Veda2.0 Released!


Updated VEDA-FE and now model wont run
#1
Hello,

I recently updated my VEDA-FE to 4.5.5. I rectified the additional import log problems which came up as a result of this update and I dont think there is anything else in there which should stop the model from working properly. I have also updated my TIMES code to the latest given by VEDA-FE.

However when I try to run a BASE version of the model the iterations yielding the objective function get to about 520,000,000 (for reference my previous BASE and other people using our TIAM model generally get an ObjZ of about 300,000,000). The following then comes up and carries on (essentially for as long as I leave the command window open).

Primal crossover.
  Primal:  Fixing 509372 variables.
   509371 PMoves:  Infeasibility 1.39847156e+000  Objective 5.17532841e+008

Given I havent changed anything in the model and it seems to have Synced in to VEDA-FE, I was wondering whether anyone had any ideas as to why this is happening?

Any help would be greatly appreciated. 

Dan
Reply
#2
Hi Dan,

You have to compare the DataOnly gdx of both versions to know the differences.

Create DataOnly gdx by selecting the associated check box under User Options > General.
Thereafter, select the Create DD Only option in the Case Manager.

Refer to the attached images for more details.


Attached Files Thumbnail(s)
       
Reply
#3
(16-11-2017, 04:33 PM)Vikrant Wrote: Hi Dan,

You have to compare the DataOnly gdx of both versions to know the differences.

Create DataOnly gdx by selecting the associated check box under User Options > General.
Thereafter, select the Create DD Only option in the Case Manager.

Refer to the attached images for more details.

Hi Vikrant,

Thanks for getting back to me!
Given that I have not changed anything in these models and just updated I'm abit confused as to what is going on.

I was about to try and compare a BASE run of my TIMES model I did before the update but all the gdx files etc. have disappeared. Instead, I asked someone at UCL to run the TIMES Demo model and send me on the results. I've attached both zipped files of the DataOnly gdx and screenshots of the model runs in the command window. My OBJZ is higher than the other demo run, and it seems to only be optimal after using an extreme point solution (or something similar).

Not having my original runs from before I updated hasn't helped - ideally I'd check the results in VEDA-BE but I cannot do that as I cant run any of the models to check them against the old results. 

Any other advice would be really appreciated!

Thanks,

Dan


Attached Files
.zip   TIMES_DEMO_DaTa_OnLy_1711161524OldVEDAUCL.gdx.zip (Size: 28.94 KB / Downloads: 5)
.zip   TIMES_DEMO~DaTa_OnLy_1711161419NewVEDA.gdx.zip (Size: 32.09 KB / Downloads: 1)
.docx   Screenshot_CMDGAMSRunDemo.docx (Size: 190.89 KB / Downloads: 4)
Reply
#4
Interesting problem.
I tried to look at the GDX files, but got the following error for both:

Problem reading GDX file:
GDX file version not supported


That was the result with all versions GAMS 23.7, GAMS 24.7, GAMS 24.8 and GAMS 24.9.
It seems the GDX files may be corrupt?  Can you read them yourself with your GAMS?
Reply
#5
Hi Vikrant,

I've played around with the CPLEX settings in the Case Manager and managed to get the model to solve; the OBJZ is still much higher than it should be (or at least was a couple of weeks ago!) but at least I can have a look in VEDA-BE now to the results and hopefully work out what has changed.

Thanks for your help and directing me to those GDX files.

Dan
Reply
#6
(16-11-2017, 10:21 PM)Antti-L Wrote: Interesting problem.
I tried to look at the GDX files, but got the following error for both:

Problem reading GDX file:
GDX file version not supported


That was the result with all versions GAMS 23.7, GAMS 24.7, GAMS 24.8 and GAMS 24.9.
It seems the GDX files may be corrupt?  Can you read them yourself with your GAMS?

Hi Antti,

Thanks for the response - I can read some of the GDX files with my GAMS yes, but something definitely isn't quite right. 

I'll have a more in-depth look tomorrow and get back to you if I can't work out where the issue lies.

Thanks for your input,

Dan
Reply
#7
I tried to figure out why the GDX files you provided are corrupted.
I suspect that you may have opened them in a text editor (e.g. Notepad.exe) and then saved them as *.txt.
I tested doing so, and it causes nulls to be converted into blanks, just like in the corrupted GDX files you provided.

Therefore, if you would like that someone takes a look at the differences in the model data generated from VEDA, please make sure that you ZIP the original (binary) GDX files without doing any editing to them, and then post the ZIP file(s) here. (You can of course do that only for a small model like the DEMO).
Reply
#8
(17-11-2017, 05:31 PM)Antti-L Wrote: I tried to figure out why the GDX files you provided are corrupted.
I suspect that you may have opened them in a text editor (e.g. Notepad.exe) and then saved them as *.txt.
I tested doing so, and it causes nulls to be converted into blanks, just like in the corrupted GDX files you provided.

Therefore, if you would like that someone takes a look at the differences in the model data generated from VEDA, please make sure that you ZIP the original (binary) GDX files without doing any editing to them, and then post the ZIP file(s) here. (You can of course do that only for a small model like the DEMO).

Hi Antti,

I've attached a file but I think its the same as the ones yesterday - when I run DD only and Create Data Only GDX the results appear in my GamsSave file in GAMS_WrkTimes and they are automatically there to be opened as txt files (I've taken a screenshot of the properties of the Demo model I've just done - I haven't opened it or changed the name). Is this a settings issue? The zip below is how it automatically appeared in the file without any modifications.

Apologies for my lack of knowledge on this, and thanks for your continued help.

Additionally in the word document I've attached a screenshot of the error message which comes up in Excel every time I open it since I updated - its looking for a path from the old version of VEDA for some reason?

Dan


Attached Files
.zip   DEMO_Test~DaTa_OnLy_1711171255.zip (Size: 32.14 KB / Downloads: 2)
.docx   Screenshot_FilePropertiesandError.docx (Size: 125.05 KB / Downloads: 3)
Reply
#9
Ok, the GDX file in the new ZIP was correct; it opens well with GAMS.
But there was only a single file in the new ZIP - so there is nothing to compare...
Is this the old one, or the new one?

GDX files should be associated with the GAMS GDX Viewer, certainly not Notepad.
Reply
#10
(17-11-2017, 06:41 PM)Antti-L Wrote: Ok, the GDX file in the new ZIP was correct; it opens well with GAMS.
But there was only a single file in the new ZIP - so there is nothing to compare...
Is this the old one, or the new one?

GDX files should be associated with the GAMS GDX Viewer, certainly not Notepad.

The one I just sent you is the new one (from my computer with the new VEDA).

I have attached the old (correct) DEMO run below, which I did yesterday. If that doesn't work let me know and I will do it from another computer.


Attached Files
.zip   TIMES_DEMO_DaTa_OnLy_1711161524 (3).zip (Size: 29.04 KB / Downloads: 1)
Reply
#11
Very well, I made a pretty extensive comparison and investigation.

I was first able to conclude, by comparing your GDX files, that your differences were, in the first place, caused by using different DEMO models in your runs. Second, in order to see if there might be any other factors at play, I made my own test runs with the DEMO model, using the following steps:

1. I first tested the DEMOv50 model with VEDA v4.5.512 (the latest version), and TIMES v4.1.5.  I obtained exactly the same value for the objective function as you did with VEDA 4.5.5 (OBJz = 61509042.520702).

2. Then, to check for any possible differences caused by different TIMES versions, I tested the same model input files (obtained in Step 1) with TIMES v3.6.1 (from 2014, over three years old). I still got exactly the same value for the objective function as in step 1. So, TIMES versions 3.6.1–4.1.5 appear to produce a fully equivalent model and the same value for the objective function.

3. Finally, I tested also importing the same model into an old VEDA-FE v4.3.5 (nearly 5 years old!). Importing the model successfully to such an old version required adding some ~FI_COMM tables into the Base templates (they were in SysSettings in the DEMO). But after doing that, the import was fine, and I solved the model from the DD files generated by VEDA-FE v4.3.5 with both TIMES v4.1.5 and TIMES v3.6.1.  Once again, the solutions were identical.

In summary, my conclusion is that VEDA v4.5.5 (with TIMES v4.1.5) produces the same formulation and the same solution for the DEMO model as a five year old VEDA v4.3.5 with TIMES v3.6.1. I thus believe that the differences you were seeing were entirely caused by running different DEMO models with your old and new versions of VEDA.

Finally, I was astounded to realize that you had, in fact, been using a very old TIMES version (3.6.1 or below based on my investigation). I don't think there can be support for such old versions...
Reply
#12
(17-11-2017, 11:19 PM)Antti-L Wrote: Very well, I made a pretty extensive comparison and investigation.

I was first able to conclude, by comparing your GDX files, that your differences were, in the first place, caused by using different DEMO models in your runs. Second, in order to see if there might be any other factors at play, I made my own test runs with the DEMO model, using the following steps:

1. I first tested the DEMOv50 model with VEDA v4.5.512 (the latest version), and TIMES v4.1.5.  I obtained exactly the same value for the objective function as you did with VEDA 4.5.5 (OBJz = 61509042.520702).

2. Then, to check for any possible differences caused by different TIMES versions, I tested the same model input files (obtained in Step 1) with TIMES v3.6.1 (from 2014, over three years old). I still got exactly the same value for the objective function as in step 1. So, TIMES versions 3.6.1–4.1.5 appear to produce a fully equivalent model and the same value for the objective function.

3. Finally, I tested also importing the same model into an old VEDA-FE v4.3.5 (nearly 5 years old!). Importing the model successfully to such an old version required adding some ~FI_COMM tables into the Base templates (they were in SysSettings in the DEMO). But after doing that, the import was fine, and I solved the model from the DD files generated by VEDA-FE v4.3.5 with both TIMES v4.1.5 and TIMES v3.6.1.  Once again, the solutions were identical.

In summary, my conclusion is that VEDA v4.5.5 (with TIMES v4.1.5) produces the same formulation and the same solution for the DEMO model as a five year old VEDA v4.3.5 with TIMES v3.6.1. I thus believe that the differences you were seeing were entirely caused by running different DEMO models with your old and new versions of VEDA.

Finally, I was astounded to realize that you had, in fact, been using a very old TIMES version (3.6.1 or below based on my investigation). I don't think there can be support for such old versions...

Antti,

That's good news that you got the same OBJZ for the DEMO.

I'll keep an eye on results etc. - when I looked in VEDA-BE at a run I did yesterday with my TIMES model using an identical run to a pre-update scenario, the Objective Function was > 200,000,000 higher in the run yesterday. However, in general the results broadly match up (emissions and consumption of the fossil fuels).

Thanks very much for looking into this in so much detail, and sorry for using up so much of your time.

Dan
Reply


Forum Jump:


Users browsing this thread: 1 Guest(s)