0 ×

05_​AL_​Re-label_​Uncertain_​Classes

Workflow

Update Data Sets Emil the TeacherBot - WebPortal Interface - Re-Labeling IThis workflow is part of a number of other workflows that address a data mining scenario at the intersection of active learning, text mining, stream mining and service-oriented knowledge discoveryarchitectures.This workflow, in particular, provides a graphical interface on the Webportal for a KNIME specialist to re-label the question with the most uncertain predicted classes.It starts by first reading a subset of the training set (10% of the most uncertain predicted classes). Then, it loops over all the questions, and for each one those, it allows the specialist to choosebetween one of the predicted classes or the option "Something Else". The labeling phase takes place in the "Choose Answer" webpage. To complete the execution of the loop the specialist has tocomplete the labeling for all the no-processed questions of the subset of the training set, or to click "Exit". If the specialist clicks "Exit" the workflow saves the last step of the loop iteration. Thus, whenthe specialist starts again the execution of the workflow on the WebPortal, he/she will be able to start labeling the questions from the last loop iteration.After the Variable Condition Loop End the data get split between questions that have been classified as "Something Else" and all the other categories. These two datasets are then saved into twodifferent tables. Reading Data Process the QuestionAt this stage the workflow processes the questions and predicts the class for each one of those. The answers aredisplayed on a webpage and the specilalist can select one. no_valid_answerrow at current Iteration onlynext versionTraining setRead subset of the uncertaineddataset Table Writer Variable ConditionLoop End Generic Loop Start Row Filter Appending class Table Writer Get_Class_Probabilities Prepare Datafor Writing Reading Data Choose Answer Update Data Sets Emil the TeacherBot - WebPortal Interface - Re-Labeling IThis workflow is part of a number of other workflows that address a data mining scenario at the intersection of active learning, text mining, stream mining and service-oriented knowledge discoveryarchitectures.This workflow, in particular, provides a graphical interface on the Webportal for a KNIME specialist to re-label the question with the most uncertain predicted classes.It starts by first reading a subset of the training set (10% of the most uncertain predicted classes). Then, it loops over all the questions, and for each one those, it allows the specialist to choosebetween one of the predicted classes or the option "Something Else". The labeling phase takes place in the "Choose Answer" webpage. To complete the execution of the loop the specialist has tocomplete the labeling for all the no-processed questions of the subset of the training set, or to click "Exit". If the specialist clicks "Exit" the workflow saves the last step of the loop iteration. Thus, whenthe specialist starts again the execution of the workflow on the WebPortal, he/she will be able to start labeling the questions from the last loop iteration.After the Variable Condition Loop End the data get split between questions that have been classified as "Something Else" and all the other categories. These two datasets are then saved into twodifferent tables. Reading Data Process the QuestionAt this stage the workflow processes the questions and predicts the class for each one of those. The answers aredisplayed on a webpage and the specilalist can select one. no_valid_answerrow at current Iteration onlynext versionTraining setRead subset of the uncertaineddataset Table Writer Variable ConditionLoop End Generic Loop Start Row Filter Appending class Table Writer Get_Class_Probabilities Prepare Datafor Writing Reading Data Choose Answer

Download

Get this workflow from the following link: Download

Nodes

05_​AL_​Re-label_​Uncertain_​Classes consists of the following 120 nodes(s):

Plugins

05_​AL_​Re-label_​Uncertain_​Classes contains nodes provided by the following 7 plugin(s):