PDA

View Full Version : information



Sabrina
10-08-2007, 11:33 AM
Hi there,
I have a problem with Weka.
I have many dataset and I want to carry out some experiments by using these at the same time. I think I can use Experimenter but I want to obtain as output the number of true positive,true negative, false positive and false negative.
When I used j48 in Explorer I obtained:
=== Confusion Matrix ===

a b <-- classified as
10893 604 | a = weekday
2963 1088 | b = weekend
TP = 10893

When I used j48 in Experimenter I obtained:
Tester: weka.experiment.PairedCorrectedTTester
Analysing: Num_true_positives
Datasets: 1
Resultsets: 1
Confidence: 0.05 (two tailed)
Sorted by: -
Date: 08/10/07 17.04

Dataset (1) trees.J48 '-
------------------------------------------
'demog_trans store.txt-we (5) 2178.60 |
------------------------------------------
(v/ /*) |
Key:
(1) trees.J48 '-C 0.25 -M 2' -217733168393644444

Why the Num_true_positives is different?What is the Num_true_positives in Experimenter?
Is it possible to obtain in Experimenter the same result as Explorer?

Thank you

Mark
10-08-2007, 10:15 PM
Hi there,

Are you doing the same evaluation in both the Explorer and Experimenter - i.e. cross-validation with the same number of folds and random seed?

Cheers,
Mark.

Sabrina
10-09-2007, 02:19 AM
Yes.
In fact the TP rate,TN rate, FP rate and FN rate are the same in both the Explorer and Experimenter.

Cheers,

Sabrina

Mark
10-09-2007, 04:22 PM
The Experimenter is computing an average over the five folds, whereas the Explorer reports the sum. 5 * 2178.60 = 10893.

Cheers,
Mark

Sabrina
10-11-2007, 12:18 PM
Thank you.

Cheers,
Sabrina