{ "id": 5649, "title": "Clarification of submission format", "url": "/competitions/expedia-personalized-sort/discussion/5649", "author_name": "", "post_date": "2013-09-04T13:10:44.657Z", "votes": null, "comment_count": 4, "views": 2175, "content": "

We are predicting "booking_bool". So how does this then get converted into the submission format? thanks

\n

 

\n

ps. Not everyone uses Python - I shouldn't have to go through the benchmark code to get submission format data and find out what we are actually trying to predict

", "messages": [ { "id": "30146", "postDate": "09/04/2013 13:10:44", "content": "

We are predicting "booking_bool". So how does this then get converted into the submission format? thanks

\n

 

\n

ps. Not everyone uses Python - I shouldn't have to go through the benchmark code to get submission format data and find out what we are actually trying to predict

", "rawMarkdown": "", "votes": null }, { "id": "30149", "postDate": "09/04/2013 13:42:59", "content": "

Hi,

\n

We are interested in predicting both booking_bool and click_bool.

\n

More precisely, the goal is to rank hotels, in the context of each user query (or impression, identified by SearchId), from hotels that are most likely to be booked, followed by hotels that are most likely to be only clicked, followed by remaining hotels. If an impression was not followed by a booking, then hotels that are most likely to be clicked should appear on the top of the rank.

\n

The submission should be given as a coma-separated file with columns corresponding respectively to SearchId and PropertyId (see the "write_submission" function in https://github.com/benhamner/ExpediaPersonalizedSortCompetition/blob/master/PythonBenchmark/data_io.py). An example of a submission file is here: https://www.kaggle.com/c/expedia-personalized-sort/download/basicPythonBenchmark.zip .

\n

I hope it helps.

\n

Cheers,Adam

\n

 

\n

 

\n

X

", "rawMarkdown": "", "votes": null }, { "id": "30153", "postDate": "09/04/2013 14:44:53", "content": "

Ok thanks. A little further forward. So the "position" field in the training set is redundant ? and there are no ranks  in the training set ? We will run models on Train to predict Click_bool and boolean bool, make predictions on the test set and then rank these predictions and use the properyid field of the predictions as the submission format? X

", "rawMarkdown": "", "votes": null }, { "id": "30154", "postDate": "09/04/2013 15:00:30", "content": "

Hi,

\n

'So the "position" field in the training set is redundant ?' - I think so. In the case when the order of records in the file per each impression does not correspond to the value in the position column, then the latter information takes precedence.

\n

"and there are no ranks  in the training set ? " - no, there are not ranks, just the click_bool and purchase_bool columns.

\n

"We will run models on Train to predict Click_bool and boolean bool, make predictions on the test set and then rank these predictions and use the properyid field of the predictions as the submission format? "  - yes

\n

 

\n

Adam

\n

 

\n

 

\n

 

", "rawMarkdown": "", "votes": null }, { "id": "30157", "postDate": "09/04/2013 15:02:05", "content": "

thanks

", "rawMarkdown": "", "votes": null } ], "comments": [ { "id": 30149, "author_name": "adamwoz", "author_url": "", "post_date": "09/04/2013 13:42:59", "content": "

Hi,

\n

We are interested in predicting both booking_bool and click_bool.

\n

More precisely, the goal is to rank hotels, in the context of each user query (or impression, identified by SearchId), from hotels that are most likely to be booked, followed by hotels that are most likely to be only clicked, followed by remaining hotels. If an impression was not followed by a booking, then hotels that are most likely to be clicked should appear on the top of the rank.

\n

The submission should be given as a coma-separated file with columns corresponding respectively to SearchId and PropertyId (see the "write_submission" function in https://github.com/benhamner/ExpediaPersonalizedSortCompetition/blob/master/PythonBenchmark/data_io.py). An example of a submission file is here: https://www.kaggle.com/c/expedia-personalized-sort/download/basicPythonBenchmark.zip .

\n

I hope it helps.

\n

Cheers,Adam

\n

 

\n

 

\n

X

", "votes": null, "replies": [] }, { "id": 30153, "author_name": "domcastro", "author_url": "", "post_date": "09/04/2013 14:44:53", "content": "

Ok thanks. A little further forward. So the "position" field in the training set is redundant ? and there are no ranks  in the training set ? We will run models on Train to predict Click_bool and boolean bool, make predictions on the test set and then rank these predictions and use the properyid field of the predictions as the submission format? X

", "votes": null, "replies": [] }, { "id": 30154, "author_name": "adamwoz", "author_url": "", "post_date": "09/04/2013 15:00:30", "content": "

Hi,

\n

'So the "position" field in the training set is redundant ?' - I think so. In the case when the order of records in the file per each impression does not correspond to the value in the position column, then the latter information takes precedence.

\n

"and there are no ranks  in the training set ? " - no, there are not ranks, just the click_bool and purchase_bool columns.

\n

"We will run models on Train to predict Click_bool and boolean bool, make predictions on the test set and then rank these predictions and use the properyid field of the predictions as the submission format? "  - yes

\n

 

\n

Adam

\n

 

\n

 

\n

 

", "votes": null, "replies": [] }, { "id": 30157, "author_name": "domcastro", "author_url": "", "post_date": "09/04/2013 15:02:05", "content": "

thanks

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "30146": "", "30149": "", "30153": "", "30154": "", "30157": "" }, "source": "meta" }