{
  "id": 82057,
  "title": "22nd (some 10k feats)shareing some engineering and some trick(not used)",
  "url": "/competitions/elo-merchant-category-recommendation/writeups/grand-rookie-22nd-some-10k-feats-shareing-some-eng",
  "author_name": "",
  "post_date": "2019-02-27T03:23:29.660Z",
  "votes": 19,
  "comment_count": 25,
  "views": 0,
  "content": "<p>Fisrt of all congratulations to the winner, sec, thanks for kagglers sharing in the compete,I have learned a lot</p>\n\n<p>law: trust your local cv</p>\n\n<p>here something I want to share\ntrick!!!\nfinal_pred = binary_prediprob * -33 + pred_without_outlier * ( 1- binary_prediprob), as I see in previos post the 1st used, We use this trick too,but did not select as our final sub\neng!!!</p>\n\n<ol>\n<li>choose a good baseline\n  I take this kernel as my baseline (here --&gt; <a href=\"https://www.kaggle.com/roydatascience/elo-merchant-recommendation-fathers-day-specials),tks\">https://www.kaggle.com/roydatascience/elo-merchant-recommendation-fathers-day-specials),tks</a> to Ashish Gupta</li>\n<li>repreoduce feature with baseline feats( except holidays )\n  select subsets to reprodeuce feats mainly through → lag &gt; threadhold (-3 ,-6) as subset condiction, reproduce feats we did once such as month_diff ---&gt; lag_3_month_diff,lag_6_month_diff, this give me at least 10thousand improvements</li>\n<li>abount merchants.csv,\n   [user preference], find user preference merchant(id) , and use it, use its features which is calcaulate from merchants.csv</li>\n<li><p>ratios\n   basically new/his, feature calcu by new_merchant div by same feature calcu in his</p></li>\n<li><p>word2vec\n   through what-and-when-and-where format to build senctence</p></li>\n<li>timestamp\n use feats from use timestamp</li>\n</ol>\n\n<p>apart from modeling, I think most improtant of all, do more eda, undersand well feats , and What is the company's business? \n10k feats\n 1.\nnew_mount_sum/his\nnew_date_max_tonow/his\nnew_date_max_tonow/his\n 2.\npref_merchant_purchase_amount_percard</p>\n\n<p>pref_merchants_avg_sales_lag3_ratio</p>\n\n<p>new_pref_merchants_avg_sales_lag3_ratio</p>\n\n<p>pref_merchants_avg_sales_lag3_div_6</p>\n\n<p>new_pref_merchants_avg_sales_lag3_div_6</p>\n\n<p>new_pref_merchants_avg_sales_lag3</p>\n\n<p>pref_merchants_avg_sales_lag3</p>\n\n<p>new_pref_merchants_numerical_1</p>\n\n<p>new_pref_merchants_numerical_1_ratio</p>\n\n<p>pref_merchants_numerical_1</p>\n\n<p>pref_merchants_numerical_1_ratio</p>\n\n<p>pref_merchants_most_recent_sales_range_ratio</p>\n\n<p>new_pref_merchants_most_recent_sales_range</p>\n\n<p>new_pref_merchants_most_recent_sales_range_ratio</p>\n\n<p>pref_merchants_most_recent_sales_range</p>\n\n<p>new_pref_merchants_category_4</p>\n\n<p>pref_merchants_category_4</p>",
  "messages": [
    {
      "id": "479214",
      "postDate": "02/27/2019 02:51:47",
      "content": "<p>Fisrt of all congratulations to the winner, sec, thanks for kagglers sharing in the compete,I have learned a lot</p>\n\n<p>law: trust your local cv</p>\n\n<p>here something I want to share\ntrick!!!\nfinal_pred = binary_prediprob * -33 + pred_without_outlier * ( 1- binary_prediprob), as I see in previos post the 1st used, We use this trick too,but did not select as our final sub\neng!!!</p>\n\n<ol>\n<li>choose a good baseline\n  I take this kernel as my baseline (here --&gt; <a href=\"https://www.kaggle.com/roydatascience/elo-merchant-recommendation-fathers-day-specials),tks\">https://www.kaggle.com/roydatascience/elo-merchant-recommendation-fathers-day-specials),tks</a> to Ashish Gupta</li>\n<li>repreoduce feature with baseline feats( except holidays )\n  select subsets to reprodeuce feats mainly through → lag &gt; threadhold (-3 ,-6) as subset condiction, reproduce feats we did once such as month_diff ---&gt; lag_3_month_diff,lag_6_month_diff, this give me at least 10thousand improvements</li>\n<li>abount merchants.csv,\n   [user preference], find user preference merchant(id) , and use it, use its features which is calcaulate from merchants.csv</li>\n<li><p>ratios\n   basically new/his, feature calcu by new_merchant div by same feature calcu in his</p></li>\n<li><p>word2vec\n   through what-and-when-and-where format to build senctence</p></li>\n<li>timestamp\n use feats from use timestamp</li>\n</ol>\n\n<p>apart from modeling, I think most improtant of all, do more eda, undersand well feats , and What is the company's business? \n10k feats\n 1.\nnew_mount_sum/his\nnew_date_max_tonow/his\nnew_date_max_tonow/his\n 2.\npref_merchant_purchase_amount_percard</p>\n\n<p>pref_merchants_avg_sales_lag3_ratio</p>\n\n<p>new_pref_merchants_avg_sales_lag3_ratio</p>\n\n<p>pref_merchants_avg_sales_lag3_div_6</p>\n\n<p>new_pref_merchants_avg_sales_lag3_div_6</p>\n\n<p>new_pref_merchants_avg_sales_lag3</p>\n\n<p>pref_merchants_avg_sales_lag3</p>\n\n<p>new_pref_merchants_numerical_1</p>\n\n<p>new_pref_merchants_numerical_1_ratio</p>\n\n<p>pref_merchants_numerical_1</p>\n\n<p>pref_merchants_numerical_1_ratio</p>\n\n<p>pref_merchants_most_recent_sales_range_ratio</p>\n\n<p>new_pref_merchants_most_recent_sales_range</p>\n\n<p>new_pref_merchants_most_recent_sales_range_ratio</p>\n\n<p>pref_merchants_most_recent_sales_range</p>\n\n<p>new_pref_merchants_category_4</p>\n\n<p>pref_merchants_category_4</p>",
      "rawMarkdown": "Fisrt of all congratulations to the winner, sec, thanks for kagglers sharing in the compete,I have learned a lot\n\nlaw: trust your local cv\n\nhere something I want to share\ntrick!!!\nfinal_pred = binary_prediprob * -33 + pred_without_outlier * ( 1- binary_prediprob), as I see in previos post the 1st used, We use this trick too,but did not select as our final sub\neng!!!\n\n 1.  choose a good baseline\n      I take this kernel as my baseline (here --&gt; https://www.kaggle.com/roydatascience/elo-merchant-recommendation-fathers-day-specials),tks to Ashish Gupta\n 2. repreoduce feature with baseline feats( except holidays )\n      select subsets to reprodeuce feats mainly through → lag &gt; threadhold (-3 ,-6) as subset condiction, reproduce feats we did once such as month_diff ---&gt; lag_3_month_diff,lag_6_month_diff, this give me at least 10thousand improvements\n 3. abount merchants.csv,\n       [user preference], find user preference merchant(id) , and use it, use its features which is calcaulate from merchants.csv\n 4. ratios\n       basically new/his, feature calcu by new_merchant div by same feature calcu in his\n  \n 5. word2vec\n       through what-and-when-and-where format to build senctence\n 6. timestamp\n     use feats from use timestamp\n\napart from modeling, I think most improtant of all, do more eda, undersand well feats , and What is the company's business? \n10k feats\n 1.\nnew_mount_sum/his\nnew_date_max_tonow/his\nnew_date_max_tonow/his\n 2.\npref_merchant_purchase_amount_percard\n\npref_merchants_avg_sales_lag3_ratio\n\nnew_pref_merchants_avg_sales_lag3_ratio\n\npref_merchants_avg_sales_lag3_div_6\n\nnew_pref_merchants_avg_sales_lag3_div_6\n\nnew_pref_merchants_avg_sales_lag3\n\npref_merchants_avg_sales_lag3\n\nnew_pref_merchants_numerical_1\n\nnew_pref_merchants_numerical_1_ratio\n\npref_merchants_numerical_1\n\npref_merchants_numerical_1_ratio\n\npref_merchants_most_recent_sales_range_ratio\n\nnew_pref_merchants_most_recent_sales_range\n\nnew_pref_merchants_most_recent_sales_range_ratio\n\npref_merchants_most_recent_sales_range\n\nnew_pref_merchants_category_4\n\npref_merchants_category_4",
      "votes": null
    },
    {
      "id": "479230",
      "postDate": "02/27/2019 03:03:01",
      "content": "<p>Thanks.</p>",
      "rawMarkdown": "Thanks.",
      "votes": null
    },
    {
      "id": "479236",
      "postDate": "02/27/2019 03:05:42",
      "content": "<p>congratulations and thanks for sharing!</p>",
      "rawMarkdown": "congratulations and thanks for sharing!",
      "votes": null
    },
    {
      "id": "479242",
      "postDate": "02/27/2019 03:07:28",
      "content": "<p>TOP 50th feats</p>\n\n<ol>\n<li>user_his-new_transactions_purchase_amount_mean</li>\n<li>user_new_transactions_purchase_amount_max</li>\n<li>card_merchant_pay_hist_card_pay_count_mean</li>\n<li>user_new_transactions_purchase_amount_mean</li>\n<li>user_new-his_transactions_purchase_timestamp2</li>\n<li>user_transactions_user_merchant_purchase_timestamp_min_skew</li>\n<li>user_his_transactions_merchant_purchase_month_frequence</li>\n<li>liu_new_hist_duration_max</li>\n<li>user_his_transactions_timedelta_skew</li>\n<li>liu_new_hist_amount_month_ratio_max</li>\n<li>user_new-his_transactions_purchase_timestamp1</li>\n<li>purchase_amount5</li>\n<li>Lag_3purchase_amount_min</li>\n<li>purchase_amount0</li>\n<li>user_his/new_transactions_purchase_amount_std</li>\n<li>purchase_amount2</li>\n<li>merchant_category_id0</li>\n<li>Lag_3purchase_amount_std</li>\n<li>user_his_transactions_month_lag_0_ratio</li>\n<li>liu_ratio_2</li>\n<li>month_lag6</li>\n<li>purchase_min</li>\n<li>user_his_transactions_timedelta_std</li>\n<li>user_history_transactions_purchase_amount_skew</li>\n<li>Lag_3purchase_amount_mean</li>\n<li>user_transactions_user_merchant_purchase_timedelta_skew</li>\n<li>liu_ratio_4</li>\n<li>user_new_transactions_purchase_amount_min</li>\n<li>user_new_transactions_month_lag_mean</li>\n<li>merchant_category_id1</li>\n<li>user_purchase_count_in_same_merchant_skew</li>\n<li>subsector_id8</li>\n<li>user_new_transactions_purchase_timestamp_mean</li>\n<li>purchase_amount4</li>\n<li>subsector_id2</li>\n<li>user_transactions_merchant_purchase_month_frequence</li>\n<li>subsector_id0</li>\n<li>month_lag5</li>\n<li>month_lag9</li>\n<li>merchant_category_id6</li>\n<li>city_id8</li>\n<li>purchase_max</li>\n<li>user_category_all_purchase_timestamp_min_std</li>\n<li>Lag_3weekend_mean</li>\n<li>user_his/new_transactions_purchase_amount_mean</li>\n<li>user_transaction_timedelta_mean_before_Black_Friday_2047</li>\n<li>user_authorized_flagN_transactions_ratio</li>\n<li>card_merchant_pay_hist_card_pay_mean_min</li>\n<li>city_id2</li>\n<li>purchase_amount8</li>\n</ol>",
      "rawMarkdown": "TOP 50th feats\n\n 1. user_his-new_transactions_purchase_amount_mean\n 2. user_new_transactions_purchase_amount_max\n 3. card_merchant_pay_hist_card_pay_count_mean\n 4. user_new_transactions_purchase_amount_mean\n 5. user_new-his_transactions_purchase_timestamp2\n 6. user_transactions_user_merchant_purchase_timestamp_min_skew\n 7. user_his_transactions_merchant_purchase_month_frequence\n 8. liu_new_hist_duration_max\n 9. user_his_transactions_timedelta_skew\n 10. liu_new_hist_amount_month_ratio_max\n 11. user_new-his_transactions_purchase_timestamp1\n 12. purchase_amount5\n 13. Lag_3purchase_amount_min\n 14. purchase_amount0\n 15. user_his/new_transactions_purchase_amount_std\n 16. purchase_amount2\n 17. merchant_category_id0\n 18. Lag_3purchase_amount_std\n 19. user_his_transactions_month_lag_0_ratio\n 29. liu_ratio_2\n 21. month_lag6\n 22. purchase_min\n 23. user_his_transactions_timedelta_std\n 24. user_history_transactions_purchase_amount_skew\n 25. Lag_3purchase_amount_mean\n 26. user_transactions_user_merchant_purchase_timedelta_skew\n 27. liu_ratio_4\n 28. user_new_transactions_purchase_amount_min\n 29. user_new_transactions_month_lag_mean\n 30. merchant_category_id1\n 31. user_purchase_count_in_same_merchant_skew\n 32. subsector_id8\n 33. user_new_transactions_purchase_timestamp_mean\n 34. purchase_amount4\n 35. subsector_id2\n 36. user_transactions_merchant_purchase_month_frequence\n 37. subsector_id0\n 38. month_lag5\n 39. month_lag9\n 40. merchant_category_id6\n 41. city_id8\n 42. purchase_max\n 43. user_category_all_purchase_timestamp_min_std\n 44. Lag_3weekend_mean\n 45. user_his/new_transactions_purchase_amount_mean\n 46. user_transaction_timedelta_mean_before_Black_Friday_2047\n 47. user_authorized_flagN_transactions_ratio\n 48. card_merchant_pay_hist_card_pay_mean_min\n 49. city_id2\n 50. purchase_amount8",
      "votes": null
    },
    {
      "id": "479285",
      "postDate": "02/27/2019 03:41:05",
      "content": "<p>thanks</p>",
      "rawMarkdown": "thanks",
      "votes": null
    },
    {
      "id": "479287",
      "postDate": "02/27/2019 03:41:25",
      "content": "<p>thank you</p>",
      "rawMarkdown": "thank you",
      "votes": null
    },
    {
      "id": "479289",
      "postDate": "02/27/2019 03:44:52",
      "content": "<p>thanks, great work. i have a question.  how much did your embedding give u cv/lb boost.  </p>",
      "rawMarkdown": "thanks, great work. i have a question.  how much did your embedding give u cv/lb boost.",
      "votes": null
    },
    {
      "id": "479300",
      "postDate": "02/27/2019 03:56:37",
      "content": "<p>w2v 2 thousand score </p>",
      "rawMarkdown": "w2v 2 thousand score",
      "votes": null
    },
    {
      "id": "479310",
      "postDate": "02/27/2019 04:04:08",
      "content": "<p>oh oh!! In my teamates model. 5k it goes, </p>",
      "rawMarkdown": "oh oh!! In my teamates model. 5k it goes,",
      "votes": null
    },
    {
      "id": "479316",
      "postDate": "02/27/2019 04:12:07",
      "content": "<p>Maybe depends on your feature sets .In fact,w2v  about 5k boost</p>",
      "rawMarkdown": "Maybe depends on your feature sets .In fact,w2v  about 5k boost",
      "votes": null
    },
    {
      "id": "479323",
      "postDate": "02/27/2019 04:19:31",
      "content": "<p>very good result, i should learn how you use this.</p>",
      "rawMarkdown": "very good result, i should learn how you use this.",
      "votes": null
    },
    {
      "id": "479371",
      "postDate": "02/27/2019 05:12:37",
      "content": "<p>Congrats and thanks for sharing. </p>\n\n<p>Here is my post on feature selection techniques <a href=\"https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/82052\">https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/82052</a></p>",
      "rawMarkdown": "Congrats and thanks for sharing. \n\nHere is my post on feature selection techniques https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/82052",
      "votes": null
    },
    {
      "id": "479533",
      "postDate": "02/27/2019 08:19:50",
      "content": "<p>Congratulations and thanks.</p>",
      "rawMarkdown": "Congratulations and thanks.",
      "votes": null
    },
    {
      "id": "479538",
      "postDate": "02/27/2019 08:22:55",
      "content": "<p>I am a newbie to Kaggle, and could you please share how to do <code>word2vec</code> on what-when-where sentences? and what is the input and output? As one <code>card-id</code> may has several what-when-where sentences, how to reduce the outputs into one vector?</p>",
      "rawMarkdown": "I am a newbie to Kaggle, and could you please share how to do `word2vec` on what-when-where sentences? and what is the input and output? As one `card-id` may has several what-when-where sentences, how to reduce the outputs into one vector?",
      "votes": null
    },
    {
      "id": "479554",
      "postDate": "02/27/2019 08:39:15",
      "content": "<p>a format is a single event, sentencs are contains of serveral events</p>",
      "rawMarkdown": "a format is a single event, sentencs are contains of serveral events",
      "votes": null
    },
    {
      "id": "479694",
      "postDate": "02/27/2019 10:47:37",
      "content": "<p>Congratulations @Orange!\nAwesome job you did there</p>",
      "rawMarkdown": "Congratulations @Orange!\nAwesome job you did there",
      "votes": null
    },
    {
      "id": "479787",
      "postDate": "02/27/2019 12:29:42",
      "content": "<p>thanks</p>",
      "rawMarkdown": "thanks",
      "votes": null
    },
    {
      "id": "480267",
      "postDate": "02/28/2019 02:02:22",
      "content": "<p>thanks! great tips!\nbut I didn't understand the following code.\nfinalpred = binaryprediprob * -33 + predwithoutoutlier * ( 1- binary_prediprob)</p>\n\n<p>could you explain its details please?\nex) specifically  what means ' -33 ' ? </p>",
      "rawMarkdown": "thanks! great tips!\nbut I didn't understand the following code.\nfinalpred = binaryprediprob * -33 + predwithoutoutlier * ( 1- binary_prediprob)\n\ncould you explain its details please?\nex) specifically  what means ' -33 ' ?",
      "votes": null
    },
    {
      "id": "480326",
      "postDate": "02/28/2019 03:42:58",
      "content": "<p>two parts ，outlier value multiply it‘s prob add inlier value multiply it's prob</p>",
      "rawMarkdown": "two parts ，outlier value multiply it‘s prob add inlier value multiply it's prob",
      "votes": null
    },
    {
      "id": "480338",
      "postDate": "02/28/2019 04:17:25",
      "content": "<p>great. i failed to get this work in this competition , although i did same like u according to your description. <br>\nmaybe devil is in details.</p>",
      "rawMarkdown": "great. i failed to get this work in this competition , although i did same like u according to your description.  \nmaybe devil is in details.",
      "votes": null
    },
    {
      "id": "481973",
      "postDate": "03/02/2019 05:54:11",
      "content": "<p>Congratulations First Medal!!! And Thanks for sharing Idea. Can you tell more about 'Reproduce Feature` ?? You did Subset dataset? or make a features with shift(-3)? ? ( I don't understand it haha...)</p>",
      "rawMarkdown": "Congratulations First Medal!!! And Thanks for sharing Idea. Can you tell more about 'Reproduce Feature` ?? You did Subset dataset? or make a features with shift(-3)? ? ( I don't understand it haha...)",
      "votes": null
    },
    {
      "id": "481974",
      "postDate": "03/02/2019 05:59:34",
      "content": "<p>it's something like slides window teck in time series, mainly throuh Lag variable's values to choose subsets to redo what has bend done on total dateset</p>",
      "rawMarkdown": "it's something like slides window teck in time series, mainly throuh Lag variable's values to choose subsets to redo what has bend done on total dateset",
      "votes": null
    },
    {
      "id": "481977",
      "postDate": "03/02/2019 06:06:59",
      "content": "<p>Thanks for answering. :D</p>",
      "rawMarkdown": "Thanks for answering. :D",
      "votes": null
    },
    {
      "id": "482137",
      "postDate": "03/02/2019 11:42:52",
      "content": "<p>Hi Orange. If possible, can you show the code in the Word2Vec part? If you feel burdened, you can refuse. Thanks</p>",
      "rawMarkdown": "Hi Orange. If possible, can you show the code in the Word2Vec part? If you feel burdened, you can refuse. Thanks",
      "votes": null
    },
    {
      "id": "482180",
      "postDate": "03/02/2019 13:07:36",
      "content": "<p>sure. but I am doing microsoft competition now,  not a good time now . I will make an kernel sharing it with you later</p>",
      "rawMarkdown": "sure. but I am doing microsoft competition now,  not a good time now . I will make an kernel sharing it with you later",
      "votes": null
    },
    {
      "id": "482192",
      "postDate": "03/02/2019 13:26:18",
      "content": "<p>Really Thank you!!! i wait it !!! </p>",
      "rawMarkdown": "Really Thank you!!! i wait it !!!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 479230,
      "author_name": "longyin2",
      "author_url": "",
      "post_date": "02/27/2019 03:03:01",
      "content": "<p>Thanks.</p>",
      "votes": null,
      "replies": [
        {
          "id": 479285,
          "author_name": "youwbpyr",
          "author_url": "",
          "post_date": "02/27/2019 03:41:05",
          "content": "<p>thanks</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 479236,
      "author_name": "jiazhuang",
      "author_url": "",
      "post_date": "02/27/2019 03:05:42",
      "content": "<p>congratulations and thanks for sharing!</p>",
      "votes": null,
      "replies": [
        {
          "id": 479287,
          "author_name": "youwbpyr",
          "author_url": "",
          "post_date": "02/27/2019 03:41:25",
          "content": "<p>thank you</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 479242,
      "author_name": "youwbpyr",
      "author_url": "",
      "post_date": "02/27/2019 03:07:28",
      "content": "<p>TOP 50th feats</p>\n\n<ol>\n<li>user_his-new_transactions_purchase_amount_mean</li>\n<li>user_new_transactions_purchase_amount_max</li>\n<li>card_merchant_pay_hist_card_pay_count_mean</li>\n<li>user_new_transactions_purchase_amount_mean</li>\n<li>user_new-his_transactions_purchase_timestamp2</li>\n<li>user_transactions_user_merchant_purchase_timestamp_min_skew</li>\n<li>user_his_transactions_merchant_purchase_month_frequence</li>\n<li>liu_new_hist_duration_max</li>\n<li>user_his_transactions_timedelta_skew</li>\n<li>liu_new_hist_amount_month_ratio_max</li>\n<li>user_new-his_transactions_purchase_timestamp1</li>\n<li>purchase_amount5</li>\n<li>Lag_3purchase_amount_min</li>\n<li>purchase_amount0</li>\n<li>user_his/new_transactions_purchase_amount_std</li>\n<li>purchase_amount2</li>\n<li>merchant_category_id0</li>\n<li>Lag_3purchase_amount_std</li>\n<li>user_his_transactions_month_lag_0_ratio</li>\n<li>liu_ratio_2</li>\n<li>month_lag6</li>\n<li>purchase_min</li>\n<li>user_his_transactions_timedelta_std</li>\n<li>user_history_transactions_purchase_amount_skew</li>\n<li>Lag_3purchase_amount_mean</li>\n<li>user_transactions_user_merchant_purchase_timedelta_skew</li>\n<li>liu_ratio_4</li>\n<li>user_new_transactions_purchase_amount_min</li>\n<li>user_new_transactions_month_lag_mean</li>\n<li>merchant_category_id1</li>\n<li>user_purchase_count_in_same_merchant_skew</li>\n<li>subsector_id8</li>\n<li>user_new_transactions_purchase_timestamp_mean</li>\n<li>purchase_amount4</li>\n<li>subsector_id2</li>\n<li>user_transactions_merchant_purchase_month_frequence</li>\n<li>subsector_id0</li>\n<li>month_lag5</li>\n<li>month_lag9</li>\n<li>merchant_category_id6</li>\n<li>city_id8</li>\n<li>purchase_max</li>\n<li>user_category_all_purchase_timestamp_min_std</li>\n<li>Lag_3weekend_mean</li>\n<li>user_his/new_transactions_purchase_amount_mean</li>\n<li>user_transaction_timedelta_mean_before_Black_Friday_2047</li>\n<li>user_authorized_flagN_transactions_ratio</li>\n<li>card_merchant_pay_hist_card_pay_mean_min</li>\n<li>city_id2</li>\n<li>purchase_amount8</li>\n</ol>",
      "votes": null,
      "replies": []
    },
    {
      "id": 479289,
      "author_name": "ddongjian0002",
      "author_url": "",
      "post_date": "02/27/2019 03:44:52",
      "content": "<p>thanks, great work. i have a question.  how much did your embedding give u cv/lb boost.  </p>",
      "votes": null,
      "replies": [
        {
          "id": 479300,
          "author_name": "youwbpyr",
          "author_url": "",
          "post_date": "02/27/2019 03:56:37",
          "content": "<p>w2v 2 thousand score </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 479310,
          "author_name": "youwbpyr",
          "author_url": "",
          "post_date": "02/27/2019 04:04:08",
          "content": "<p>oh oh!! In my teamates model. 5k it goes, </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 479316,
          "author_name": "process531",
          "author_url": "",
          "post_date": "02/27/2019 04:12:07",
          "content": "<p>Maybe depends on your feature sets .In fact,w2v  about 5k boost</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 479323,
          "author_name": "codlife",
          "author_url": "",
          "post_date": "02/27/2019 04:19:31",
          "content": "<p>very good result, i should learn how you use this.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 480338,
          "author_name": "ddongjian0002",
          "author_url": "",
          "post_date": "02/28/2019 04:17:25",
          "content": "<p>great. i failed to get this work in this competition , although i did same like u according to your description. <br>\nmaybe devil is in details.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 479371,
      "author_name": "sheriytm",
      "author_url": "",
      "post_date": "02/27/2019 05:12:37",
      "content": "<p>Congrats and thanks for sharing. </p>\n\n<p>Here is my post on feature selection techniques <a href=\"https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/82052\">https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/82052</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 479533,
      "author_name": "liewmanchoi",
      "author_url": "",
      "post_date": "02/27/2019 08:19:50",
      "content": "<p>Congratulations and thanks.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 479538,
      "author_name": "liewmanchoi",
      "author_url": "",
      "post_date": "02/27/2019 08:22:55",
      "content": "<p>I am a newbie to Kaggle, and could you please share how to do <code>word2vec</code> on what-when-where sentences? and what is the input and output? As one <code>card-id</code> may has several what-when-where sentences, how to reduce the outputs into one vector?</p>",
      "votes": null,
      "replies": [
        {
          "id": 479554,
          "author_name": "youwbpyr",
          "author_url": "",
          "post_date": "02/27/2019 08:39:15",
          "content": "<p>a format is a single event, sentencs are contains of serveral events</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 479694,
      "author_name": "tpthegreat",
      "author_url": "",
      "post_date": "02/27/2019 10:47:37",
      "content": "<p>Congratulations @Orange!\nAwesome job you did there</p>",
      "votes": null,
      "replies": [
        {
          "id": 479787,
          "author_name": "youwbpyr",
          "author_url": "",
          "post_date": "02/27/2019 12:29:42",
          "content": "<p>thanks</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 480267,
      "author_name": "mokmushroom",
      "author_url": "",
      "post_date": "02/28/2019 02:02:22",
      "content": "<p>thanks! great tips!\nbut I didn't understand the following code.\nfinalpred = binaryprediprob * -33 + predwithoutoutlier * ( 1- binary_prediprob)</p>\n\n<p>could you explain its details please?\nex) specifically  what means ' -33 ' ? </p>",
      "votes": null,
      "replies": [
        {
          "id": 480326,
          "author_name": "youwbpyr",
          "author_url": "",
          "post_date": "02/28/2019 03:42:58",
          "content": "<p>two parts ，outlier value multiply it‘s prob add inlier value multiply it's prob</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 481973,
      "author_name": "chocozzz",
      "author_url": "",
      "post_date": "03/02/2019 05:54:11",
      "content": "<p>Congratulations First Medal!!! And Thanks for sharing Idea. Can you tell more about 'Reproduce Feature` ?? You did Subset dataset? or make a features with shift(-3)? ? ( I don't understand it haha...)</p>",
      "votes": null,
      "replies": [
        {
          "id": 481974,
          "author_name": "youwbpyr",
          "author_url": "",
          "post_date": "03/02/2019 05:59:34",
          "content": "<p>it's something like slides window teck in time series, mainly throuh Lag variable's values to choose subsets to redo what has bend done on total dateset</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 481977,
          "author_name": "chocozzz",
          "author_url": "",
          "post_date": "03/02/2019 06:06:59",
          "content": "<p>Thanks for answering. :D</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 482137,
          "author_name": "chocozzz",
          "author_url": "",
          "post_date": "03/02/2019 11:42:52",
          "content": "<p>Hi Orange. If possible, can you show the code in the Word2Vec part? If you feel burdened, you can refuse. Thanks</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 482180,
          "author_name": "youwbpyr",
          "author_url": "",
          "post_date": "03/02/2019 13:07:36",
          "content": "<p>sure. but I am doing microsoft competition now,  not a good time now . I will make an kernel sharing it with you later</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 482192,
          "author_name": "chocozzz",
          "author_url": "",
          "post_date": "03/02/2019 13:26:18",
          "content": "<p>Really Thank you!!! i wait it !!! </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "479214": "Fisrt of all congratulations to the winner, sec, thanks for kagglers sharing in the compete,I have learned a lot\n\nlaw: trust your local cv\n\nhere something I want to share\ntrick!!!\nfinal_pred = binary_prediprob * -33 + pred_without_outlier * ( 1- binary_prediprob), as I see in previos post the 1st used, We use this trick too,but did not select as our final sub\neng!!!\n\n 1.  choose a good baseline\n      I take this kernel as my baseline (here --&gt; https://www.kaggle.com/roydatascience/elo-merchant-recommendation-fathers-day-specials),tks to Ashish Gupta\n 2. repreoduce feature with baseline feats( except holidays )\n      select subsets to reprodeuce feats mainly through → lag &gt; threadhold (-3 ,-6) as subset condiction, reproduce feats we did once such as month_diff ---&gt; lag_3_month_diff,lag_6_month_diff, this give me at least 10thousand improvements\n 3. abount merchants.csv,\n       [user preference], find user preference merchant(id) , and use it, use its features which is calcaulate from merchants.csv\n 4. ratios\n       basically new/his, feature calcu by new_merchant div by same feature calcu in his\n  \n 5. word2vec\n       through what-and-when-and-where format to build senctence\n 6. timestamp\n     use feats from use timestamp\n\napart from modeling, I think most improtant of all, do more eda, undersand well feats , and What is the company's business? \n10k feats\n 1.\nnew_mount_sum/his\nnew_date_max_tonow/his\nnew_date_max_tonow/his\n 2.\npref_merchant_purchase_amount_percard\n\npref_merchants_avg_sales_lag3_ratio\n\nnew_pref_merchants_avg_sales_lag3_ratio\n\npref_merchants_avg_sales_lag3_div_6\n\nnew_pref_merchants_avg_sales_lag3_div_6\n\nnew_pref_merchants_avg_sales_lag3\n\npref_merchants_avg_sales_lag3\n\nnew_pref_merchants_numerical_1\n\nnew_pref_merchants_numerical_1_ratio\n\npref_merchants_numerical_1\n\npref_merchants_numerical_1_ratio\n\npref_merchants_most_recent_sales_range_ratio\n\nnew_pref_merchants_most_recent_sales_range\n\nnew_pref_merchants_most_recent_sales_range_ratio\n\npref_merchants_most_recent_sales_range\n\nnew_pref_merchants_category_4\n\npref_merchants_category_4",
    "479230": "Thanks.",
    "479236": "congratulations and thanks for sharing!",
    "479242": "TOP 50th feats\n\n 1. user_his-new_transactions_purchase_amount_mean\n 2. user_new_transactions_purchase_amount_max\n 3. card_merchant_pay_hist_card_pay_count_mean\n 4. user_new_transactions_purchase_amount_mean\n 5. user_new-his_transactions_purchase_timestamp2\n 6. user_transactions_user_merchant_purchase_timestamp_min_skew\n 7. user_his_transactions_merchant_purchase_month_frequence\n 8. liu_new_hist_duration_max\n 9. user_his_transactions_timedelta_skew\n 10. liu_new_hist_amount_month_ratio_max\n 11. user_new-his_transactions_purchase_timestamp1\n 12. purchase_amount5\n 13. Lag_3purchase_amount_min\n 14. purchase_amount0\n 15. user_his/new_transactions_purchase_amount_std\n 16. purchase_amount2\n 17. merchant_category_id0\n 18. Lag_3purchase_amount_std\n 19. user_his_transactions_month_lag_0_ratio\n 29. liu_ratio_2\n 21. month_lag6\n 22. purchase_min\n 23. user_his_transactions_timedelta_std\n 24. user_history_transactions_purchase_amount_skew\n 25. Lag_3purchase_amount_mean\n 26. user_transactions_user_merchant_purchase_timedelta_skew\n 27. liu_ratio_4\n 28. user_new_transactions_purchase_amount_min\n 29. user_new_transactions_month_lag_mean\n 30. merchant_category_id1\n 31. user_purchase_count_in_same_merchant_skew\n 32. subsector_id8\n 33. user_new_transactions_purchase_timestamp_mean\n 34. purchase_amount4\n 35. subsector_id2\n 36. user_transactions_merchant_purchase_month_frequence\n 37. subsector_id0\n 38. month_lag5\n 39. month_lag9\n 40. merchant_category_id6\n 41. city_id8\n 42. purchase_max\n 43. user_category_all_purchase_timestamp_min_std\n 44. Lag_3weekend_mean\n 45. user_his/new_transactions_purchase_amount_mean\n 46. user_transaction_timedelta_mean_before_Black_Friday_2047\n 47. user_authorized_flagN_transactions_ratio\n 48. card_merchant_pay_hist_card_pay_mean_min\n 49. city_id2\n 50. purchase_amount8",
    "479285": "thanks",
    "479287": "thank you",
    "479289": "thanks, great work. i have a question.  how much did your embedding give u cv/lb boost.",
    "479300": "w2v 2 thousand score",
    "479310": "oh oh!! In my teamates model. 5k it goes,",
    "479316": "Maybe depends on your feature sets .In fact,w2v  about 5k boost",
    "479323": "very good result, i should learn how you use this.",
    "479371": "Congrats and thanks for sharing. \n\nHere is my post on feature selection techniques https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/82052",
    "479533": "Congratulations and thanks.",
    "479538": "I am a newbie to Kaggle, and could you please share how to do `word2vec` on what-when-where sentences? and what is the input and output? As one `card-id` may has several what-when-where sentences, how to reduce the outputs into one vector?",
    "479554": "a format is a single event, sentencs are contains of serveral events",
    "479694": "Congratulations @Orange!\nAwesome job you did there",
    "479787": "thanks",
    "480267": "thanks! great tips!\nbut I didn't understand the following code.\nfinalpred = binaryprediprob * -33 + predwithoutoutlier * ( 1- binary_prediprob)\n\ncould you explain its details please?\nex) specifically  what means ' -33 ' ?",
    "480326": "two parts ，outlier value multiply it‘s prob add inlier value multiply it's prob",
    "480338": "great. i failed to get this work in this competition , although i did same like u according to your description.  \nmaybe devil is in details.",
    "481973": "Congratulations First Medal!!! And Thanks for sharing Idea. Can you tell more about 'Reproduce Feature` ?? You did Subset dataset? or make a features with shift(-3)? ? ( I don't understand it haha...)",
    "481974": "it's something like slides window teck in time series, mainly throuh Lag variable's values to choose subsets to redo what has bend done on total dateset",
    "481977": "Thanks for answering. :D",
    "482137": "Hi Orange. If possible, can you show the code in the Word2Vec part? If you feel burdened, you can refuse. Thanks",
    "482180": "sure. but I am doing microsoft competition now,  not a good time now . I will make an kernel sharing it with you later",
    "482192": "Really Thank you!!! i wait it !!!"
  },
  "source": "meta"
}