Role of localized_alerts_data and submission score
by csosuteam - Thursday, August 1, 2019, 17:54:57

I am having difficulty understanding how the data table describing localized alerts relates to training and testing data. Also what type of score do we need in the submission files?

RE: Role of localized_alerts_data and submission score
by daniel_kaluza - Friday, August 02, 2019, 09:08:52

Hello,


Have you seen the post Andrzej wrote about this table? https://knowledgepit.ml/post/449/

The data table describing localized alerts contains additional features for every alert from both training set and test set. You can match the original data with them using ' alert_ids' column.

The scores that should be placed in the submission file indicate the certainty that the particular alert should be notified(is in a positive prediction class). Their range should be from [0, 1]. Then your submission will be evaluated using the AUC measure.

Please ask if you have any more questions,

Daniel

RE: Role of localized_alerts_data and submission score
by daniel_kaluza - Monday, August 05, 2019, 11:25:36

I have checked and there was a mistake in my previous post, the evaluation does not assume that scores are in range [0, 1]. The only thing that matters is the ordering of the scores, so alert that is considered to be more likely to be notified should have a greater score that alert that is less likely to be notified.

RE: Role of localized_alerts_data and submission score
by andrzej - Tuesday, August 06, 2019, 17:47:06

Hello,

let me add to Daniel's post. Tables describing localized alerts and log events can be used to define additional features for the training and test records. However, those features need to be computed - they are not given directly. Usually, many records from the additional tables correspond to a single record from the training or test data.

You may also try to use the additional data in another way, e.g., they can be used directly as an input to a recurrent neural network :-) Be creative!

Good luck,
Andrzej