Data
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1098915%2F7fe1db545b78d6ded19e76a3c8497052%2Fdata.jpg?generation=1661566215137056&alt=media\")
Model
First time be a solo winner, I must say there is luck in winning the competition.
\nMy best result is a heavy ensemble with LGB and NN.
\nData
\n
Model
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1098915%2F3c5fa34a54082283ce20f7b373d36524%2Fmodel.jpg?generation=1661562190912958&alt=media\")
for NN model, all data fillna(0) and using nn.utils.rnn.pack_padded_sequence to pad.
\nI have to look for my early stage model, sorry for small font size in figure.
\nupdate: I release a clean code at https://github.com/jxzly/Kaggle-American-Express-Default-Prediction-1st-solution.
\nnote: You may not be able to reproduce the best result due to random fluctuations.
First time be a solo winner, I must say there is luck in winning the competition.
\nMy best result is a heavy ensemble with LGB and NN.
\nData
\n
Model
for NN model, all data fillna(0) and using nn.utils.rnn.pack_padded_sequence to pad.
\nI have to look for my early stage model, sorry for small font size in figure.
\nupdate: I release a clean code at https://github.com/jxzly/Kaggle-American-Express-Default-Prediction-1st-solution.
\nnote: You may not be able to reproduce the best result due to random fluctuations.
A huge congratulation Daishu!
\n1st time solo winner is an achievement to be remembered. Mostly in this AMEX competition.
\nThough I can't even understand any solutions since I'm a beginner. Thank you for sharing yours.
Thanks for sharing and congratulations for your first solo winner! Some questions from my side:
\nHow user-based and month-based rank features are calculated? Simply applying scipy.stats.rankdata to raw data per user and per month before making the aggregations?
diff: Is it last value - nth value or nth value - nth-1 value?
LGB series oof: Let's call the number of customers in train set N_1 and the number of features N_2. The feature matrix you used to calculate LGB series oof has size (N_1, N_2*13), correct?
What are GreedyBins? Could you point some resource to take a look at it?
How was your iteration process for trying new features? I mean, how did you decided which features to try/add to your already selected features each time?
Answers from anybody are welcome :)
", "rawMarkdown": "Thanks for sharing and congratulations for your first solo winner! Some questions from my side:\n\n1. How `user-based` and `month-based` rank features are calculated? Simply applying `scipy.stats.rankdata` to raw data per user and per month before making the aggregations?\n\n2. `diff`: Is it `last value` - `nth value` or `nth value` - `nth-1 value`?\n\n3. `LGB series oof`: Let's call the number of customers in train set `N_1` and the number of features `N_2`. The feature matrix you used to calculate `LGB series oof` has size `(N_1, N_2*13)`, correct?\n\n4. What are `GreedyBins`? Could you point some resource to take a look at it?\n\n5. How was your iteration process for trying new features? I mean, how did you decided which features to try/add to your already selected features each time?\n\nAnswers from anybody are welcome :)", "votes": 4, "replies": [ { "id": 1918942, "postDate": "2022-08-30T02:04:55.497Z", "content": "df.groupby('cid')[num_features].rank(pct=True), df.groupby('year-month')[num_features].rank(pct=True)
nth value - nth-1 value
the training set is train_data.merge(train_y,how='left',on='cid')
GreedyBins is a operation in LGB.
\n https://blog.katastros.com/a?ID=01800-4e3a4f7c-6981-40af-b4dd-3224074d705a
when cv and lb boosting, the feature were selected.
Thanks for sharing this amazing way to summarize the solution
", "rawMarkdown": "Thanks for sharing this amazing way to summarize the solution", "votes": 1 }, { "id": 1926570, "postDate": "2022-09-05T00:36:29.467Z", "content": "Congrats on the win! Thank you for sharing！
", "rawMarkdown": "Congrats on the win! Thank you for sharing！", "votes": 1 }, { "id": 1926568, "postDate": "2022-09-05T00:35:26.227Z", "content": "Well done!!
", "rawMarkdown": "Well done!!", "votes": 1 }, { "id": 1921322, "postDate": "2022-08-31T18:01:09.383Z", "content": "Well done!!
", "rawMarkdown": "Well done!!", "votes": 1 }, { "id": 1920529, "postDate": "2022-08-31T07:44:51.527Z", "content": "Congrats!
", "rawMarkdown": "Congrats!\n", "votes": 1 }, { "id": 1920495, "postDate": "2022-08-31T07:28:14.970Z", "content": "Congrats, well done! 💪
", "rawMarkdown": "Congrats, well done! 💪", "votes": 1 }, { "id": 1919865, "postDate": "2022-08-30T17:56:07.570Z", "content": "Congratulations on the win!
", "rawMarkdown": "Congratulations on the win!", "votes": 1 }, { "id": 1919086, "postDate": "2022-08-30T05:59:50.160Z", "content": "Congrats on the solo win🎉
", "rawMarkdown": "Congrats on the solo win🎉", "votes": 1 }, { "id": 1918984, "postDate": "2022-08-30T03:23:43.777Z", "content": "Thanks for sharing amazing way and l look forward to your github code!👍
", "rawMarkdown": "Thanks for sharing amazing way and l look forward to your github code!👍", "votes": 1 }, { "id": 1918876, "postDate": "2022-08-29T23:59:19.337Z", "content": "Thanks for sharing amazing way to summarize solution
", "rawMarkdown": "Thanks for sharing amazing way to summarize solution", "votes": 1 }, { "id": 1918861, "postDate": "2022-08-29T23:17:20.443Z", "content": "Congrats on the win! 🎉Thank U for sharing！
", "rawMarkdown": "Congrats on the win! 🎉Thank U for sharing！", "votes": 1 }, { "id": 1918303, "postDate": "2022-08-29T13:31:20.883Z", "content": "Thanks for sharing amazing way to summarize solution
", "rawMarkdown": "Thanks for sharing amazing way to summarize solution", "votes": 1 }, { "id": 1918293, "postDate": "2022-08-29T13:21:48.477Z", "content": "Really lucky shake )
\nMy congratulation and thanks for sharing!
Thank you for sharing! And congratulations for the win
", "rawMarkdown": "Thank you for sharing! And congratulations for the win", "votes": 1 }, { "id": 1918054, "postDate": "2022-08-29T09:12:44.733Z", "content": "congratulation @daishu and Thank you for sharing your interesting solution.
", "rawMarkdown": "congratulation @daishu and Thank you for sharing your interesting solution.", "votes": 1 }, { "id": 1917641, "postDate": "2022-08-29T00:09:09.233Z", "content": "Congrats on the win! 🎉
", "rawMarkdown": "Congrats on the win! 🎉\n\n", "votes": 1 }, { "id": 1916588, "postDate": "2022-08-28T02:48:23.063Z", "content": "Very interesting @daishu. Enjoy your first solo. Im also beginning in competitions and is very instructive to see your workflow. By the way, feature selection didnt improve my public also. Anyway… a tough competition in my opinion. Congrats!!!
", "rawMarkdown": "Very interesting @daishu. Enjoy your first solo. Im also beginning in competitions and is very instructive to see your workflow. By the way, feature selection didnt improve my public also. Anyway... a tough competition in my opinion. Congrats!!!", "votes": 1 }, { "id": 1916537, "postDate": "2022-08-28T00:59:01.283Z", "content": "Congrats on the win! 🎉
", "rawMarkdown": "Congrats on the win! 🎉", "votes": 1 }, { "id": 1916319, "postDate": "2022-08-27T18:56:16.843Z", "content": "Thanks for sharing!
\nMy first competition, I'm not sure what's normal: are you planning to share any Notebook(s) of the above work?
", "rawMarkdown": "Thanks for sharing!\n\nMy first competition, I'm not sure what's normal: are you planning to share any Notebook(s) of the above work?", "votes": 1, "replies": [ { "id": 1916560, "postDate": "2022-08-28T02:05:14.100Z", "content": "I run all on my local machine. Maybe I will share code at github.
", "rawMarkdown": "I run all on my local machine. Maybe I will share code at github.", "votes": 6, "replies": [ { "id": 1919177, "postDate": "2022-08-30T07:31:23.843Z", "content": "Please, do it
", "rawMarkdown": "Please, do it" } ] }, { "id": 1919411, "postDate": "2022-08-30T11:39:33.493Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 1916019, "postDate": "2022-08-27T15:26:52.357Z", "content": "Congratulations! Interesting solution👍
", "rawMarkdown": "Congratulations! Interesting solution👍", "votes": 1 }, { "id": 1915851, "postDate": "2022-08-27T12:35:39.973Z", "content": "Thanks for sharing your solution @daishu! Very insightful.
", "rawMarkdown": "Thanks for sharing your solution @daishu! Very insightful.", "votes": 1 }, { "id": 1915607, "postDate": "2022-08-27T05:38:53.560Z", "content": "Hearty congratulations for the result!! The achievent is fantastic!! Great approach though!!
", "rawMarkdown": "Hearty congratulations for the result!! The achievent is fantastic!! Great approach though!! ", "votes": 1 }, { "id": 1915463, "postDate": "2022-08-27T02:05:18.147Z", "content": "Congrats! Thank you for sharing. I'm interested in the \"global rank\" feature. How is it different from the raw feature? I thought they are the same for xgb/lgb. Thanks.
", "rawMarkdown": "Congrats! Thank you for sharing. I'm interested in the \"global rank\" feature. How is it different from the raw feature? I thought they are the same for xgb/lgb. Thanks.", "votes": 1, "replies": [ { "id": 1915466, "postDate": "2022-08-27T02:08:58.900Z", "content": "Sorry for my mistake, it's user-based rank.
", "rawMarkdown": "Sorry for my mistake, it's user-based rank.", "votes": 4 } ] }, { "id": 1921489, "postDate": "2022-08-31T21:45:48.733Z", "content": "Congratulations on the win!
", "rawMarkdown": "Congratulations on the win!", "votes": 2 }, { "id": 1919037, "postDate": "2022-08-30T04:37:54.607Z", "content": "Congrats! l look forward to your github code!👍
", "rawMarkdown": "Congrats! l look forward to your github code!👍", "votes": 2 }, { "id": 1918891, "postDate": "2022-08-30T00:43:36.477Z", "content": "Congratulation! That luck met its owner well :) @daishu
", "rawMarkdown": "Congratulation! That luck met its owner well :) @daishu ", "votes": 2 }, { "id": 1915775, "postDate": "2022-08-27T10:49:48.107Z", "content": "Congratulations on the seemingly simple solution. Can you briefly explain why you applied some transformations in the last 3 and 6 rows?
", "rawMarkdown": "Congratulations on the seemingly simple solution. Can you briefly explain why you applied some transformations in the last 3 and 6 rows?", "votes": 2 }, { "id": 2931165, "postDate": "2024-07-21T17:40:04.273Z", "content": "Congrats on the win
", "rawMarkdown": "Congrats on the win " }, { "id": 2669323, "postDate": "2024-02-26T08:01:47.877Z", "content": "why did you do this: df[col] = np.floor(df[col]*100)?? what was the idea behid this??
", "rawMarkdown": "why did you do this: df[col] = np.floor(df[col]*100)?? what was the idea behid this??" }, { "id": 2529211, "postDate": "2023-11-18T03:46:51.800Z", "content": "well this is a late comment but what machine do you have can you please specify?
", "rawMarkdown": "well this is a late comment but what machine do you have can you please specify?" }, { "id": 2204955, "postDate": "2023-04-01T05:18:07.813Z", "content": "great approach and congrats. One question, how did you decide on the weights allotment for different models and what led to you to use these models here? I am learning ensembling and would love some insight.
", "rawMarkdown": "great approach and congrats. One question, how did you decide on the weights allotment for different models and what led to you to use these models here? I am learning ensembling and would love some insight.", "replies": [ { "id": 2207136, "postDate": "2023-04-03T06:20:04.077Z", "content": "I used the public score to determine the best weights.
", "rawMarkdown": "I used the public score to determine the best weights." } ] }, { "id": 2132984, "postDate": "2023-02-07T07:07:16.077Z", "content": "Congratulations folks!
", "rawMarkdown": "Congratulations folks!" }, { "id": 2021309, "postDate": "2022-11-08T06:55:23.853Z", "content": "what's the thinking behind trainning and predicting series datas since Y labels created based on custer_id not on S_2 ,if each customer has uncentain length series datas,for instance,each customer received a lot of calls,would you still do it?
", "rawMarkdown": "what's the thinking behind trainning and predicting series datas since Y labels created based on custer_id not on S_2 ,if each customer has uncentain length series datas,for instance,each customer received a lot of calls,would you still do it?", "replies": [ { "id": 2207134, "postDate": "2023-04-03T06:18:36.200Z", "content": "This is mainly to try to summarize the features using the model, which is different from hand-made.
", "rawMarkdown": "This is mainly to try to summarize the features using the model, which is different from hand-made." } ] }, { "id": 1959880, "postDate": "2022-09-28T11:19:21.157Z", "content": "The best and brief summary of the solution . Congrats on winning this competition
", "rawMarkdown": "The best and brief summary of the solution . Congrats on winning this competition " }, { "id": 1945820, "postDate": "2022-09-19T12:27:20.163Z", "content": "Congrats and thanks for your sharing! If you don't mind, I would like to know the specs of the local machine you used for this competition (I've just run your github codes and I suppose my machine seems to be insufficient…)
", "rawMarkdown": "Congrats and thanks for your sharing! If you don't mind, I would like to know the specs of the local machine you used for this competition (I've just run your github codes and I suppose my machine seems to be insufficient...)", "replies": [ { "id": 1948481, "postDate": "2022-09-21T05:39:53.910Z", "content": "I have a big machine. You can reduce memory by using float16/32 and int8/16.
", "rawMarkdown": "I have a big machine. You can reduce memory by using float16/32 and int8/16.\n", "votes": 2 }, { "id": 1949234, "postDate": "2022-09-21T15:26:14.487Z", "content": "Thanks! I'll try it!
", "rawMarkdown": "Thanks! I'll try it!" } ] }, { "id": 1945670, "postDate": "2022-09-19T09:54:44.157Z", "content": "Congrats for 1st! Thanks for your sharing!
", "rawMarkdown": "Congrats for 1st! Thanks for your sharing!" }, { "id": 1945584, "postDate": "2022-09-19T08:39:29.147Z", "content": "Hi, great solution. I have a question:- Does calculating mean and std. on categorical columns (OneHotcategory) makes sense?
", "rawMarkdown": "Hi, great solution. I have a question:- Does calculating mean and std. on categorical columns (OneHotcategory) makes sense?", "replies": [ { "id": 1948476, "postDate": "2022-09-21T05:35:34.837Z", "content": "Mean stands for probability of one category. Std. maybe needless.
", "rawMarkdown": "Mean stands for probability of one category. Std. maybe needless.", "votes": 1 } ] }, { "id": 1944018, "postDate": "2022-09-18T02:00:20.610Z", "content": "Congratulations!!
", "rawMarkdown": "Congratulations!!" }, { "id": 1942476, "postDate": "2022-09-16T17:05:21.560Z", "content": "Well done!
", "rawMarkdown": "Well done!\n" }, { "id": 1936342, "postDate": "2022-09-12T17:10:35.103Z", "content": "Congratulations on the Solo win.
", "rawMarkdown": "Congratulations on the Solo win." }, { "id": 1936326, "postDate": "2022-09-12T16:58:02.637Z", "content": "Thank you for sharing your solution! They are very helpful! Just wondering if you have done any hyperparam tuning, and if yes what’s your procedure like? Thanks!
", "rawMarkdown": "Thank you for sharing your solution! They are very helpful! Just wondering if you have done any hyperparam tuning, and if yes what’s your procedure like? Thanks!", "replies": [ { "id": 1941370, "postDate": "2022-09-16T02:02:16.060Z", "content": "I have tuned learning_rate, max_depth, num_leaves, bagging_fraction, feature_fraction, min_data_in_leaf, max_bin, min_data_in_bin, lambda_l1, lambda_l2 for LGB and layer_num, hidden_size for NN. Usually tune to boost cv, but also public score in this competition.
", "rawMarkdown": "I have tuned learning_rate, max_depth, num_leaves, bagging_fraction, feature_fraction, min_data_in_leaf, max_bin, min_data_in_bin, lambda_l1, lambda_l2 for LGB and layer_num, hidden_size for NN. Usually tune to boost cv, but also public score in this competition.", "votes": 3 } ] }, { "id": 1930615, "postDate": "2022-09-08T04:40:50.650Z", "content": "Why do you use last 3 rows / last 6 rows for one_hot_category / numerical feature building? Is it because you think more recent data is more important?
", "rawMarkdown": "Why do you use last 3 rows / last 6 rows for one_hot_category / numerical feature building? Is it because you think more recent data is more important?", "replies": [ { "id": 1930862, "postDate": "2022-09-08T09:21:30.487Z", "content": "Yes. My assumption is that the more recent more weight of the data.
", "rawMarkdown": "Yes. My assumption is that the more recent more weight of the data.", "votes": 1 } ] }, { "id": 1930566, "postDate": "2022-09-08T03:46:01.283Z", "content": "Thank you for sharing! Awesome work and congratulations!!!
", "rawMarkdown": "Thank you for sharing! Awesome work and congratulations!!!" }, { "id": 1930298, "postDate": "2022-09-07T17:55:55.887Z", "content": "Well done.. congratulations.. this will help for sure in future.
", "rawMarkdown": "Well done.. congratulations.. this will help for sure in future." }, { "id": 1929444, "postDate": "2022-09-07T04:25:24.440Z", "content": "well done👍
", "rawMarkdown": "well done👍" }, { "id": 1929396, "postDate": "2022-09-07T03:22:44.443Z", "content": "Well done!!
", "rawMarkdown": "Well done!!" }, { "id": 1927677, "postDate": "2022-09-05T21:45:31.113Z", "content": "The diagram explains all, thanks @daishu , beautiful solution.
", "rawMarkdown": "The diagram explains all, thanks @daishu , beautiful solution." }, { "id": 1927441, "postDate": "2022-09-05T16:08:57.047Z", "content": "Congratulations mate, well done !!
", "rawMarkdown": "Congratulations mate, well done !!" }, { "id": 1926915, "postDate": "2022-09-05T08:32:04.873Z", "content": "well down, that's amazing way to express the solution. cograts on the win!
", "rawMarkdown": "well down, that's amazing way to express the solution. cograts on the win!" }, { "id": 1926914, "postDate": "2022-09-05T08:30:13.060Z", "content": "wow, that is an amazing way to summarize the solution. thank you for sharing
", "rawMarkdown": "wow, that is an amazing way to summarize the solution. thank you for sharing" }, { "id": 1926596, "postDate": "2022-09-05T01:35:08.053Z", "content": "Congrats on the win and thanks for sharing the solution!
", "rawMarkdown": "Congrats on the win and thanks for sharing the solution!" }, { "id": 1925572, "postDate": "2022-09-04T05:26:19.480Z", "content": "Congrats and thank you for sharing.
\nI have a question. How was your CV score?
best single model: 0.799
", "rawMarkdown": "best single model: 0.799", "votes": 1 } ] }, { "id": 1925379, "postDate": "2022-09-03T22:57:54.490Z", "content": "Well done!! Congrats
", "rawMarkdown": "Well done!! Congrats" }, { "id": 1925186, "postDate": "2022-09-03T17:39:10.440Z", "content": "Congrats, well done!
", "rawMarkdown": "Congrats, well done!" }, { "id": 1924677, "postDate": "2022-09-03T09:36:10.563Z", "content": "Congratulations!!
", "rawMarkdown": "Congratulations!!" }, { "id": 1924675, "postDate": "2022-09-03T09:34:14.910Z", "content": "Congrats on the solo win!!, looks complicated
", "rawMarkdown": "Congrats on the solo win!!, looks complicated" }, { "id": 1924309, "postDate": "2022-09-02T23:40:51.877Z", "content": "Can someone explain how is he extracting mean, std and sum from the onehot categorical features?
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2171038%2F40623c84861036dc940cbc7c9849ab73%2Fimg.png?generation=1662162008896785&alt=media\")
For each one-hot feature, it's numerical 1s and 0s. So if a single column D_62 or whatever expands to six one-hots, each has a mean, std, and sum (6*3=18 columns). Mean and sum would be redundant if all customers had 13 entries, but are both there as they handle variance in sequence length differently.
", "rawMarkdown": "For each one-hot feature, it's numerical 1s and 0s. So if a single column D_62 or whatever expands to six one-hots, each has a mean, std, and sum (6*3=18 columns). Mean and sum would be redundant if all customers had 13 entries, but are both there as they handle variance in sequence length differently. " } ] }, { "id": 1924139, "postDate": "2022-09-02T18:49:03.523Z", "content": "congratulations
", "rawMarkdown": "congratulations" }, { "id": 1924117, "postDate": "2022-09-02T18:17:39.840Z", "content": "congradulations !
", "rawMarkdown": "congradulations !" }, { "id": 1923282, "postDate": "2022-09-02T05:38:05.783Z", "content": "An interesting solution. Thank you very much!
", "rawMarkdown": "An interesting solution. Thank you very much!" }, { "id": 1923256, "postDate": "2022-09-02T05:04:27.833Z", "content": "Congratulations!! @daishu, very well executed!
", "rawMarkdown": "Congratulations!! @daishu, very well executed!" }, { "id": 1923090, "postDate": "2022-09-02T02:05:45.033Z", "content": "hello, I'm very interested in your best single model. I believe it is an LGB model. I'm wondering if you could share an inference only version of that model as a kaggle notebook. Thank you so much.
", "rawMarkdown": "hello, I'm very interested in your best single model. I believe it is an LGB model. I'm wondering if you could share an inference only version of that model as a kaggle notebook. Thank you so much.", "replies": [ { "id": 1925514, "postDate": "2022-09-04T03:48:10.803Z", "content": "I release a clean code at https://github.com/jxzly/Kaggle-American-Express-Default-Prediction-1st-solution
", "rawMarkdown": "I release a clean code at https://github.com/jxzly/Kaggle-American-Express-Default-Prediction-1st-solution", "votes": 1, "replies": [ { "id": 1926810, "postDate": "2022-09-05T06:35:16.220Z", "content": "Thank you. Nice solution.
", "rawMarkdown": "Thank you. Nice solution." } ] } ] }, { "id": 1922197, "postDate": "2022-09-01T10:36:34.810Z", "content": "well done!!
", "rawMarkdown": "well done!!" }, { "id": 1921908, "postDate": "2022-09-01T06:22:04.180Z", "content": "Congrats, well done
", "rawMarkdown": "Congrats, well done" }, { "id": 1919713, "postDate": "2022-08-30T16:13:10.317Z", "content": "Thank you for sharing your solution @daishu 😊, and congratulations🎉!!!
\nCan you please share with us your code source ?
Maybe, hah
", "rawMarkdown": "Maybe, hah", "votes": 1 } ] }, { "id": 1919659, "postDate": "2022-08-30T15:26:24.560Z", "content": "@daishu congrats on the solo win! May I ask
\n) [EDIT: resolved]
b) do you recall if you have removed any features (and their agg) from adversarial validation, like B_29, R_1 etc ?
\nc) final ensemble is with ranked probs?
", "rawMarkdown": "@daishu congrats on the solo win! May I ask \n~~a) what Greedybins are? is like binned numerical features (?~~) [EDIT: resolved]\n\nb) do you recall if you have removed any features (and their agg) from adversarial validation, like B_29, R_1 etc ? \n\nc) final ensemble is with ranked probs?", "replies": [ { "id": 1920760, "postDate": "2022-08-31T11:43:22.053Z", "content": "ensemble with probs not ranked
", "rawMarkdown": "ensemble with probs not ranked", "votes": 2 } ] }, { "id": 1919188, "postDate": "2022-08-30T07:39:30.247Z", "content": "Congratulation!,Thanks for sharing the idea, but how can i see the figure,in my website, i can only see the blank space , how to see the figure
", "rawMarkdown": "Congratulation!,Thanks for sharing the idea, but how can i see the figure,in my website, i can only see the blank space , how to see the figure", "replies": [ { "id": 1919505, "postDate": "2022-08-30T13:11:31.343Z", "content": "Maybe you need VPN.
", "rawMarkdown": "Maybe you need VPN.", "votes": 1 }, { "id": 1919544, "postDate": "2022-08-30T13:56:16.310Z", "content": "yeah, i could see it , thanks a lot
", "rawMarkdown": "yeah, i could see it , thanks a lot" } ] }, { "id": 1919174, "postDate": "2022-08-30T07:27:18.157Z", "content": "weight sum not equal to 1 is not the matter in this competition, since competition eval on case ranking rather than it true possibility, as ranking will not mess up with the issue
", "rawMarkdown": "weight sum not equal to 1 is not the matter in this competition, since competition eval on case ranking rather than it true possibility, as ranking will not mess up with the issue", "replies": [ { "id": 1919189, "postDate": "2022-08-30T07:39:31.313Z", "content": "But I guess if the author find the total weight is not 1, and if he wants to sum up 1, then he may distribute the extra 0.1 to the four models, which may or may not affect the final scores :)
", "rawMarkdown": "But I guess if the author find the total weight is not 1, and if he wants to sum up 1, then he may distribute the extra 0.1 to the four models, which may or may not affect the final scores :)" }, { "id": 1919197, "postDate": "2022-08-30T07:47:31.993Z", "content": "Therefore keep it in 0.9 is fine for the result.
\nFor 0.1, equal dist to models will get same result.
Thanks for sharing! I have a question, the single GRU model is public 0.790 which is pretty poor comparing with other models. Have you tried discard this model, will that boost the public and private score?
", "rawMarkdown": "Thanks for sharing! I have a question, the single GRU model is public 0.790 which is pretty poor comparing with other models. Have you tried discard this model, will that boost the public and private score?", "replies": [ { "id": 1919509, "postDate": "2022-08-30T13:14:18.703Z", "content": "It's very beneficial for score due to difference.
", "rawMarkdown": "It's very beneficial for score due to difference. ", "votes": 3 } ] }, { "id": 1918777, "postDate": "2022-08-29T20:43:35.380Z", "content": "Congrats @daishu your solution is simple and sweet . In terms of picking the last 3 rows/months for diff/raw and last 6 months for raw aggs what was the reasoning behind it .. for us 1-2 months diff and first last and middle features somehow worked . going 3 months diff also worked as that had the max population excel 1-2 months which will be mostly non default ones I assume .
", "rawMarkdown": "Congrats @daishu your solution is simple and sweet . In terms of picking the last 3 rows/months for diff/raw and last 6 months for raw aggs what was the reasoning behind it .. for us 1-2 months diff and first last and middle features somehow worked . going 3 months diff also worked as that had the max population excel 1-2 months which will be mostly non default ones I assume .", "replies": [ { "id": 1918932, "postDate": "2022-08-30T01:57:55.140Z", "content": "My assumption is that the more recent more weight of the data, so I picked the last 3/6 rows and choose features by cv and lb.
", "rawMarkdown": "My assumption is that the more recent more weight of the data, so I picked the last 3/6 rows and choose features by cv and lb.", "votes": 2 } ] }, { "id": 1918592, "postDate": "2022-08-29T17:28:09.987Z", "content": "I run all on my local machine. Maybe I will share code at github. - Great thanks, but what is your git hub link
\nthere are many this this nick name
\nhttps://github.com/search?l=Python&q=daishu&type=users
@daishu congrats again - huge achievement
", "rawMarkdown": "https://github.com/jxzly\n\n@daishu congrats again - huge achievement ", "votes": 3 } ] }, { "id": 1918464, "postDate": "2022-08-29T15:40:12.390Z", "content": "Congratulations and thank you for sharing your solution!
\nDid the mistake in weight sum was in purpose or divine luck? 😄
", "rawMarkdown": "Congratulations and thank you for sharing your solution! \n\nDid the mistake in weight sum was in purpose or divine luck? 😄", "replies": [ { "id": 1918943, "postDate": "2022-08-30T02:06:47.703Z", "content": "Weights is not very important in private. But it also is divine luck.
", "rawMarkdown": "Weights is not very important in private. But it also is divine luck.", "votes": 1 } ] }, { "id": 1916352, "postDate": "2022-08-27T19:29:31.960Z", "content": "How much RAM and VRAM did your models take? What was your hardware stack?
", "rawMarkdown": "How much RAM and VRAM did your models take? What was your hardware stack?", "replies": [ { "id": 1916559, "postDate": "2022-08-28T02:02:59.567Z", "content": "128g RAM+3090
", "rawMarkdown": "128g RAM+3090", "votes": 2 } ] }, { "id": 1915795, "postDate": "2022-08-27T11:30:49.440Z", "content": "Congratulations!! May i ask how may features you used (the columns of dataframe) ?
", "rawMarkdown": "Congratulations!! May i ask how may features you used (the columns of dataframe) ?", "replies": [ { "id": 1915900, "postDate": "2022-08-27T13:08:17.657Z", "content": "About 6000
", "rawMarkdown": "About 6000", "votes": 5 } ] }, { "id": 1915657, "postDate": "2022-08-27T06:55:41.880Z", "content": "Very interesting, thank you for sharing! Do you try any feature selection methods, I find it didn't work in my case. (I think this method is quiet confusing, it seems to work in some competition, but it doesn't work in most of cases I experienced. Maybe I didn't use it properly )
", "rawMarkdown": "Very interesting, thank you for sharing! Do you try any feature selection methods, I find it didn't work in my case. (I think this method is quiet confusing, it seems to work in some competition, but it doesn't work in most of cases I experienced. Maybe I didn't use it properly )", "replies": [ { "id": 1915670, "postDate": "2022-08-27T07:11:30.660Z", "content": "Yes, I tried feature selection, but it didn't work for public. So I didn't use it for final submission.
", "rawMarkdown": "Yes, I tried feature selection, but it didn't work for public. So I didn't use it for final submission.", "votes": 3 } ] }, { "id": 2137388, "postDate": "2023-02-09T23:59:53.220Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2095443, "postDate": "2023-01-11T11:13:37.950Z", "rawMarkdown": "", "isDeleted": true }, { "id": 1928352, "postDate": "2022-09-06T12:36:50.157Z", "rawMarkdown": "", "isDeleted": true }, { "id": 1924896, "postDate": "2022-09-03T13:58:45.410Z", "rawMarkdown": "", "isDeleted": true }, { "id": 1922401, "postDate": "2022-09-01T13:11:26.563Z", "rawMarkdown": "", "isDeleted": true }, { "id": 1927208, "postDate": "2022-09-05T12:57:29.433Z", "content": "Congrats! Thank you for sharing！
", "rawMarkdown": "Congrats! Thank you for sharing！" } ], "comments": [ { "id": 1916435, "author_name": "Marília Prata", "author_url": "", "post_date": "2022-08-27T21:56:11.107000", "content": "A huge congratulation Daishu!
\n1st time solo winner is an achievement to be remembered. Mostly in this AMEX competition.
\nThough I can't even understand any solutions since I'm a beginner. Thank you for sharing yours.
Thanks for sharing and congratulations for your first solo winner! Some questions from my side:
\nHow user-based and month-based rank features are calculated? Simply applying scipy.stats.rankdata to raw data per user and per month before making the aggregations?
diff: Is it last value - nth value or nth value - nth-1 value?
LGB series oof: Let's call the number of customers in train set N_1 and the number of features N_2. The feature matrix you used to calculate LGB series oof has size (N_1, N_2*13), correct?
What are GreedyBins? Could you point some resource to take a look at it?
How was your iteration process for trying new features? I mean, how did you decided which features to try/add to your already selected features each time?
Answers from anybody are welcome :)
", "votes": 4, "replies": [ { "id": 1918942, "author_name": "Correlation", "author_url": "", "post_date": "2022-08-30T02:04:55.497000", "content": "df.groupby('cid')[num_features].rank(pct=True), df.groupby('year-month')[num_features].rank(pct=True)
nth value - nth-1 value
the training set is train_data.merge(train_y,how='left',on='cid')
GreedyBins is a operation in LGB.
\n https://blog.katastros.com/a?ID=01800-4e3a4f7c-6981-40af-b4dd-3224074d705a
when cv and lb boosting, the feature were selected.
Thanks for sharing this amazing way to summarize the solution
", "votes": 1, "replies": [] }, { "id": 1926570, "author_name": "guansuo", "author_url": "", "post_date": "2022-09-05T00:36:29.467000", "content": "Congrats on the win! Thank you for sharing！
", "votes": 1, "replies": [] }, { "id": 1926568, "author_name": "peterclerk", "author_url": "", "post_date": "2022-09-05T00:35:26.227000", "content": "Well done!!
", "votes": 1, "replies": [] }, { "id": 1921322, "author_name": "Mezue", "author_url": "", "post_date": "2022-08-31T18:01:09.383000", "content": "Well done!!
", "votes": 1, "replies": [] }, { "id": 1920529, "author_name": "Stefan Burger", "author_url": "", "post_date": "2022-08-31T07:44:51.527000", "content": "Congrats!
", "votes": 1, "replies": [] }, { "id": 1920495, "author_name": "Ed Kent-Hazledine", "author_url": "", "post_date": "2022-08-31T07:28:14.970000", "content": "Congrats, well done! 💪
", "votes": 1, "replies": [] }, { "id": 1919865, "author_name": "Gizachew Alemu", "author_url": "", "post_date": "2022-08-30T17:56:07.570000", "content": "Congratulations on the win!
", "votes": 1, "replies": [] }, { "id": 1919086, "author_name": "Ranjeet shrivastav", "author_url": "", "post_date": "2022-08-30T05:59:50.160000", "content": "Congrats on the solo win🎉
", "votes": 1, "replies": [] }, { "id": 1918984, "author_name": "SgangX", "author_url": "", "post_date": "2022-08-30T03:23:43.777000", "content": "Thanks for sharing amazing way and l look forward to your github code!👍
", "votes": 1, "replies": [] }, { "id": 1918876, "author_name": "Lamsaoub", "author_url": "", "post_date": "2022-08-29T23:59:19.337000", "content": "Thanks for sharing amazing way to summarize solution
", "votes": 1, "replies": [] }, { "id": 1918861, "author_name": "", "author_url": "", "post_date": "2022-08-29T23:17:20.443000", "content": "Congrats on the win! 🎉Thank U for sharing！
", "votes": 1, "replies": [] }, { "id": 1918303, "author_name": "Athar Sayed", "author_url": "", "post_date": "2022-08-29T13:31:20.883000", "content": "Thanks for sharing amazing way to summarize solution
", "votes": 1, "replies": [] }, { "id": 1918293, "author_name": "Ivan Kireev", "author_url": "", "post_date": "2022-08-29T13:21:48.477000", "content": "Really lucky shake )
\nMy congratulation and thanks for sharing!
Thank you for sharing! And congratulations for the win
", "votes": 1, "replies": [] }, { "id": 1918054, "author_name": "Matthew Setiadi", "author_url": "", "post_date": "2022-08-29T09:12:44.733000", "content": "congratulation @daishu and Thank you for sharing your interesting solution.
", "votes": 1, "replies": [] }, { "id": 1917641, "author_name": "Ahmed Gomaa", "author_url": "", "post_date": "2022-08-29T00:09:09.233000", "content": "Congrats on the win! 🎉
", "votes": 1, "replies": [] }, { "id": 1916588, "author_name": "Felipe", "author_url": "", "post_date": "2022-08-28T02:48:23.063000", "content": "Very interesting @daishu. Enjoy your first solo. Im also beginning in competitions and is very instructive to see your workflow. By the way, feature selection didnt improve my public also. Anyway… a tough competition in my opinion. Congrats!!!
", "votes": 1, "replies": [] }, { "id": 1916537, "author_name": "shahilpravind", "author_url": "", "post_date": "2022-08-28T00:59:01.283000", "content": "Congrats on the win! 🎉
", "votes": 1, "replies": [] }, { "id": 1916319, "author_name": "Robert Hatch", "author_url": "", "post_date": "2022-08-27T18:56:16.843000", "content": "Thanks for sharing!
\nMy first competition, I'm not sure what's normal: are you planning to share any Notebook(s) of the above work?
", "votes": 1, "replies": [ { "id": 1916560, "author_name": "Correlation", "author_url": "", "post_date": "2022-08-28T02:05:14.100000", "content": "I run all on my local machine. Maybe I will share code at github.
", "votes": 6, "replies": [ { "id": 1919177, "author_name": "Santiago Mota", "author_url": "", "post_date": "2022-08-30T07:31:23.843000", "content": "Please, do it
", "votes": 0, "replies": [] } ] }, { "id": 1919411, "author_name": "", "author_url": "", "post_date": "2022-08-30T11:39:33.493000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 1916019, "author_name": "Mike Mazurov", "author_url": "", "post_date": "2022-08-27T15:26:52.357000", "content": "Congratulations! Interesting solution👍
", "votes": 1, "replies": [] }, { "id": 1915851, "author_name": "Oscar Aguilar", "author_url": "", "post_date": "2022-08-27T12:35:39.973000", "content": "Thanks for sharing your solution @daishu! Very insightful.
", "votes": 1, "replies": [] }, { "id": 1915607, "author_name": "Ravi Ramakrishnan", "author_url": "", "post_date": "2022-08-27T05:38:53.560000", "content": "Hearty congratulations for the result!! The achievent is fantastic!! Great approach though!!
", "votes": 1, "replies": [] }, { "id": 1915463, "author_name": "Jiwei Liu", "author_url": "", "post_date": "2022-08-27T02:05:18.147000", "content": "Congrats! Thank you for sharing. I'm interested in the \"global rank\" feature. How is it different from the raw feature? I thought they are the same for xgb/lgb. Thanks.
", "votes": 1, "replies": [ { "id": 1915466, "author_name": "Correlation", "author_url": "", "post_date": "2022-08-27T02:08:58.900000", "content": "Sorry for my mistake, it's user-based rank.
", "votes": 4, "replies": [] } ] }, { "id": 1921489, "author_name": "Astraz93", "author_url": "", "post_date": "2022-08-31T21:45:48.733000", "content": "Congratulations on the win!
", "votes": 2, "replies": [] }, { "id": 1919037, "author_name": "SgangX", "author_url": "", "post_date": "2022-08-30T04:37:54.607000", "content": "Congrats! l look forward to your github code!👍
", "votes": 2, "replies": [] }, { "id": 1918891, "author_name": "Making TARS", "author_url": "", "post_date": "2022-08-30T00:43:36.477000", "content": "Congratulation! That luck met its owner well :) @daishu
", "votes": 2, "replies": [] }, { "id": 1915775, "author_name": "Raimondo Melis", "author_url": "", "post_date": "2022-08-27T10:49:48.107000", "content": "Congratulations on the seemingly simple solution. Can you briefly explain why you applied some transformations in the last 3 and 6 rows?
", "votes": 2, "replies": [] }, { "id": 2931165, "author_name": "", "author_url": "", "post_date": "2024-07-21T17:40:04.273000", "content": "", "votes": 0, "replies": [] }, { "id": 2669323, "author_name": "", "author_url": "", "post_date": "2024-02-26T08:01:47.877000", "content": "", "votes": 0, "replies": [] }, { "id": 2529211, "author_name": "", "author_url": "", "post_date": "2023-11-18T03:46:51.800000", "content": "", "votes": 0, "replies": [] }, { "id": 2204955, "author_name": "", "author_url": "", "post_date": "2023-04-01T05:18:07.813000", "content": "", "votes": 0, "replies": [ { "id": 2207136, "author_name": "", "author_url": "", "post_date": "2023-04-03T06:20:04.077000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 2132984, "author_name": "", "author_url": "", "post_date": "2023-02-07T07:07:16.077000", "content": "", "votes": 0, "replies": [] }, { "id": 2021309, "author_name": "", "author_url": "", "post_date": "2022-11-08T06:55:23.853000", "content": "", "votes": 0, "replies": [ { "id": 2207134, "author_name": "", "author_url": "", "post_date": "2023-04-03T06:18:36.200000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 1959880, "author_name": "", "author_url": "", "post_date": "2022-09-28T11:19:21.157000", "content": "", "votes": 0, "replies": [] }, { "id": 1945820, "author_name": "", "author_url": "", "post_date": "2022-09-19T12:27:20.163000", "content": "", "votes": 0, "replies": [ { "id": 1948481, "author_name": "", "author_url": "", "post_date": "2022-09-21T05:39:53.910000", "content": "", "votes": 2, "replies": [] }, { "id": 1949234, "author_name": "", "author_url": "", "post_date": "2022-09-21T15:26:14.487000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 1945670, "author_name": "", "author_url": "", "post_date": "2022-09-19T09:54:44.157000", "content": "", "votes": 0, "replies": [] }, { "id": 1945584, "author_name": "", "author_url": "", "post_date": "2022-09-19T08:39:29.147000", "content": "", "votes": 0, "replies": [ { "id": 1948478, "author_name": "", "author_url": "", "post_date": "2022-09-21T05:38:00.077000", "content": "", "votes": 1, "replies": [] } ] }, { "id": 1945094, "author_name": "", "author_url": "", "post_date": "2022-09-18T20:36:02.880000", "content": "", "votes": 0, "replies": [ { "id": 1948476, "author_name": "", "author_url": "", "post_date": "2022-09-21T05:35:34.837000", "content": "", "votes": 1, "replies": [] } ] }, { "id": 1944018, "author_name": "", "author_url": "", "post_date": "2022-09-18T02:00:20.610000", "content": "", "votes": 0, "replies": [] }, { "id": 1942476, "author_name": "", "author_url": "", "post_date": "2022-09-16T17:05:21.560000", "content": "", "votes": 0, "replies": [] }, { "id": 1936342, "author_name": "", "author_url": "", "post_date": "2022-09-12T17:10:35.103000", "content": "", "votes": 0, "replies": [] }, { "id": 1936326, "author_name": "", "author_url": "", "post_date": "2022-09-12T16:58:02.637000", "content": "", "votes": 0, "replies": [ { "id": 1941370, "author_name": "", "author_url": "", "post_date": "2022-09-16T02:02:16.060000", "content": "", "votes": 3, "replies": [] } ] }, { "id": 1930615, "author_name": "", "author_url": "", "post_date": "2022-09-08T04:40:50.650000", "content": "", "votes": 0, "replies": [ { "id": 1930862, "author_name": "", "author_url": "", "post_date": "2022-09-08T09:21:30.487000", "content": "", "votes": 1, "replies": [] } ] }, { "id": 1930566, "author_name": "", "author_url": "", "post_date": "2022-09-08T03:46:01.283000", "content": "", "votes": 0, "replies": [] }, { "id": 1930298, "author_name": "", "author_url": "", "post_date": "2022-09-07T17:55:55.887000", "content": "", "votes": 0, "replies": [] }, { "id": 1929444, "author_name": "", "author_url": "", "post_date": "2022-09-07T04:25:24.440000", "content": "", "votes": 0, "replies": [] }, { "id": 1929396, "author_name": "", "author_url": "", "post_date": "2022-09-07T03:22:44.443000", "content": "", "votes": 0, "replies": [] }, { "id": 1927677, "author_name": "", "author_url": "", "post_date": "2022-09-05T21:45:31.113000", "content": "", "votes": 0, "replies": [] }, { "id": 1927441, "author_name": "", "author_url": "", "post_date": "2022-09-05T16:08:57.047000", "content": "", "votes": 0, "replies": [] }, { "id": 1926915, "author_name": "", "author_url": "", "post_date": "2022-09-05T08:32:04.873000", "content": "", "votes": 0, "replies": [] }, { "id": 1926914, "author_name": "", "author_url": "", "post_date": "2022-09-05T08:30:13.060000", "content": "", "votes": 0, "replies": [] }, { "id": 1926596, "author_name": "", "author_url": "", "post_date": "2022-09-05T01:35:08.053000", "content": "", "votes": 0, "replies": [] }, { "id": 1925572, "author_name": "", "author_url": "", "post_date": "2022-09-04T05:26:19.480000", "content": "", "votes": 0, "replies": [ { "id": 1926567, "author_name": "", "author_url": "", "post_date": "2022-09-05T00:33:03.700000", "content": "", "votes": 1, "replies": [] } ] }, { "id": 1925379, "author_name": "", "author_url": "", "post_date": "2022-09-03T22:57:54.490000", "content": "", "votes": 0, "replies": [] }, { "id": 1925186, "author_name": "", "author_url": "", "post_date": "2022-09-03T17:39:10.440000", "content": "", "votes": 0, "replies": [] }, { "id": 1924677, "author_name": "", "author_url": "", "post_date": "2022-09-03T09:36:10.563000", "content": "", "votes": 0, "replies": [] }, { "id": 1924675, "author_name": "", "author_url": "", "post_date": "2022-09-03T09:34:14.910000", "content": "", "votes": 0, "replies": [] }, { "id": 1924309, "author_name": "", "author_url": "", "post_date": "2022-09-02T23:40:51.877000", "content": "", "votes": 0, "replies": [ { "id": 1924325, "author_name": "", "author_url": "", "post_date": "2022-09-03T00:43:12.253000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 1924139, "author_name": "", "author_url": "", "post_date": "2022-09-02T18:49:03.523000", "content": "", "votes": 0, "replies": [] }, { "id": 1924117, "author_name": "", "author_url": "", "post_date": "2022-09-02T18:17:39.840000", "content": "", "votes": 0, "replies": [] }, { "id": 1923282, "author_name": "", "author_url": "", "post_date": "2022-09-02T05:38:05.783000", "content": "", "votes": 0, "replies": [] }, { "id": 1923256, "author_name": "", "author_url": "", "post_date": "2022-09-02T05:04:27.833000", "content": "", "votes": 0, "replies": [] }, { "id": 1923090, "author_name": "", "author_url": "", "post_date": "2022-09-02T02:05:45.033000", "content": "", "votes": 0, "replies": [ { "id": 1925514, "author_name": "", "author_url": "", "post_date": "2022-09-04T03:48:10.803000", "content": "", "votes": 1, "replies": [ { "id": 1926810, "author_name": "", "author_url": "", "post_date": "2022-09-05T06:35:16.220000", "content": "", "votes": 0, "replies": [] } ] } ] }, { "id": 1922197, "author_name": "", "author_url": "", "post_date": "2022-09-01T10:36:34.810000", "content": "", "votes": 0, "replies": [] }, { "id": 1921908, "author_name": "", "author_url": "", "post_date": "2022-09-01T06:22:04.180000", "content": "", "votes": 0, "replies": [] }, { "id": 1919713, "author_name": "", "author_url": "", "post_date": "2022-08-30T16:13:10.317000", "content": "", "votes": 0, "replies": [ { "id": 1920764, "author_name": "", "author_url": "", "post_date": "2022-08-31T11:44:13.097000", "content": "", "votes": 1, "replies": [] } ] }, { "id": 1919659, "author_name": "", "author_url": "", "post_date": "2022-08-30T15:26:24.560000", "content": "", "votes": 0, "replies": [ { "id": 1920760, "author_name": "", "author_url": "", "post_date": "2022-08-31T11:43:22.053000", "content": "", "votes": 2, "replies": [] } ] }, { "id": 1919188, "author_name": "", "author_url": "", "post_date": "2022-08-30T07:39:30.247000", "content": "", "votes": 0, "replies": [ { "id": 1919505, "author_name": "", "author_url": "", "post_date": "2022-08-30T13:11:31.343000", "content": "", "votes": 1, "replies": [] }, { "id": 1919544, "author_name": "", "author_url": "", "post_date": "2022-08-30T13:56:16.310000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 1919174, "author_name": "", "author_url": "", "post_date": "2022-08-30T07:27:18.157000", "content": "", "votes": 0, "replies": [ { "id": 1919189, "author_name": "", "author_url": "", "post_date": "2022-08-30T07:39:31.313000", "content": "", "votes": 0, "replies": [] }, { "id": 1919197, "author_name": "", "author_url": "", "post_date": "2022-08-30T07:47:31.993000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 1919167, "author_name": "", "author_url": "", "post_date": "2022-08-30T07:25:00.430000", "content": "", "votes": 0, "replies": [ { "id": 1919509, "author_name": "", "author_url": "", "post_date": "2022-08-30T13:14:18.703000", "content": "", "votes": 3, "replies": [] } ] }, { "id": 1918777, "author_name": "", "author_url": "", "post_date": "2022-08-29T20:43:35.380000", "content": "", "votes": 0, "replies": [ { "id": 1918932, "author_name": "", "author_url": "", "post_date": "2022-08-30T01:57:55.140000", "content": "", "votes": 2, "replies": [] } ] }, { "id": 1918592, "author_name": "", "author_url": "", "post_date": "2022-08-29T17:28:09.987000", "content": "", "votes": 0, "replies": [ { "id": 1918597, "author_name": "", "author_url": "", "post_date": "2022-08-29T17:30:00.287000", "content": "", "votes": 3, "replies": [] } ] }, { "id": 1918464, "author_name": "", "author_url": "", "post_date": "2022-08-29T15:40:12.390000", "content": "", "votes": 0, "replies": [ { "id": 1918943, "author_name": "", "author_url": "", "post_date": "2022-08-30T02:06:47.703000", "content": "", "votes": 1, "replies": [] } ] }, { "id": 1916352, "author_name": "", "author_url": "", "post_date": "2022-08-27T19:29:31.960000", "content": "", "votes": 0, "replies": [ { "id": 1916559, "author_name": "", "author_url": "", "post_date": "2022-08-28T02:02:59.567000", "content": "", "votes": 2, "replies": [] } ] }, { "id": 1915795, "author_name": "", "author_url": "", "post_date": "2022-08-27T11:30:49.440000", "content": "", "votes": 0, "replies": [ { "id": 1915900, "author_name": "", "author_url": "", "post_date": "2022-08-27T13:08:17.657000", "content": "", "votes": 5, "replies": [] } ] }, { "id": 1915657, "author_name": "", "author_url": "", "post_date": "2022-08-27T06:55:41.880000", "content": "", "votes": 0, "replies": [ { "id": 1915670, "author_name": "", "author_url": "", "post_date": "2022-08-27T07:11:30.660000", "content": "", "votes": 3, "replies": [] } ] }, { "id": 2137388, "author_name": "", "author_url": "", "post_date": "2023-02-09T23:59:53.220000", "content": "", "votes": 0, "replies": [] }, { "id": 2095443, "author_name": "", "author_url": "", "post_date": "2023-01-11T11:13:37.950000", "content": "", "votes": 0, "replies": [] }, { "id": 1928352, "author_name": "", "author_url": "", "post_date": "2022-09-06T12:36:50.157000", "content": "", "votes": 0, "replies": [] }, { "id": 1924896, "author_name": "", "author_url": "", "post_date": "2022-09-03T13:58:45.410000", "content": "", "votes": 0, "replies": [] }, { "id": 1922401, "author_name": "", "author_url": "", "post_date": "2022-09-01T13:11:26.563000", "content": "", "votes": 0, "replies": [] }, { "id": 1927208, "author_name": "", "author_url": "", "post_date": "2022-09-05T12:57:29.433000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1915428": "First time be a solo winner, I must say there is luck in winning the competition. \n\nMy best result is a heavy ensemble with LGB and NN.\n\n1. Data\n\n\n2. Model\n\n\n\nfor NN model, all data fillna(0) and using nn.utils.rnn.pack_padded_sequence to pad.\n\nI have to look for my early stage model, sorry for small font size in figure.\n\nupdate: I release a clean code at https://github.com/jxzly/Kaggle-American-Express-Default-Prediction-1st-solution.\nnote: You may not be able to reproduce the best result due to random fluctuations.", "1916435": "A huge congratulation Daishu! \n\n1st time solo winner is an achievement to be remembered. Mostly in this AMEX competition.\nThough I can't even understand any solutions since I'm a beginner. Thank you for sharing yours.\n", "1918663": "Thanks for sharing and congratulations for your first solo winner! Some questions from my side:\n\n1. How `user-based` and `month-based` rank features are calculated? Simply applying `scipy.stats.rankdata` to raw data per user and per month before making the aggregations?\n\n2. `diff`: Is it `last value` - `nth value` or `nth value` - `nth-1 value`?\n\n3. `LGB series oof`: Let's call the number of customers in train set `N_1` and the number of features `N_2`. The feature matrix you used to calculate `LGB series oof` has size `(N_1, N_2*13)`, correct?\n\n4. What are `GreedyBins`? Could you point some resource to take a look at it?\n\n5. How was your iteration process for trying new features? I mean, how did you decided which features to try/add to your already selected features each time?\n\nAnswers from anybody are welcome :)", "1927740": "Thanks for sharing this amazing way to summarize the solution", "1926570": "Congrats on the win! Thank you for sharing！", "1926568": "Well done!!", "1921322": "Well done!!", "1920529": "Congrats!\n", "1920495": "Congrats, well done! 💪", "1919865": "Congratulations on the win!", "1919086": "Congrats on the solo win🎉", "1918984": "Thanks for sharing amazing way and l look forward to your github code!👍", "1918876": "Thanks for sharing amazing way to summarize solution", "1918861": "Congrats on the win! 🎉Thank U for sharing！", "1918303": "Thanks for sharing amazing way to summarize solution", "1918293": "Really lucky shake )\nMy congratulation and thanks for sharing!", "1918184": "Thank you for sharing! And congratulations for the win", "1918054": "congratulation @daishu and Thank you for sharing your interesting solution.", "1917641": "Congrats on the win! 🎉\n\n", "1916588": "Very interesting @daishu. Enjoy your first solo. Im also beginning in competitions and is very instructive to see your workflow. By the way, feature selection didnt improve my public also. Anyway... a tough competition in my opinion. Congrats!!!", "1916537": "Congrats on the win! 🎉", "1916319": "Thanks for sharing!\n\nMy first competition, I'm not sure what's normal: are you planning to share any Notebook(s) of the above work?", "1916019": "Congratulations! Interesting solution👍", "1915851": "Thanks for sharing your solution @daishu! Very insightful.", "1915607": "Hearty congratulations for the result!! The achievent is fantastic!! Great approach though!! ", "1915463": "Congrats! Thank you for sharing. I'm interested in the \"global rank\" feature. How is it different from the raw feature? I thought they are the same for xgb/lgb. Thanks.", "1921489": "Congratulations on the win!", "1919037": "Congrats! l look forward to your github code!👍", "1918891": "Congratulation! That luck met its owner well :) @daishu ", "1915775": "Congratulations on the seemingly simple solution. Can you briefly explain why you applied some transformations in the last 3 and 6 rows?", "2931165": "Congrats on the win ", "2669323": "why did you do this: df[col] = np.floor(df[col]*100)?? what was the idea behid this??", "2529211": "well this is a late comment but what machine do you have can you please specify?", "2204955": "great approach and congrats. One question, how did you decide on the weights allotment for different models and what led to you to use these models here? I am learning ensembling and would love some insight.", "2132984": "Congratulations folks!", "2021309": "what's the thinking behind trainning and predicting series datas since Y labels created based on custer_id not on S_2 ,if each customer has uncentain length series datas,for instance,each customer received a lot of calls,would you still do it?", "1959880": "The best and brief summary of the solution . Congrats on winning this competition ", "1945820": "Congrats and thanks for your sharing! If you don't mind, I would like to know the specs of the local machine you used for this competition (I've just run your github codes and I suppose my machine seems to be insufficient...)", "1945670": "Congrats for 1st! Thanks for your sharing!", "1945584": "1. To what degree did the history features improve your score? \n2. Did you use the NNs/GRU with the non time series featurized inputs as well?", "1945094": "Hi, great solution. I have a question:- Does calculating mean and std. on categorical columns (OneHotcategory) makes sense?", "1944018": "Congratulations!!", "1942476": "Well done!\n", "1936342": "Congratulations on the Solo win.", "1936326": "Thank you for sharing your solution! They are very helpful! Just wondering if you have done any hyperparam tuning, and if yes what’s your procedure like? Thanks!", "1930615": "Why do you use last 3 rows / last 6 rows for one_hot_category / numerical feature building? Is it because you think more recent data is more important?", "1930566": "Thank you for sharing! Awesome work and congratulations!!!", "1930298": "Well done.. congratulations.. this will help for sure in future.", "1929444": "well done👍", "1929396": "Well done!!", "1927677": "The diagram explains all, thanks @daishu , beautiful solution.", "1927441": "Congratulations mate, well done !!", "1926915": "well down, that's amazing way to express the solution. cograts on the win!", "1926914": "wow, that is an amazing way to summarize the solution. thank you for sharing", "1926596": "Congrats on the win and thanks for sharing the solution!", "1925572": "Congrats and thank you for sharing. \nI have a question. How was your CV score?", "1925379": "Well done!! Congrats", "1925186": "Congrats, well done!", "1924677": "Congratulations!!", "1924675": "Congrats on the solo win!!, looks complicated", "1924309": "Can someone explain how is he extracting mean, std and sum from the onehot categorical features?\n", "1924139": "congratulations", "1924117": "congradulations !", "1923282": "An interesting solution. Thank you very much!", "1923256": "Congratulations!! @daishu, very well executed!", "1923090": "hello, I'm very interested in your best single model. I believe it is an LGB model. I'm wondering if you could share an inference only version of that model as a kaggle notebook. Thank you so much.", "1922197": "well done!!", "1921908": "Congrats, well done", "1919713": "Thank you for sharing your solution @daishu 😊, and congratulations🎉!!!\nCan you please share with us your code source ? ", "1919659": "@daishu congrats on the solo win! May I ask \n~~a) what Greedybins are? is like binned numerical features (?~~) [EDIT: resolved]\n\nb) do you recall if you have removed any features (and their agg) from adversarial validation, like B_29, R_1 etc ? \n\nc) final ensemble is with ranked probs?", "1919188": "Congratulation!,Thanks for sharing the idea, but how can i see the figure,in my website, i can only see the blank space , how to see the figure", "1919174": "weight sum not equal to 1 is not the matter in this competition, since competition eval on case ranking rather than it true possibility, as ranking will not mess up with the issue", "1919167": "Thanks for sharing! I have a question, the single GRU model is public 0.790 which is pretty poor comparing with other models. Have you tried discard this model, will that boost the public and private score?", "1918777": "Congrats @daishu your solution is simple and sweet . In terms of picking the last 3 rows/months for diff/raw and last 6 months for raw aggs what was the reasoning behind it .. for us 1-2 months diff and first last and middle features somehow worked . going 3 months diff also worked as that had the max population excel 1-2 months which will be mostly non default ones I assume .", "1918592": "I run all on my local machine. Maybe I will share code at github. - Great thanks, but what is your git hub link\nthere are many this this nick name \nhttps://github.com/search?l=Python&q=daishu&type=users\n", "1918464": "Congratulations and thank you for sharing your solution! \n\nDid the mistake in weight sum was in purpose or divine luck? 😄", "1916352": "How much RAM and VRAM did your models take? What was your hardware stack?", "1915795": "Congratulations!! May i ask how may features you used (the columns of dataframe) ?", "1915657": "Very interesting, thank you for sharing! Do you try any feature selection methods, I find it didn't work in my case. (I think this method is quiet confusing, it seems to work in some competition, but it doesn't work in most of cases I experienced. Maybe I didn't use it properly )", "2137388": "", "2095443": "", "1928352": "", "1924896": "", "1922401": "", "1927208": "Congrats! Thank you for sharing！" } }