Simple Comparison of Weka Interfaces

Introduction#

Weka has many interfaces, Explorer, KnowledgeFlow, Experimenter, SimpleCLI, Workbench.

All of them share mostly can do the same tasks, with different focus and flexibility.

Here, we are going to explore their different focuses and flexibilities.

Remarks#

Explorer

pro:

1. do all things quickly
2. give a quick and comprehensive view of data structure

cos: can’t save the process;

Experimenter

pro:

1. compare several models at once, e.g., run 3 different classifiers against 5 datasets all together, and see the compared result at one place;
2. experiment can be saved

KnowledgeFlow

pro:

1. do almost all things that Explorer can do
2. can save the process

cos:

1. KF can’t do Experimenter’s job, as it doesn’t support loops, but ADAMS can help;
2. KF can’t access low-level functionalities inside Weka API;

simpleCLI

pro: run similar tasks of what Explorer does using command line

cos: it can’t access all functionalities of Weka API, Jython or Groovy scripting is recommended for this task.

Workbench

pro: it gathers all other interfaces together into one place

simpleCLI and Jython examples

simpleCLI

go to simpleCLI, enter the following code

java weka.classifiers.rules.ZeroR -t path/to/a-file-of-dataset

Jython Example

codes from Advanced Weka MOOC course lesson 5.1

# imports
import weka.core.converters.ConverterUtils.DataSource as DS
import weka.filters.Filter as Filter
import weka.filters.unsupervised.attribute.Remove as Remove
import os

# load data
data = DS.read(os.environ.get("MOOC_DATA") + os.sep + "iris.arff")

# remove class attribute
rem = Remove()
rem.setOptions(["-R", "last"])
rem.setInputFormat(data)
dataNew = Filter.useFilter(data, rem)

# output filtered dataset
print(dataNew)