Icon

03_​Active_​Learning_​Uncertainty_​Sampling

Active Learning - Uncertainty Sampling
Active Learning - Uncertainty Sampling: This workflow shows examples of ActiveLearning with Uncertainty Sampling. Legacy Example - before KNIME AP 4.1 (Dec. 2019)Before KNIME 4.1 the Active Learning extension included nodes that did not support interactive JavaScript views for KNIME WebPortal. This example in particular uses"Auto Active Learn Loop End" which does not enable the user at all to label the instances, but it is still quite useful for testing and benchmark purposes. To automaticallylabel the selected instances by uncertainty sampling at each iteration the workflow uses the ground truth in the test data. Current Example - since KNIME AP 4.1 (Dec. 2019)With KNIME Analytics Platform 4.1 the Active Learning extension was updated to support interactive JavaScript views for KNIME WebPortal. In this example you caninteractively label the instances using KNIME WebPortal, which comes with KNIME Server. To test the workflow until the second iteration dothe following: 1) Right click Labeling View >"Execute and OpenViews"2) Label the instance using the in-view buttons3) Save new labels by clicking "Apply" and "Close"in the View lower left corner4) Right Click the Active Learning Loop End node>"Step Loop Execution" twice5) Open the Labeling View again.To apply interactions after the first iteration youneed the KNIME WebPortal, where the workflow ishidden and the user can access the view as aremotely accessible web-based application with"Next" and "Back" buttons. Simply execute the entire workflow to select aninstance by uncertainty sampling at each iterationand retrieve the right label from the orginal testdata. To run the human-in-the-loop applicationreplace Auto Active Learn Loop End (Legacy) withActive Learning Loop End (Legacy) and open itsview while it is executing.This workflow cannot be deployed on KNIMEWebPortal via KNIME Server. initialize Tree EnsembleLearner Data Preparation Createartifical data Active Learn LoopStart (Legacy) Learn + Predict Margin UncertaintyScorer Top k Selector Auto Active LearnLoop End (Legacy) Createartifical data Data Preparation Active LearningLoop End Active LearningLoop Start Learn + Predict Row Splitter Labeling View forKNIME WebPortal Active Learning - Uncertainty Sampling: This workflow shows examples of ActiveLearning with Uncertainty Sampling. Legacy Example - before KNIME AP 4.1 (Dec. 2019)Before KNIME 4.1 the Active Learning extension included nodes that did not support interactive JavaScript views for KNIME WebPortal. This example in particular uses"Auto Active Learn Loop End" which does not enable the user at all to label the instances, but it is still quite useful for testing and benchmark purposes. To automaticallylabel the selected instances by uncertainty sampling at each iteration the workflow uses the ground truth in the test data. Current Example - since KNIME AP 4.1 (Dec. 2019)With KNIME Analytics Platform 4.1 the Active Learning extension was updated to support interactive JavaScript views for KNIME WebPortal. In this example you caninteractively label the instances using KNIME WebPortal, which comes with KNIME Server. To test the workflow until the second iteration dothe following: 1) Right click Labeling View >"Execute and OpenViews"2) Label the instance using the in-view buttons3) Save new labels by clicking "Apply" and "Close"in the View lower left corner4) Right Click the Active Learning Loop End node>"Step Loop Execution" twice5) Open the Labeling View again.To apply interactions after the first iteration youneed the KNIME WebPortal, where the workflow ishidden and the user can access the view as aremotely accessible web-based application with"Next" and "Back" buttons. Simply execute the entire workflow to select aninstance by uncertainty sampling at each iterationand retrieve the right label from the orginal testdata. To run the human-in-the-loop applicationreplace Auto Active Learn Loop End (Legacy) withActive Learning Loop End (Legacy) and open itsview while it is executing.This workflow cannot be deployed on KNIMEWebPortal via KNIME Server. initialize Tree EnsembleLearner Data Preparation Createartifical data Active Learn LoopStart (Legacy) Learn + Predict Margin UncertaintyScorer Top k Selector Auto Active LearnLoop End (Legacy) Createartifical data Data Preparation Active LearningLoop End Active LearningLoop Start Learn + Predict Row Splitter Labeling View forKNIME WebPortal

Nodes

Extensions

Links