Icon

01_​Feature_​Elimination_​Done_​Manually

Manual Feature Selection

These workflows demonstrate interesting way to use loops. They replicate functionality that's available as a set of nodes

Perform model evaluation on randomly chosen feature subsetsand aggregate average accurarcy of those models for selected features. Gradually reduce number of features based on ranking in inner loop. Recursive Feature Elimination. Done manually. Recursive Feature Construction. Done manually. Perform model evaluation on existing feature subset + each one of the remaining featuresand aggregate average accurarcy of those models for the selected feature. Add features one-by-one based on ranking in inner loop. Initialization What you always wanted to do in KNIME but never dared to: Manual Feature Selection.DISLAIMER: These workflows demonstrate interesting way to use loops.They replicate functionality that's available as a set of nodesDO NOT USE IN REAL LIFE - use the nodes on the right instead!The workflow on the top recursively picks features, the one on the bottom reduces a set of features to a specified count. start featurereductionfeedbackreduced featureset to start node.initial listof featuresstart featureimportance evalpick 10% ofcolumns randomlyisolate classcolumnremove classcolumncollect...add classcolumnadd accuracyto each chosenfeature...andsummarizefeature statssory byaccuracyleave onlytop 90% featuresreduce datato (column) subsetremove classcolumncomplete listof featuresisolate classcolumnartificial datawith 4 usefuland 100 noisyfeaturescreate emptystart feature listpick x+1selected columnsfrom data tableiterate overall unusedfeaturescreate list of not yetincluded featuresx+1 features totry in this runtrain and evalclassifiercollect statsfor potentialnew featuressort byaccuracypick topfeaturecleantableadd selectedfeature to listappendclass columnfor speed upartificial datawith 4 usefuland 100 noisyfeaturestrain and evalclassifiercollectresultsturn var intoone new rowextract featureand accuracy Delegating LoopStart (deprecated) Delegating LoopEnd (deprecated) List Columns Counting Loop Start Row Sampling Column Filter Column Filter Loop End Joiner Variable to Tablecolumn (deprecated) GroupBy Sorter Row Sampling Reference ColumnFilter (from Column) Column Filter List Columns Column Filter Data Generation Table Creator Reference ColumnFilter (from Column) Table Row ToVariable Loop Start ReferenceRow Filter Concatenate EvaluateNaive Bayes Loop End Sorter Row Sampling Column Filter Concatenate Column Appender Cache Data Generation EvaluateNaive Bayes RecursiveLoop Start Recursive Loop End Variable toTable Row Variable toTable Row Feature SelectionLoop Start (1:1) Feature SelectionLoop End Feature SelectionFilter Perform model evaluation on randomly chosen feature subsetsand aggregate average accurarcy of those models for selected features. Gradually reduce number of features based on ranking in inner loop. Recursive Feature Elimination. Done manually. Recursive Feature Construction. Done manually. Perform model evaluation on existing feature subset + each one of the remaining featuresand aggregate average accurarcy of those models for the selected feature. Add features one-by-one based on ranking in inner loop. Initialization What you always wanted to do in KNIME but never dared to: Manual Feature Selection.DISLAIMER: These workflows demonstrate interesting way to use loops.They replicate functionality that's available as a set of nodesDO NOT USE IN REAL LIFE - use the nodes on the right instead!The workflow on the top recursively picks features, the one on the bottom reduces a set of features to a specified count. start featurereductionfeedbackreduced featureset to start node.initial listof featuresstart featureimportance evalpick 10% ofcolumns randomlyisolate classcolumnremove classcolumncollect...add classcolumnadd accuracyto each chosenfeature...andsummarizefeature statssory byaccuracyleave onlytop 90% featuresreduce datato (column) subsetremove classcolumncomplete listof featuresisolate classcolumnartificial datawith 4 usefuland 100 noisyfeaturescreate emptystart feature listpick x+1selected columnsfrom data tableiterate overall unusedfeaturescreate list of not yetincluded featuresx+1 features totry in this runtrain and evalclassifiercollect statsfor potentialnew featuressort byaccuracypick topfeaturecleantableadd selectedfeature to listappendclass columnfor speed upartificial datawith 4 usefuland 100 noisyfeaturestrain and evalclassifiercollectresultsturn var intoone new rowextract featureand accuracy Delegating LoopStart (deprecated) Delegating LoopEnd (deprecated) List Columns Counting Loop Start Row Sampling Column Filter Column Filter Loop End Joiner Variable to Tablecolumn (deprecated) GroupBy Sorter Row Sampling Reference ColumnFilter (from Column) Column Filter List Columns Column Filter Data Generation Table Creator Reference ColumnFilter (from Column) Table Row ToVariable Loop Start ReferenceRow Filter Concatenate EvaluateNaive Bayes Loop End Sorter Row Sampling Column Filter Concatenate Column Appender Cache Data Generation EvaluateNaive Bayes RecursiveLoop Start Recursive Loop End Variable toTable Row Variable toTable Row Feature SelectionLoop Start (1:1) Feature SelectionLoop End Feature SelectionFilter

Nodes

Extensions

Links