{ "id": 73091, "title": "Sample_submission should not be classified ...", "url": "/competitions/ga-customer-revenue-prediction/discussion/73091", "author_name": "", "post_date": "2018-11-29T18:19:52.772995Z", "votes": -3, "comment_count": 7, "views": 0, "content": "

For this contest to maintain the minimum of dignity in this sad end, I suggest not rewarding submissions that contains only zeros (ie, sample_submission)\nNo money, no medals and no points ...

", "messages": [ { "id": "430047", "postDate": "11/29/2018 18:19:52", "content": "

For this contest to maintain the minimum of dignity in this sad end, I suggest not rewarding submissions that contains only zeros (ie, sample_submission)\nNo money, no medals and no points ...

", "rawMarkdown": "For this contest to maintain the minimum of dignity in this sad end, I suggest not rewarding submissions that contains only zeros (ie, sample_submission)\nNo money, no medals and no points ...", "votes": null }, { "id": "430209", "postDate": "11/30/2018 02:57:05", "content": "

Well, I have seen competiton where all 0 sample submission gets top 15. We would hope this is not the case here, but lottary tickets are free.

", "rawMarkdown": "Well, I have seen [competiton](https://www.kaggle.com/c/msk-redefining-cancer-treatment/leaderboard) where all 0 sample submission gets top 15. We would hope this is not the case here, but lottary tickets are free.", "votes": null }, { "id": "430437", "postDate": "11/30/2018 11:36:51", "content": "

Untill now, about 80 teams (10% of all teams) have an identical 2.1138 score, which is likely the all-zeros submission. It might get a high place, but it's highly un-likely that it would be in the money, as there are enough teams (as myself) who did extract usefull features from the data, and would get at least slightly better results than all-zeros.

", "rawMarkdown": "Untill now, about 80 teams (10% of all teams) have an identical 2.1138 score, which is likely the all-zeros submission. It might get a high place, but it's highly un-likely that it would be in the money, as there are enough teams (as myself) who did extract usefull features from the data, and would get at least slightly better results than all-zeros.", "votes": null }, { "id": "430441", "postDate": "11/30/2018 11:41:48", "content": "

Indeed, this is all about getting bit better than only 0's :)

", "rawMarkdown": "Indeed, this is all about getting bit better than only 0's :)", "votes": null }, { "id": "430443", "postDate": "11/30/2018 11:44:46", "content": "

Maybe final result is all 0！

", "rawMarkdown": "Maybe final result is all 0！", "votes": null }, { "id": "430505", "postDate": "11/30/2018 13:37:48", "content": "", "rawMarkdown": "", "votes": null }, { "id": "430506", "postDate": "11/30/2018 13:40:02", "content": "

I believe the conditions below are not bad, however, the output is filled with zeros.\nhttps://www.kaggle.com/tchristophern/base-model-v2-with-with-full-features

\n\n

It seems that the problem is ill-defined because the likelihood of repeated transactions is so low that the all-zero strategy works well. The key features of my kernel are as follows;

\n\n
    \n
  1. Dataset is divided in such a way that the revenue in Dec, 2018 -Jan, 2019 is predicted appropriately.\n(dividing the data into\n X[23 months training period] <-(1.5 months )->y [2 months validation period]\nand rolling the window in an appropriate manner)
    2. \n
  2. Xgboost with cross validation and hyper-parameter optimization is implemented
    3. \n
  3. Data is aggregated by fullVisitorId\nto pick up the latest data for categorical data\nto sum numerical data
    4. \n
  4. Categorical data is label-encoded
    5. \n
  5. Date is converted into \"the last date of the training period - date\"
    6. \n
", "rawMarkdown": "I believe the conditions below are not bad, however, the output is filled with zeros.\nhttps://www.kaggle.com/tchristophern/base-model-v2-with-with-full-features\n\nIt seems that the problem is ill-defined because the likelihood of repeated transactions is so low that the all-zero strategy works well. The key features of my kernel are as follows;\n\n1. Dataset is divided in such a way that the revenue in Dec, 2018 -Jan, 2019 is predicted appropriately.\n (dividing the data into\n X[23 months training period] <-(1.5 months )->y [2 months validation period]\n and rolling the window in an appropriate manner)\n2. Xgboost with cross validation and hyper-parameter optimization is implemented\n3. Data is aggregated by fullVisitorId\n to pick up the latest data for categorical data\n to sum numerical data\n4. Categorical data is label-encoded\n5. Date is converted into \"the last date of the training period - date\"", "votes": null }, { "id": "430665", "postDate": "11/30/2018 20:05:14", "content": "

If you did not like this post, you probably will like this:

\n\n

https://www.kaggle.com/paulorzp/one-function-in-r-script-to-predict-stage-2

", "rawMarkdown": "If you did not like this post, you probably will like this:\n\nhttps://www.kaggle.com/paulorzp/one-function-in-r-script-to-predict-stage-2", "votes": null } ], "comments": [ { "id": 430209, "author_name": "ryanzhang", "author_url": "", "post_date": "11/30/2018 02:57:05", "content": "

Well, I have seen competiton where all 0 sample submission gets top 15. We would hope this is not the case here, but lottary tickets are free.

", "votes": null, "replies": [] }, { "id": 430437, "author_name": "aryerez", "author_url": "", "post_date": "11/30/2018 11:36:51", "content": "

Untill now, about 80 teams (10% of all teams) have an identical 2.1138 score, which is likely the all-zeros submission. It might get a high place, but it's highly un-likely that it would be in the money, as there are enough teams (as myself) who did extract usefull features from the data, and would get at least slightly better results than all-zeros.

", "votes": null, "replies": [ { "id": 430441, "author_name": "mikematusiak", "author_url": "", "post_date": "11/30/2018 11:41:48", "content": "

Indeed, this is all about getting bit better than only 0's :)

", "votes": null, "replies": [] } ] }, { "id": 430443, "author_name": "zjp2origin", "author_url": "", "post_date": "11/30/2018 11:44:46", "content": "

Maybe final result is all 0！

", "votes": null, "replies": [] }, { "id": 430505, "author_name": "tchristophern", "author_url": "", "post_date": "11/30/2018 13:37:48", "content": "", "votes": null, "replies": [] }, { "id": 430506, "author_name": "tchristophern", "author_url": "", "post_date": "11/30/2018 13:40:02", "content": "

I believe the conditions below are not bad, however, the output is filled with zeros.\nhttps://www.kaggle.com/tchristophern/base-model-v2-with-with-full-features

\n\n

It seems that the problem is ill-defined because the likelihood of repeated transactions is so low that the all-zero strategy works well. The key features of my kernel are as follows;

\n\n
    \n
  1. Dataset is divided in such a way that the revenue in Dec, 2018 -Jan, 2019 is predicted appropriately.\n(dividing the data into\n X[23 months training period] <-(1.5 months )->y [2 months validation period]\nand rolling the window in an appropriate manner)
    2. \n
  2. Xgboost with cross validation and hyper-parameter optimization is implemented
    3. \n
  3. Data is aggregated by fullVisitorId\nto pick up the latest data for categorical data\nto sum numerical data
    4. \n
  4. Categorical data is label-encoded
    5. \n
  5. Date is converted into \"the last date of the training period - date\"
    6. \n
", "votes": null, "replies": [] }, { "id": 430665, "author_name": "paulorzp", "author_url": "", "post_date": "11/30/2018 20:05:14", "content": "

If you did not like this post, you probably will like this:

\n\n

https://www.kaggle.com/paulorzp/one-function-in-r-script-to-predict-stage-2

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "430047": "For this contest to maintain the minimum of dignity in this sad end, I suggest not rewarding submissions that contains only zeros (ie, sample_submission)\nNo money, no medals and no points ...", "430209": "Well, I have seen [competiton](https://www.kaggle.com/c/msk-redefining-cancer-treatment/leaderboard) where all 0 sample submission gets top 15. We would hope this is not the case here, but lottary tickets are free.", "430437": "Untill now, about 80 teams (10% of all teams) have an identical 2.1138 score, which is likely the all-zeros submission. It might get a high place, but it's highly un-likely that it would be in the money, as there are enough teams (as myself) who did extract usefull features from the data, and would get at least slightly better results than all-zeros.", "430441": "Indeed, this is all about getting bit better than only 0's :)", "430443": "Maybe final result is all 0！", "430505": "", "430506": "I believe the conditions below are not bad, however, the output is filled with zeros.\nhttps://www.kaggle.com/tchristophern/base-model-v2-with-with-full-features\n\nIt seems that the problem is ill-defined because the likelihood of repeated transactions is so low that the all-zero strategy works well. The key features of my kernel are as follows;\n\n1. Dataset is divided in such a way that the revenue in Dec, 2018 -Jan, 2019 is predicted appropriately.\n (dividing the data into\n X[23 months training period] <-(1.5 months )->y [2 months validation period]\n and rolling the window in an appropriate manner)\n2. Xgboost with cross validation and hyper-parameter optimization is implemented\n3. Data is aggregated by fullVisitorId\n to pick up the latest data for categorical data\n to sum numerical data\n4. Categorical data is label-encoded\n5. Date is converted into \"the last date of the training period - date\"", "430665": "If you did not like this post, you probably will like this:\n\nhttps://www.kaggle.com/paulorzp/one-function-in-r-script-to-predict-stage-2" }, "source": "meta" }