Closed martcaro closed 1 year ago
Looks good. Are model access and needsTrainingData object properties or would it be better to have them as data type properties with ranges xsd:anyURI and xsd:Boolean?
On Fri, 12 May 2023, 17:53 Marta Caro Martínez, @.***> wrote:
Metadata for exaplainers are required to do the filtering step properly. The metada we should include is the following:
- modelAccess: Object property. Description: if the explainer execution is made by file or url. Values: Any, File, URL
- supportsBWImage: New DatasetType instance (in explainer.owl). Description: only for images, if supports RAW images.
- needsTrainingData: Object property. Description: if the explainer needs training data, the user has to upload it. Values: True-False
More details about how to use these properties (and other ones already in the ontology) here: https://docs.google.com/spreadsheets/d/1v8d89Ex_Fm_4JOiAmagaawQA1UppfVMKwKgVGm-Wnf8/edit#gid=1337559317
— Reply to this email directly, view it on GitHub https://github.com/isee4xai/iSeeOnto/issues/50, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAJHPMOE4I7HS64RRVPJPG3XFZTJFANCNFSM6AAAAAAX7YUATQ . You are receiving this because you are subscribed to this thread.Message ID: @.***>
Model access type can be non-URI. How about - Explainer ‘has model access’ ModelAccessType. And ModelAccessType has individuals: Any, File and URL
@dcorsar I guess it's better to have has model access and needsTrainingData as object properties, because they are going to appear always, right? Data property is a good option, though.
@ike01 I agree
Yes, this looks like a good option to model that requirement. Will we also be capturing the URL and/or the location/name of the file or is it just that the model is accessed through a File or URL?
On Fri, 12 May 2023 at 18:16, Ikechukwu Nkisi-Orji @.***> wrote:
Model access type can be non-URI.
How about - Explainer ‘has model access’ ModelAccessType. And ModelAccessType has individuals: Any, File and URL
— Reply to this email directly, view it on GitHub https://github.com/isee4xai/iSeeOnto/issues/50#issuecomment-1546050875, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAJHPMMMGY4QVFBBCELJSK3XFZWADANCNFSM6AAAAAAX7YUATQ . You are receiving this because you commented.Message ID: @.***>
What would the value of needsTrainingData be if its an ObjectProperty? Will it be pointing to details of the training data?
Regarding modelAccess: we only need to capture if the model is uploaded by file or shared by URL, no need of adding more information. Regarding needsTrainingData: we are going to represent that property with True or False, so and idea could be to create a class TrainingDataRequirement, with two instances TrainingDataRequired (True), and NoTrainingDataRequired (False). Let me know if you think this is good, or we might look for another idea!
It's probably best to use data properties since it's ever going to be either true or false strings. We could go with 'needs training data' with xsd:string data type and include {"True", "False"} for the data range. Thoughts?
<owl:DatatypeProperty rdf:about="http://www.w3id.org/iSeeOnto/explainer#needs_training_data">
<rdfs:range>
<rdfs:Datatype>
<owl:oneOf>
<rdf:Description>
<rdf:type rdf:resource="http://www.w3.org/1999/02/22-rdf-syntax-ns#List"/>
<rdf:first>False</rdf:first>
<rdf:rest>
<rdf:Description>
<rdf:type rdf:resource="http://www.w3.org/1999/02/22-rdf-syntax-ns#List"/>
<rdf:first>True</rdf:first>
<rdf:rest rdf:resource="http://www.w3.org/1999/02/22-rdf-syntax-ns#nil"/>
</rdf:Description>
</rdf:rest>
</rdf:Description>
</owl:oneOf>
</rdfs:Datatype>
</rdfs:range>
<rdfs:label xml:lang="en">needs training data</rdfs:label>
</owl:DatatypeProperty>
Okay, yes, maybe it is better option to use data property, TrainingDataRequirement might be a very specific class. If we use data property, I think we could use xsd:boolean instead of string???
Metadata for exaplainers are required to do the filtering step properly. The metada we should include is the following:
More details about how to use these properties (and other ones already in the ontology) here: https://docs.google.com/spreadsheets/d/1v8d89Ex_Fm_4JOiAmagaawQA1UppfVMKwKgVGm-Wnf8/edit#gid=1337559317