Closed obi-bot closed 7 years ago
two paras mention clustering: 1. revised text below, added output. checked refs to OBI class names - OK
For example, the OBI class hierarchical clustering is defined as a process which takes as input a collection of objects and builds a hierarchy of clusters and outputs a clustered data set. Hierarchical clustering, in practice, can be used to achieve two different objectives; to cluster data (OBI class: class discovery data transformation) and to partition data (OBI class: partitioning data transformation).
2. gene pattern use case also a figure to be updated - in progress
Original comment by: helenp
Original comment by: helenp
discussed on call, issue with objective and the logical def not matching the text needs attention The clustering objective is about clustering and not partioning tho the process may do both. JM needs to look through the dt notes. There may be other cases where the logical def of objective is similar.
Original comment by: helenp
1) Need to clarify what objective is being achieved. Should it be partitioning ever? 2) Need to verify that queries etc. will work with logical definition if revised. 3) Need to clarify what the input is. The definition implies 'objects'
Original comment by: bpeters42
discussed by JM, CS, HP after call to check objectives are correct for processes. All dts with multiple objectives checked.
actions:
1. remove partioning objective from logical definitions and text defs - these now just have class discovery objective hierarchical clustering k nearest neighbours k means clustering
2. remove partioning objective from logical and text defs - these now just have 'class discovery objectives' aggolmerative h clustering divisive h clustering gating
3. discriminant function analysis/linear discriminant function analysis - logical definition - achieves planned objective - class discovery. spec output - clustered data set. CS suggests - the outputs are predictors of outcome, therefore not class discovery. Proposal to remove class discovery objective.
4. similarity calculation - objectives class prediction/class discovery CS:objectives incorrect - needs other dt to get to these objectives. Proposal to remove both objectives
5. text definition which implied 'objects' for h'al clustering to be reworded to use 'data item'
----------------------------------------------------------------------------------------------------------- REVISED TEXT FOR PAPER based on suggested changes. removed section on multiple objectives as this no longer works.
For example, the OBI class hierarchical clustering is defined as a process which takes as input data and builds a hierarchy of clusters and outputs a clustered data set.
Original comment by: helenp
Original comment by: helenp
I'd like to amend the first 2 actions. K nearest neighbors, K means clustering, and gating do partition so they should not have that objective removed.
Original comment by: cstoeckert
changes committed to rev 2065.
revised text should read:
For example, the OBI classes k-means clustering is defined as a processes which takes as input data item and outputs a clustered data set. K means clustering, in practice, can be used to achieve two different objectives; to cluster data (OBI class: class discovery data transformation) and to partition data (OBI class: partitioning data transformation).
Original comment by: helenp
Original comment by: helenp
text not the same as what is in the file, needs to be re-aligned. There are some duplicated classes in OBI and IAO - may not have been resolved. Chris/AR/James need to resolve these while dealing with this para.
Reported by: helenp
Original Ticket: obi/obi-terms/203