Converting separate text files containing training data and its labels to ARFF format

Question

I need to perform a classification task in Weka for a dataset. The dataset contains of 3 text files training.txt, label_training.txt and testing.txt. The format of training.txt and testing.txt is as follows:

InformationID  FeatureID  Value
1                6         1.00
1               160       31.00
1               438        1.00
1               479        1.00
2              6457        2.00
2              6664        0.65
2              6761        0.46
2              6762        1.00

The label_training.txt contains the class labels for the training data and each row represents a data point in the training set.

1
1
1
-1
1
-1
-1

Does this mean row 1 of label_training.txt file corresponds to all rows in the training.txt file which has InformationID 1? I would like to make sure if I am understanding it right. So one data point in the training set corresponds to InformationID 1 with values of 4 features with ID 6,160,438,479?

Now, how do I create an ARFF file which combines the training data and the labels for it to derive a classifier? Any help would be appreciated.

Answer 1

Well, it seems that your dataset is in an sparse format in which InformationID identifies the instance, and FeatureID identifies the feature, being Value the value for each couple of instance/feature.

Lets us assume that the label_training.txt is explicit (an instance is identified by the line, for example line #1 identifies instance #1, which corresponds to InformationID 1.

In this case, you need to generate ARFF files like the following one:

@relation my-relation

@attribute my-class {-1,1}
@attribute 1 numeric
@attribute 2 numeric
../..

@data
{0 1, 6 1.00, 160 31.00, 438 1.00, 479 1.00}
{0 1, 6457 2.00, 6664 0.65, 6761 0.46, 6762 1.00}
../..

This a WEKA sparse ARFF format in which each couple of numbers correspond to an attribute number and to its value.

I suggest to write an script to perform this transformation.