search

ariesteam / aries

http://www.ariesonline.org
GNU General Public License v3.0
6 stars 1 forks source link

no output results from Bayesian statements on trained model runs #62

Closed kbagstad closed 12 years ago

kbagstad commented 12 years ago

@fvilla

For models I've trained where I needed to change all concepts to the local ontology (puget:Concept, sanPedro:Concept, etc), when I run the models they now generate outputs values plus uncertainty for the undiscretizer's concept, and values of zero for the concept in the Bayesian statement! This is often leaving values of zero to pass on to the flow models, which is obviously problematic.

Here are a couple of statements that have given me this type of problem (pull my latest changes to see for yourself):

model -o floodsinksj.nc core.models.flood-puget/sink-annual core.contexts.puget/wria-sanjuan model -o sedsourcesj.nc core.models.sediment-puget/source-puget core.contexts.puget/wria-sanjuan model -o spcsourcetest.nc core.models.carbon-san-pedro/source core.contexts.san-pedro/san-pedro-watershed-us-real

You can see the same behavior using model -d but obviously model -o is good for looking at the outputs.

Thanks, Ken

fvilla commented 12 years ago

As far as I can understand, the calculations are all OK but you're getting the Bayesian output wrong because you're asking the BN model to "keep" the wrong concept. Again, unless you say "puget" in the concept, it will go to the default namespace set in the "namespace-ontology" statement, which for this model is floodService. In your first example, when I set the concept in the keep list to "puget:GreenInfrastructureSink" I get that in the output, and it looks OK.

Please check if there's anything else wrong with it and you can close this one if not.

kbagstad commented 12 years ago

That seems to do the job for most of it - easy fix, thanks. One last question, then I think we're good to close this one out (pull my changes first):

When I try this one: model -o spwatersinktest.nc core.models.water-san-pedro/et-sink core.contexts.san-pedro/san-pedro-watershed-us-real It returns values of "null". This should be the last piece to having all trained Puget, San Pedro, and Madagascar models up and running properly...

kbagstad commented 12 years ago

Closing this one as the San Pedro water model was better off with the untrained than the trained network. Hopefully this one doesn't rear its ugly head again.