You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
It seems like the RelevancyIndex interface and intent is just a specific use of a SupervisedClassifier instance upon a set of DescriptorElements. The IqrSession structure is the only thing currently (within SMQTK at least) that calls the RelevancyIndex.rank method (aside from tests), thus the only thing that requires the specific positive/negative classifier training and application to a DescriptorElement set. A more generalized structure here would be for the IqrSession to take in a SupervisedClassifier implementation config (instead of a RelevancyIndex) and IqrSession.refine to define the training and application logic that is specific to the IQR process instead of fragmenting it out into a separate class (the RelevancyIndex).
This would also work to address opening up the use of other supervised classifiers for use in IQR refinement instead of currently needing to implement an analogous RelevancyIndex.
The text was updated successfully, but these errors were encountered:
Purg
changed the title
Replace the use of the RelevancyIndex with just a Classifier instance
Replace the use of the RelevancyIndex with a Classifier instance
Mar 19, 2019
It seems like the
RelevancyIndex
interface and intent is just a specific use of aSupervisedClassifier
instance upon a set ofDescriptorElements
. TheIqrSession
structure is the only thing currently (within SMQTK at least) that calls theRelevancyIndex.rank
method (aside from tests), thus the only thing that requires the specific positive/negative classifier training and application to aDescriptorElement
set. A more generalized structure here would be for theIqrSession
to take in aSupervisedClassifier
implementation config (instead of aRelevancyIndex
) andIqrSession.refine
to define the training and application logic that is specific to the IQR process instead of fragmenting it out into a separate class (theRelevancyIndex
).This would also work to address opening up the use of other supervised classifiers for use in IQR refinement instead of currently needing to implement an analogous
RelevancyIndex
.The text was updated successfully, but these errors were encountered: