Opened 7 months ago

Last modified 3 weeks ago

#2690 assigned feature request

Implement views for decision tree models (random forest and gradient boosted trees)

Reported by: gkronber Owned by: gkronber
Priority: medium Milestone: HeuristicLab 3.3.15
Component: Algorithms.DataAnalysis.Views Version: 3.3.14
Keywords: Cc:

Description

It would be nice if we could inspect decision tree models

Change History (10)

comment:1 Changed 7 months ago by gkronber

  • Status changed from new to accepted

comment:2 Changed 7 months ago by gkronber

r14345: implemented methods to generate symbolic expression tree solutions for decision tree models (random forest and gradient boosted) as well as views which make it possible to inspect each of the individual trees in a GBT and RF solution

comment:3 Changed 7 months ago by gkronber

  • Owner changed from gkronber to mkommend
  • Status changed from accepted to reviewing

comment:4 Changed 7 months ago by gkronber

r14346: removed unnecessary plugin dependency

Last edited 7 months ago by gkronber (previous) (diff)

comment:5 follow-up: Changed 7 months ago by mkommend

  • Owner changed from mkommend to gkronber
  • Status changed from reviewing to assigned

Bug: At least RF models are not stored directly, but rather recreated when necessary. However the new model view does not trigger this recreation and throws an out-of-bounds exception if a deserialized model is displayed.

comment:6 in reply to: ↑ 5 Changed 7 months ago by gkronber

Replying to mkommend:

Bug: At least RF models are not stored directly, but rather recreated when necessary. However the new model view does not trigger this recreation and throws an out-of-bounds exception if a deserialized model is displayed.

Thanks. Should be fixed with r14368.

comment:7 Changed 7 months ago by gkronber

  • Owner changed from gkronber to mkommend
  • Status changed from assigned to reviewing

comment:8 Changed 6 months ago by mkommend

Review comments:

  • Double click in the list view does not open the symbolic regression solution.
  • The model view is the last entry in the solution view.
  • Why are evaluation views misused to display the individual trees. Would not it be possible to create a view for IRandomForestModel / GBTModel?
  • Why are GBTSolutionView and RF Regression/Classifation View necessary? Only to remove the menu entry?
  • Excel format for the symbolic models does not work correctly. There is no infinity symbol in excel and additionally 1/(1+EXP(-Infinity * ($D1-44)) is not easy to interpret correctly. Especially for RF where ignoreSlope == true a standard if-condition should be exported.
  • Depending on whether ignoreSlope is set the condition is negated (> changes to <).
  • Slope of +/- infinity has the same meaning as ignoreSlope=true and not slope == 0 (VariableConditionTreeNode.ToString).
  • VariableConditionView should show the ignoreSlope flag
  • IMHO IgnoreSlope should be the default for variable condition symbols.

gkronber: After discussion we decided it would be best to introduce a new type of symbol especially for this (split var <= threshold). This also means that we need to extend interpreters and formatters to support this symbol.

Last edited 3 weeks ago by gkronber (previous) (diff)

comment:9 Changed 5 months ago by mkommend

  • Owner changed from mkommend to gkronber
  • Status changed from reviewing to assigned

comment:10 Changed 7 weeks ago by mkommend

r14839: Adapated content attribute of solution views for GBT and RF.

This change has been necessary to show the solution name text box from the NamedDataAnalysisSolutionView.

Note: See TracTickets for help on using tickets.