Scaling #15
Scaling #15
Conversation
YoungFaithful
commented
May 24, 2019
YoungFaithful
commented
- Scaling included for better numerical properties
- Lines adjusted as there were lines defined twice with opposite start and end node
This should fix it to 0.3 explicitly, right?
Update
Explicit dependency on ClustForOpt version
|
The fails are OK, because:
|
| @@ -39,7 +40,8 @@ function OptVariable(cep::OptModelCEP, | |||
| end | |||
| end | |||
| end | |||
| OptVariable(round.(jumparray.data;digits=round_sigdigits),jumparray.axes...; axes_names=axes_names, type=type) | |||
| unscaled_data=jumparray.data*scale[variable] #Unscale the jumparray data based on the scaling parameters in Dictionary scale | |||
There was a problem hiding this comment.
I am not entirely sure if I follow fully. You unscale the optimal variables here and store them. Where do you scale them previously?
There was a problem hiding this comment.
And does that mean that if I just take the optimization results as is form the JuMP model, the variable units would not be correct, right?
There was a problem hiding this comment.
The JuMP variables and equations are scaled within the problem formulation (one can not scale them separately like in GAMS - idea being to give you more control of what is actually given to optimizer)
The JuMP variables themselves are scaled and unscaling happens afterwards.
There was a problem hiding this comment.
Ok, I see. Let's walk through this in person together, I think this is a great change in terms of computational time reduction, but needs to be documented well, because the units do not hold anymore. Make sure to document this in the function header and also where scale is applied: What is the reason for using scale, where is it unscaled.
|
That's a significant improvement, awesome! |
… tech_l in consistence with thesis
