Mistakes easily made when using KnowledgeFlow

Introduction#

Weka KnowledgeFlow(KF) is a great interface to use. However, Weka manual does not cover every little details of using KF. Here would be a place for collecting those little tricks or details I learnt from those errors I did or will make as time goes. Many thanks to people at Wekalist (especially Mark Hall, Eibe Frank) for building a wonderful learning environment for Weka!

Remarks#

TrainingSetMaker and TestSetMaker

1. a ClassAssigner must be linked between ArffLoader and TrainingSetMaker or TestSetMaker.

ArffSaver

2. In order to save dataset into arff file successfully, it is safer to set relationNameForFilename to False inside configuration of ArffSaver.

How to use TimeSeriesForecasting in KnowledgeFlow?

1. Open knowledgeFlow, load dataset with ArffLoader
2. go to setting, check time series forecasting perspective, right-click ArffLoader to send to all perspective
3. go to time series forecasting perspective to set up a model
4. run the model and copy the model to clipboard
5. ctrl + v, and click to paste model to Data mining process canvas
6. save prediction along with original data with ArffSaver

How to open KnowledgeFlow file directly from terminal

1. add the following function into .bash_profile, save and exit

function wekaflstart() {
export R_HOME=/Library/Frameworks/R.framework/Resources
java -Xss10M -Xmx4096M -cp :weka.jar weka.gui.knowledgeflow.KnowledgeFlow “$1”
}

2. inside a directory with a weka.jar file, open its terminal, run wekastart "path to a knowledgeflow file"