first question: when I calculate the H1 as H1=1/(1+exp(-(intercept + Sum(H1:Xi*Xi)) is not the H1 value JMP provided.

My second question; I assume the H1 and H2 values from JMP calculation are correct and using the linear combination of them get the correct value YHat1.

However, how to use the YHat1 to calculate the Y. JMP's guide said: there is a

S_y(X) function as "The Identity function". From math, I assume it is just one since I only test one Output Y. The result I can not get right comparing with JMP prediced value.

I feel that part of Guide is too brief without many math definition and seem to me there is no easy way to implement the Neural model to predict without writing JSL.

I do want to write an independent scripts (out of JMP) to predict.

Appreciat someone can offer any help! Thanks in Advance! Ming

3 REPLIES 3

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Re: How do I use Neural Net model to predict data?

When you have run the Neural Net platform, from the Neural Net red triangle menu choose "Save Profile Formulas". That will create a new column with a prediction formula you can use to predict future Y values from future X1, X2, X3, X4 values.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Re: How do I use Neural Net model to predict data?

Thanks! But can you be more specific? I do not really understand how to use that parameter from saving profile formulas? is that linear combination or have to get active (logistic) function involved?

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Re: How do I use Neural Net model to predict data?

Thanks MPB. I got the formula, which is embeded in the column definition.

also, can you explain why the RSquare Cross Validation (CV R2) can be negative?

I got trainind set R2 = 0.9 and CV R2 = -3.2 and R2 (both training and cross validation) is 0.95. That does not make sense to me!

also, can you explain why the RSquare Cross Validation (CV R2) can be negative?

I got trainind set R2 = 0.9 and CV R2 = -3.2 and R2 (both training and cross validation) is 0.95. That does not make sense to me!