![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F83267%2F03e7f381efdce85fb0ae98931166a7cd%2Ffeatures_to_generate.png?generation=1669901142143022&alt=media\")
I think sharing the number of candidates and recall score can help us.
\nMine is: (Do not confused with this below recall and LB recall. This is just calculated by |candidates ⋂ targets| / |targets|. I used @cdeotte 's codes for calculating recall.)
# of candidates: 67.17\n=============\nclicks recall = 0.5342\ncarts recall = 0.4374\norders recall = 0.6677\n=============\nOverall Recall = 0.5852\n=============\nHow about yours? I really want to know.
", "messages": [ { "id": "2051113", "postDate": "12/01/2022 08:03:18", "content": "I think sharing the number of candidates and recall score can help us.
\nMine is: (Do not confused with this below recall and LB recall. This is just calculated by |candidates ⋂ targets| / |targets|. I used @cdeotte 's codes for calculating recall.)
# of candidates: 67.17\n=============\nclicks recall = 0.5342\ncarts recall = 0.4374\norders recall = 0.6677\n=============\nOverall Recall = 0.5852\n=============\nHow about yours? I really want to know.
", "rawMarkdown": "I think sharing the number of candidates and recall score can help us.\n\nMine is: (Do not confused with this below recall and LB recall. This is just calculated by `|candidates ⋂ targets| / |targets|`. I used @cdeotte 's codes for calculating recall.)\n\n```\n# of candidates: 67.17\n=============\nclicks recall = 0.5342\ncarts recall = 0.4374\norders recall = 0.6677\n=============\nOverall Recall = 0.5852\n=============\n```\n\nHow about yours? I really want to know.", "votes": null }, { "id": "2051143", "postDate": "12/01/2022 08:42:00", "content": "Hey @baekseungyun! That is an interesting question 🙂 I am generating on average 85.86 candidates per session. This is how I got to my current position on the LB.
\nHope this helps! 🙌 Hope others will reply as well, curious about what other people are doing 🙂
", "rawMarkdown": "Hey @baekseungyun! That is an interesting question 🙂 I am generating on average 85.86 candidates per session. This is how I got to my current position on the LB.\n\nHope this helps! 🙌 Hope others will reply as well, curious about what other people are doing 🙂", "votes": null }, { "id": "2051154", "postDate": "12/01/2022 08:54:50", "content": "Oh, that's smaller than I think! Thanks you for sharing.
", "rawMarkdown": "Oh, that's smaller than I think! Thanks you for sharing.", "votes": null }, { "id": "2051158", "postDate": "12/01/2022 08:57:40", "content": "np 🙂 just to add a bit more, I am only generating candidates from \"strong sources\" so far (the co-visitation matrices, popularity-based candidates)
\nI would really love to hear what people at the top of the LB are doing 😄 I am thinking they maybe be training with any more candidates per session but I might be wrong!
", "rawMarkdown": "np 🙂 just to add a bit more, I am only generating candidates from \"strong sources\" so far (the co-visitation matrices, popularity-based candidates)\n\nI would really love to hear what people at the top of the LB are doing 😄 I am thinking they maybe be training with any more candidates per session but I might be wrong!", "votes": null }, { "id": "2051451", "postDate": "12/01/2022 12:28:00", "content": "I have built a model with almost 100 candidates, but it has not worked well.
\nCandidates are the same as @radek1
\n", "rawMarkdown": "I have built a model with almost 100 candidates, but it has not worked well.\nCandidates are the same as @radek1 \n\n> of candidates: 96\nclicks recall =clicks: 0.5091\ncarts recall = 0.4105\norders recall = 0.6499\nOverall Recall = 0.5640", "votes": null }, { "id": "2051510", "postDate": "12/01/2022 13:28:08", "content": "of candidates: 96
\n
\n clicks recall =clicks: 0.5091
\n carts recall = 0.4105
\n orders recall = 0.6499
\n Overall Recall = 0.5640
I suspect we need to study this comment by @cdeotte @dehokanta and generate a bunch of better candidates 🙂
\n
I am not sure where I found this on Kaggle but here is another one that is pure gold:
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F83267%2F5c4be47de8d468e97e66b40bf9743354%2Freranking_by_Chris.png?generation=1669901197754319&alt=media\")
Anyhow, if you got to 0.5640 which would be around 0.576 on the LB by using covisitation matrix features and a ranking model, I believe you are off to a great start 🙂
\nCrossing my fingers for the next set of good candidates you will be able to find! 🙂 (in that regard, we are in the same boat! 🙂)
", "rawMarkdown": "I suspect we need to study this comment by @cdeotte @dehokanta and generate a bunch of better candidates 🙂\n\n\n\nI am not sure where I found this on Kaggle but here is another one that is pure gold:\n\n\n\nAnyhow, if you got to 0.5640 which would be around 0.576 on the LB by using covisitation matrix features and a ranking model, I believe you are off to a great start 🙂\n\nCrossing my fingers for the next set of good candidates you will be able to find! 🙂 (in that regard, we are in the same boat! 🙂)", "votes": null }, { "id": "2052097", "postDate": "12/01/2022 23:02:30", "content": "Thanks for your kind comments.
\nI did not see the second @cdeotte comment, so that helps.
Hey @baekseungyun! That is an interesting question 🙂 I am generating on average 85.86 candidates per session. This is how I got to my current position on the LB.
\nHope this helps! 🙌 Hope others will reply as well, curious about what other people are doing 🙂
", "votes": null, "replies": [ { "id": 2051154, "author_name": "baekseungyun", "author_url": "", "post_date": "12/01/2022 08:54:50", "content": "Oh, that's smaller than I think! Thanks you for sharing.
", "votes": null, "replies": [] }, { "id": 2051158, "author_name": "radek1", "author_url": "", "post_date": "12/01/2022 08:57:40", "content": "np 🙂 just to add a bit more, I am only generating candidates from \"strong sources\" so far (the co-visitation matrices, popularity-based candidates)
\nI would really love to hear what people at the top of the LB are doing 😄 I am thinking they maybe be training with any more candidates per session but I might be wrong!
", "votes": null, "replies": [] } ] }, { "id": 2051451, "author_name": "dehokanta", "author_url": "", "post_date": "12/01/2022 12:28:00", "content": "I have built a model with almost 100 candidates, but it has not worked well.
\nCandidates are the same as @radek1
\n", "votes": null, "replies": [ { "id": 2051510, "author_name": "radek1", "author_url": "", "post_date": "12/01/2022 13:28:08", "content": "of candidates: 96
\n
\n clicks recall =clicks: 0.5091
\n carts recall = 0.4105
\n orders recall = 0.6499
\n Overall Recall = 0.5640
I suspect we need to study this comment by @cdeotte @dehokanta and generate a bunch of better candidates 🙂
\n
I am not sure where I found this on Kaggle but here is another one that is pure gold:
\n
Anyhow, if you got to 0.5640 which would be around 0.576 on the LB by using covisitation matrix features and a ranking model, I believe you are off to a great start 🙂
\nCrossing my fingers for the next set of good candidates you will be able to find! 🙂 (in that regard, we are in the same boat! 🙂)
", "votes": null, "replies": [] }, { "id": 2052097, "author_name": "dehokanta", "author_url": "", "post_date": "12/01/2022 23:02:30", "content": "Thanks for your kind comments.
\nI did not see the second @cdeotte comment, so that helps.