Greetings!
Please see model attached.
I use "time" to get values from an array, "investment profile," into the variable "investment".
1) The value of "investment" is slightly different from that of the values in the "investment profile" array; since the value of the time step is one full period, the simulation is run over 10 periods and there are 10 elements in the array - should there be any interpolation?
2) I can't figure out how to optimize the lookup table; the control panel gives me the following error when I choose "investment profile"as a "currently active parameter" - "ERROR: No usable constant name. 0<=investment profile<=10".
Thanks!
Optimize Lookup
-
- Senior Member
- Posts: 107
- Joined: Wed Nov 26, 2008 6:12 am
Optimize Lookup
- Attachments
-
- optimizelookup.zip
- (10.66 KiB) Downloaded 206 times
vector lookup optimize
Hi Karan
I do not see where is the problem.
You just have to set the optimizer to policy (not calibration) with the vpd file and pen to optimize with the weight -1 and take the profile array as parameters with bounds for instance between 0 and 10.
Joined the vpd, voc and vdf files.
Regards.
JJ
I do not see where is the problem.
You just have to set the optimizer to policy (not calibration) with the vpd file and pen to optimize with the weight -1 and take the profile array as parameters with bounds for instance between 0 and 10.
Joined the vpd, voc and vdf files.
Regards.
JJ
- Attachments
-
- opt vector lookup2.zip
- (1.95 KiB) Downloaded 192 times
-
- Senior Member
- Posts: 107
- Joined: Wed Nov 26, 2008 6:12 am
Hi Karan
I try to avoid optimizing lookups. Befor doing it I try to fit the lookup by an algebraic formula. The advantage is that you have generally much less parameters to optimize (the parameters of the algebraic formula) and most important it keeps the shape of the lookup consistent. Otherwise when you optimize directly a lookup you may get strange results totally independant from one another, generating an irrealistic curve. I do this when the parameters of the lookup are not independant (the parameters draw a specific curve shape). If the parameters are totally independant which to my opinion is estremely rare for a lookup, you are obliged to optimize directly the lookup parameters.
This remark is especillay valid for multidimensional lookups.
Or lookups with more than one input variable.
It eases too multivariate sensibility analysis and using synthesim.
Regards.
JJ
I try to avoid optimizing lookups. Befor doing it I try to fit the lookup by an algebraic formula. The advantage is that you have generally much less parameters to optimize (the parameters of the algebraic formula) and most important it keeps the shape of the lookup consistent. Otherwise when you optimize directly a lookup you may get strange results totally independant from one another, generating an irrealistic curve. I do this when the parameters of the lookup are not independant (the parameters draw a specific curve shape). If the parameters are totally independant which to my opinion is estremely rare for a lookup, you are obliged to optimize directly the lookup parameters.
This remark is especillay valid for multidimensional lookups.
Or lookups with more than one input variable.
It eases too multivariate sensibility analysis and using synthesim.
Regards.
JJ
-
- Senior Member
- Posts: 107
- Joined: Wed Nov 26, 2008 6:12 am
Another obvious candidate for a functional form is a polynomial of time, e.g.
decision = constant + a*time + b*time^2 ...
A much better option, if possible, is to create a feedback rule, e.g.
decision = constant + a*level_1 + b*level_2 ...
decision = constant + a*time + b*time^2 ...
A much better option, if possible, is to create a feedback rule, e.g.
decision = constant + a*level_1 + b*level_2 ...
/*
Advice to posters (it really helps us to help you)
http://www.ventanasystems.co.uk/forum/v ... f=2&t=4391
Blog: http://blog.metasd.com
Model library: http://models.metasd.com
Bookmarks: http://delicious.com/tomfid/SystemDynamics
*/
Advice to posters (it really helps us to help you)
http://www.ventanasystems.co.uk/forum/v ... f=2&t=4391
Blog: http://blog.metasd.com
Model library: http://models.metasd.com
Bookmarks: http://delicious.com/tomfid/SystemDynamics
*/