Log entry "not enough messages to train a classifier"

BMerz · June 19, 2020, 3:32pm

Hello,
since last update of the mail app I have following warning in my nextcloud log:
“not enough messages to train a classifier”
It is shown ~ every hour.

Is something wrong with the installation? Should this be really a warning?

Thanks!

Nextcloud version: 18.0.6
Mail-App version: 1.4.0

ChristophWurst · June 19, 2020, 4:49pm

Good point. I guess it should be a info or debug statement. The code in question is at mail/lib/Service/Classification/ImportanceClassifier.php at 7d695701631060bf1d929d9e15de99e4b6e2fb68 · nextcloud/mail · GitHub. Mind submitting your first pull request to our repo?

Cheers

BMerz · June 20, 2020, 3:44pm

Yes, I can submit a pull request. Is there a small HowTo? Do I need to have access rights?
I think this line should also be changed:

github.com

nextcloud/mail/blob/7d695701631060bf1d929d9e15de99e4b6e2fb68/lib/Service/Classification/ImportanceClassifier.php#L195


		 * How many of the most recent messages are excluded from training?
		 */
		$validationThreshold = max(
			5,
			(int)(count($dataSet) * 0.1)
		);
		$validationSet = array_slice($dataSet, 0, $validationThreshold);
		$trainingSet = array_slice($dataSet, $validationThreshold);
		$logger->debug('data set split into ' . count($trainingSet) . ' training and ' . count($validationSet) . ' validation sets');
		if (empty($validationSet) || empty($trainingSet)) {
			$logger->warning('not enough messages to train a classifier');
			$perf->end();
			return;
		}
		$validationEstimator = $this->trainClassifier($trainingSet);
		try {
			$classifier = $this->validateClassifier(
				$validationEstimator,
				$trainingSet,
				$validationSet,
				$logger

ChristophWurst · June 21, 2020, 3:00pm

See https://help.github.com/en/github/collaborating-with-issues-and-pull-requests/about-pull-requests

ChristophWurst · June 22, 2020, 7:23am

Good catch! Indeed both lines should be changed.

Klemens · July 24, 2020, 3:36pm

I have the some error every hour. Thanks for proceeding the request!

h3rb3rt · September 25, 2020, 7:40am

Is there a link to the github issue ?
Or any details explanation why this happens, I am quite certain that my mail app should have enough data for the classifiers

thanks

BMerz · September 25, 2020, 7:10pm

Here you can find the pull request: