NEURON version of Traub 2005 model should be tested. 82894

[nrnhines]

traubEtAl2005 downloads as a fortran version of the model (45539). The modeldb/modeldb-meta.yaml fragment is

45539: !!python/object:modeldb.data.Model
  _created: '2004-11-18T23:28:58'
  _name: A single column thalamocortical network model (Traub et al 2005)
  _object_id: 45539
  _ver_date: '2017-07-03T22:08:21.927'

We should instead get the NEURON version https://modeldb.science/download/82894 or perhaps the github version https://github.com/ModelDBRepository/82894

Perhaps this issue is related to #75

[olupton]

@ramcdougal provided some background on this here.

[nrnhines]

From a naive (or bull in a china shop) perspective, I see that

nrn-modeldb-ci$ getmodels 82894
...
Exception: Model IDs {82894} were explicitly requested, but are not known NEURON models.

On the other hand it is explicitly downloadable, eg.

curl -O https://modeldb.science/download/82894

So in the worst tradition of "hacking away", and the situation being that there are only a few (1?) of this kind of model, it would seem that a reasonable work around is to modify modeldb.py download_models to fix up its list of known NEURON model accession numbers by replacing 45539 with 82894. Is a manually maintained "special-case.yaml' file worthwhile?

[nrnhines]

The following seems to take care of the problem...

hines@hines-t7500:~/neuron/nrn-modeldb-ci/modeldb$ git diff
diff --git a/modeldb/modeldb.py b/modeldb/modeldb.py
index 87eb284..6ddafcf 100644
--- a/modeldb/modeldb.py
+++ b/modeldb/modeldb.py
@@ -126,6 +126,7 @@ class ModelDB(object):
             return requests.get(MDB_NEURON_MODELS_URL.format(model_field=field)).json()

         all_model_ids = query("id")
+        all_model_ids[all_model_ids.index(45539)] = 82894
         all_model_timestamps = query("ver_date")
         metadata = {
             model_id: timestamp

and getmodels() generates a modeldb-meta.yaml fragment of

82894: !!python/object:modeldb.data.Model
  _created: '2004-11-18T23:28:58'
  _name: A single column thalamocortical network model (Traub et al 2005)
  _object_id: 45539
  _ver_date: '2017-07-03T22:08:21.927'  

[alexsavulescu]

Not looking at the discussion, but we should rather "fix" modeldb than add an exception

[nrnhines]

I think modeldb is correct under the interpretation that the id request is for conceptual models. (the notions of model entry, conceptual model, publication(s) of that model, and model implementations are distinct. Generally one expects but does not always experience a one-to-one correspondence between the above 4 notions.) From the metadata of https://modeldb.science/api/v1/models/45539 One sees that

"modeling_application": {
    "value": [
        {"object_id": 1882, "object_name": "NEURON"},
        {"object_id": 45540, "object_name": "FORTRAN"}
    ],
    "attr_id": 114
}

which indicates there is more than one implementation (and one is for NEURON). However the default (base?) implementation is (I think things are a bit shakey at this point and it may be best to ignore this aspect)

"default_file_notes": {
    "value": "ifc fortran",
    "attr_id": 392
}

which suggest to me that the zip file for 45539 is for ifc fortran. Anyway, if there is more than one implementation and NEURON is not the default, then the one to get may appear in

"alternative_version": {
    "value": [
        {"object_id": 82894, "object_name": "NEURON"},
        {"object_id": 82895, "object_name": "g77"},
        {"object_id": 93318, "object_name": "nrntraub with pattern.mod"},
        {"object_id": 116860, "object_name": "NEURON 7"},
        {"object_id": 127353, "object_name": "neuroConstruct"}
    ],
    "attr_id": 314
}

and it this case we would wish to use 82894. Note that the metadata for https://modeldb.science/api/v1/models/45539 is identical to that of https://modeldb.science/api/v1/models/82894

Note that in unzipping, it may be worthwhile to create an id folder and unzip there since zip files with different id names can unzip to the same folder name. That happens for the 45539 and 82894 though is not an issue if we only get 82894. Nevertheless there is no guaranteed that unrelated NEURON models unzip into different folders. edit: it appears that the curators took care to make the unzip folder names different. I.e.82894 unzips to nrntraub and 116860 unzips to nrntraub_v7

[nrnhines]

Things are a bit more interesting than I supposed. I've been looking at the results from test.py

from config import *
import requests

def query(field):
    return requests.get(MDB_NEURON_MODELS_URL.format(model_field=field)).json()

nrnids = query("id")
alts = query("alternative_version") 
apps = query("modeling_application")

for i, val in enumerate(apps):
  if len(val['value']) > 1:
    # print (i, nrnids[i], val['value'])
    # print ("    ", alts[i]['value']) if alts[i] else None
    if alts[i]:
        for sim in alts[i]['value']:
            if "NEURON" in sim['object_name']:
                print(i, nrnids[i], ": replace with ", sim["object_id"],
                    "   ", sim['object_name'])

and see

hines@hines-t7500:~/neuron/nrn-modeldb-ci/modeldb$ python3 test.py
113 45539 : replace with  82894     NEURON
113 45539 : replace with  116860     NEURON 7
178 97747 : replace with  113435     NEURON
457 184142 : replace with  184144     Standalone NEURON
458 184145 : replace with  184146     Standalone NEURON
459 184148 : replace with  184149     Standalone NEURON
460 184152 : replace with  184153     Standalone NEURON
461 184157 : replace with  184158     Standalone NEURON
462 184159 : replace with  184160     Standalone NEURON
463 184161 : replace with  266824     Standalone NEURON python3
463 184161 : replace with  184162     Standalone NEURON
464 184163 : replace with  184165     Standalone NEURON
465 184166 : replace with  184167     Standalone NEURON
466 184168 : replace with  184169     Standalone NEURON
467 184171 : replace with  184172     Standalone NEURON
468 184173 : replace with  184174     Standalone NEURON
469 184176 : replace with  184177     Standalone NEURON
470 184178 : replace with  184179     Standalone NEURON
471 184180 : replace with  184181     Standalone NEURON
472 184182 : replace with  184183     Standalone NEURON
473 184184 : replace with  184185     Standalone NEURON
474 184186 : replace with  184187     Standalone NEURON
475 184196 : replace with  184197     Standalone NEURON
476 184202 : replace with  184203     Standalone NEURON
477 184231 : replace with  184232     Standalone NEURON
478 184274 : replace with  184278     Standalone NEURON
479 184297 : replace with  184298     Standalone NEURON
480 184299 : replace with  184300     Standalone NEURON
481 184301 : replace with  184302     Standalone NEURON
482 184303 : replace with  184304     Standalone NEURON
483 184305 : replace with  184306     Standalone NEURON
485 184310 : replace with  184311     Standalone NEURON
486 184312 : replace with  184313     Standalone NEURON
487 184314 : replace with  184315     Standalone NEURON
488 184316 : replace with  184317     Standalone NEURON
489 184319 : replace with  184320     Standalone NEURON
490 184321 : replace with  184322     Standalone NEURON
491 184323 : replace with  184324     Standalone NEURON
492 184325 : replace with  184326     Standalone NEURON
493 184327 : replace with  184328     Standalone NEURON
494 184329 : replace with  184330     Standalone NEURON
495 184331 : replace with  184332     Standalone NEURON
496 184333 : replace with  184334     Standalone NEURON
497 184335 : replace with  190566     Standalone NEURON
498 184337 : replace with  184338     Standalone NEURON
499 184339 : replace with  184340     Standalone NEURON
500 184341 : replace with  184342     Standalone NEURON
501 184344 : replace with  184345     Standalone NEURON
502 184346 : replace with  184347     Standalone NEURON
503 184348 : replace with  184349     Standalone NEURON
504 184350 : replace with  184351     Standalone NEURON

Also, as mentioned elsewhere, there are many cases of multiple simulators for a single id with combined files. That is not an issue in this context. The "Standalone NEURON" cases are merely preprocessed allen institute models which do not need their processing code. They are all very similar to each other and running one is probably sufficient.

[nrnhines]

it appears that the curators took care to make the unzip folder names different.

I was mistaken about that and am unfamiliar with the zip file handling in the python code so don't know if it matters, but

20212 CA1_multi
206244 CA1_multi
266844 code
267018 code
267610 code_pub
267621 code_pub
260972 Model
267184 Model
144392 modeldb
258844 modeldb
266797 modelDB
267594 modelDB
257747 ModelDB
267146 ModelDB
267646 ModelDB

[ramcdougal]

The folder names are usually whatever the model submitter chooses. No attempt is made to make them unique.

[olupton]

I believe that in nrn-modeldb-ci we add an extra {modelid}/ directory to avoid duplicates from the .zip files.