{ "id": 3595, "title": "Most popular benchmark (different than my calculation)", "url": "/competitions/event-recommendation-engine-challenge/discussion/3595", "author_name": "", "post_date": "2013-01-14T22:25:26.703Z", "votes": null, "comment_count": 3, "views": 1529, "content": "

I'm looking at the "most popular" benchmark, and it appears different from my calculations.

\r\n

For example, in python, I do

\r\n

train = pd.read_csv(my_dir+"train.csv")

\r\n

like_counts = train.groupby('event').interested.sum()

\r\n

like_counts.sort()

\r\n

 

\r\n

The second most popular event appears to be 2529072432 with 94 likes.  If this is the 2nd most popular event, I wouldn't expect it to be listed later the 2nd for any user.  But we find this is frequently the case.  For instance, consider the line

\r\n

23465780,[955398943L, 602394192L, 2529072432L, 1600413013L, 3017074183L, 2414358105L, 3807695143L]

\r\n

where it is listed after an event with only 12 likes.

\r\n

This is just one example, but I'm finding lots of instances like this.

", "messages": [ { "id": "19306", "postDate": "01/14/2013 22:25:26", "content": "

I'm looking at the "most popular" benchmark, and it appears different from my calculations.

\r\n

For example, in python, I do

\r\n

train = pd.read_csv(my_dir+"train.csv")

\r\n

like_counts = train.groupby('event').interested.sum()

\r\n

like_counts.sort()

\r\n

 

\r\n

The second most popular event appears to be 2529072432 with 94 likes.  If this is the 2nd most popular event, I wouldn't expect it to be listed later the 2nd for any user.  But we find this is frequently the case.  For instance, consider the line

\r\n

23465780,[955398943L, 602394192L, 2529072432L, 1600413013L, 3017074183L, 2414358105L, 3807695143L]

\r\n

where it is listed after an event with only 12 likes.

\r\n

This is just one example, but I'm finding lots of instances like this.

", "rawMarkdown": "", "votes": null }, { "id": "19307", "postDate": "01/14/2013 22:30:32", "content": "

[quote=DanB;19306]

\r\n

I'm looking at the "most popular" benchmark, and it appears different from my calculations.

\r\n

For example, in python, I do

\r\n

train = pd.read_csv(my_dir+"train.csv")

\r\n

like_counts = train.groupby('event').interested.sum()

\r\n

like_counts.sort()

\r\n

 

\r\n

The second most popular event appears to be 2529072432 with 94 likes.  If this is the 2nd most popular event, I wouldn't expect it to be listed later the 2nd for any user.  But we find this is frequently the case.  For instance, consider the line

\r\n

23465780,[955398943L, 602394192L, 2529072432L, 1600413013L, 3017074183L, 2414358105L, 3807695143L]

\r\n

where it is listed after an event with only 12 likes.

\r\n

This is just one example, but I'm finding lots of instances like this.

\r\n

[/quote]The benchmark measures popularity by the number of people indicating that they are attending the event, not the number of people that are interested in an event.

", "rawMarkdown": "", "votes": null }, { "id": "19309", "postDate": "01/14/2013 22:41:51", "content": "

Thanks for the quick response... though that makes me want to verify I understand the metric correctly.  The rules say 

\r\n

"Events should be ordered from the ones in which you predict the user will be most interested to those in which the user will be least be interested."

\r\n

Can I infer from your response that "interested" in the rules refers to the probability of attending, rather than probability of saying they are interested?

\r\n

Thanks,

\r\n

Dan

\r\n


\r\n

", "rawMarkdown": "", "votes": null }, { "id": "19311", "postDate": "01/14/2013 23:25:15", "content": "

[quote=DanB;19309]

\r\n

Thanks for the quick response... though that makes me want to verify I understand the metric correctly.  The rules say 

\r\n

"Events should be ordered from the ones in which you predict the user will be most interested to those in which the user will be least be interested."

\r\n

Can I infer from your response that "interested" in the rules refers to the probability of attending, rather than probability of saying they are interested?

\r\n

Thanks,

\r\n

Dan[/quote]No, "interested" refers to whether the user marked that they were interested in it (as shown in train.csv).

", "rawMarkdown": "", "votes": null } ], "comments": [ { "id": 19307, "author_name": "benhamner", "author_url": "", "post_date": "01/14/2013 22:30:32", "content": "

[quote=DanB;19306]

\r\n

I'm looking at the "most popular" benchmark, and it appears different from my calculations.

\r\n

For example, in python, I do

\r\n

train = pd.read_csv(my_dir+"train.csv")

\r\n

like_counts = train.groupby('event').interested.sum()

\r\n

like_counts.sort()

\r\n

 

\r\n

The second most popular event appears to be 2529072432 with 94 likes.  If this is the 2nd most popular event, I wouldn't expect it to be listed later the 2nd for any user.  But we find this is frequently the case.  For instance, consider the line

\r\n

23465780,[955398943L, 602394192L, 2529072432L, 1600413013L, 3017074183L, 2414358105L, 3807695143L]

\r\n

where it is listed after an event with only 12 likes.

\r\n

This is just one example, but I'm finding lots of instances like this.

\r\n

[/quote]The benchmark measures popularity by the number of people indicating that they are attending the event, not the number of people that are interested in an event.

", "votes": null, "replies": [] }, { "id": 19309, "author_name": "dansbecker", "author_url": "", "post_date": "01/14/2013 22:41:51", "content": "

Thanks for the quick response... though that makes me want to verify I understand the metric correctly.  The rules say 

\r\n

"Events should be ordered from the ones in which you predict the user will be most interested to those in which the user will be least be interested."

\r\n

Can I infer from your response that "interested" in the rules refers to the probability of attending, rather than probability of saying they are interested?

\r\n

Thanks,

\r\n

Dan

\r\n


\r\n

", "votes": null, "replies": [] }, { "id": 19311, "author_name": "benhamner", "author_url": "", "post_date": "01/14/2013 23:25:15", "content": "

[quote=DanB;19309]

\r\n

Thanks for the quick response... though that makes me want to verify I understand the metric correctly.  The rules say 

\r\n

"Events should be ordered from the ones in which you predict the user will be most interested to those in which the user will be least be interested."

\r\n

Can I infer from your response that "interested" in the rules refers to the probability of attending, rather than probability of saying they are interested?

\r\n

Thanks,

\r\n

Dan[/quote]No, "interested" refers to whether the user marked that they were interested in it (as shown in train.csv).

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "19306": "", "19307": "", "19309": "", "19311": "" }, "source": "meta" }