Closed GoogleCodeExporter closed 9 years ago
You can use the "wndchrm test" functionality to cross-validate a classifier
that you built using the "wndchrm train" command. You can have wndchrm run up
to 200 train/test splits or what we call "simulations."
You would use a command like `wndchrm test -n200 my_classifier.fit
output_report.html`, where "my_classifier.fit is a file you generated with
wndchrm train, and output_report.html is the output of the wndchrm test
cross-validation.
Opening the output_report.html, you will see the results of 200 independent
train/test splits, that is, dividing the images within the classifier
"my_classifier.fit" randomly 200 times into two pools: a training set (75% of
the images) and a test set (remaining 25% of images). Classification accuracies
are generated for each split. The standard deviation of the classification
accuracies across splits is a measure of how homogenous/heterogeneous the
morphology is in the images within "my_classifier.fit"
Does this answer your question?
Original comment by christop...@nih.gov
on 25 May 2012 at 7:56
Original comment by christop...@nih.gov
on 10 Jun 2012 at 6:15
Original issue reported on code.google.com by
valmlo...@gmail.com
on 21 May 2012 at 10:22