It changes a lot from the current competitions. I hope this one has an interesting data set, it looks very nice due to the topic and the data column descriptions provided.
\n\nSounds like a lot of features to engineer, and that will be probably really great to work in this (unusual?) competition!!
", "messages": [ { "id": 122918, "postDate": "2016-06-08T09:46:27.450Z", "content": "And then Gert will go to a camping site in Belgium and destroy us all!
", "rawMarkdown": "And then Gert will go to a camping site in Belgium and destroy us all!", "votes": 15 }, { "id": 122878, "postDate": "2016-06-08T01:04:36.463Z", "content": "It changes a lot from the current competitions. I hope this one has an interesting data set, it looks very nice due to the topic and the data column descriptions provided.
\n\nSounds like a lot of features to engineer, and that will be probably really great to work in this (unusual?) competition!!
", "rawMarkdown": "It changes a lot from the current competitions. I hope this one has an interesting data set, it looks very nice due to the topic and the data column descriptions provided.\r\n\r\nSounds like a lot of features to engineer, and that will be probably really great to work in this (unusual?) competition!!", "votes": 11 }, { "id": 122916, "postDate": "2016-06-08T09:36:11.017Z", "content": "All of us will also be waiting for you to share great and insightful information about various aspects of the data. You are simply astounding and awesome in your analysis of data.
\n\nRegards\nSayantan
", "rawMarkdown": "All of us will also be waiting for you to share great and insightful information about various aspects of the data. You are simply astounding and awesome in your analysis of data. \r\n\r\nRegards\r\nSayantan", "votes": 4 }, { "id": 124265, "postDate": "2016-06-16T15:50:14.933Z", "content": "It should be working like that - or else we would be using data from the future to predict the past.
\n\nSomething interesting, however, is that we have data of 7 weeks (3,4,5,6,7,8,9) for predicting the two following ones (10 and 11). So we shouldn't be dealing with all the test instances in the same way... ;)
", "rawMarkdown": "It should be working like that - or else we would be using data from the future to predict the past.\r\n\r\nSomething interesting, however, is that we have data of 7 weeks (3,4,5,6,7,8,9) for predicting the two following ones (10 and 11). So we shouldn't be dealing with all the test instances in the same way... ;)", "votes": 1 }, { "id": 123220, "postDate": "2016-06-10T12:28:03.560Z", "content": "[quote=GonzaloAndrésMorenoGómez;122994]
\n\n¿Se necesita saber de series de tiempo para concursar? Es mi primera vez en Kaggle y no lo tengo muy claro \nIs need to know of series of time? This is my first time in Kaggle
\n\n[/quote]
\n\nNo necesariamente, pero necesitaras saber como trabajar con problemas de regresión.
\n\nUna forma sencilla de ver este problema seria: Para cada producto, dadas las unidades vendidas en la semana 1, en la semana 2, ... en la semana N, puedes predecir las ventas en la siguiente semana? (semana N+1). Y así puedes hacer lo mismo con cada producto.
\n\nSuerte!
\n\n=============================================
\n\nNot necessarily, but you would need to know your way with regression problems.
\n\nAn easy way of formulating the problem would be: For each product, given the unit sales in week 1, week 2, ... week N, can you predict the unit sales in the next week? (week N+1). And the you can do the same with each other product.
\n\nGood luck!
", "rawMarkdown": "[quote=GonzaloAndrésMorenoGómez;122994]\r\n\r\n¿Se necesita saber de series de tiempo para concursar? Es mi primera vez en Kaggle y no lo tengo muy claro \r\nIs need to know of series of time? This is my first time in Kaggle\r\n\r\n[/quote]\r\n\r\nNo necesariamente, pero necesitaras saber como trabajar con problemas de regresión.\r\n\r\nUna forma sencilla de ver este problema seria: Para cada producto, dadas las unidades vendidas en la semana 1, en la semana 2, ... en la semana N, puedes predecir las ventas en la siguiente semana? (semana N+1). Y así puedes hacer lo mismo con cada producto.\r\n\r\nSuerte!\r\n\r\n=============================================\r\n\r\nNot necessarily, but you would need to know your way with regression problems.\r\n\r\nAn easy way of formulating the problem would be: For each product, given the unit sales in week 1, week 2, ... week N, can you predict the unit sales in the next week? (week N+1). And the you can do the same with each other product.\r\n\r\nGood luck!\r\n\r\n", "votes": 1 }, { "id": 124886, "postDate": "2016-06-23T04:48:15.517Z", "content": "[quote=Jesse Burström;124826]
\n\nIt worked!!! I got without any hassle 0.49 on LB. Thanks to @Misfyre for the averages script which i used as input to xgboost. I tried to use custom objective function creating gradient and hessian for the competition loss function. It worked not due to the fact it assumes only positive values. But if the evaluation function is custom with only change to alway put predictions to absolute value it worked nicely(objectie standard rmse): R feval function i used:
\n\nevalerror <- function(preds, dtrain) {\n labels <- getinfo(dtrain, "label")\n #preds[preds<0] <- 0\n preds <- abs(preds)\n err <- rmsell(preds, labels)\n return(list(metric = "rmsell", value = err))\n}\n[/quote]
\n\nAre you using using linear regression with xgboost?
", "rawMarkdown": "[quote=Jesse Burström;124826]\r\n\r\nIt worked!!! I got without any hassle 0.49 on LB. Thanks to @Misfyre for the averages script which i used as input to xgboost. I tried to use custom objective function creating gradient and hessian for the competition loss function. It worked not due to the fact it assumes only positive values. But if the evaluation function is custom with only change to alway put predictions to absolute value it worked nicely(objectie standard rmse): R feval function i used:\r\n\r\n evalerror <- function(preds, dtrain) {\r\n labels <- getinfo(dtrain, \"label\")\r\n #preds[preds<0] <- 0\r\n preds <- abs(preds)\r\n err <- rmsell(preds, labels)\r\n return(list(metric = \"rmsell\", value = err))\r\n }\r\n\r\n\r\n\r\n[/quote]\r\n\r\n\r\nAre you using using linear regression with xgboost?" }, { "id": 124848, "postDate": "2016-06-22T18:50:23.887Z", "content": "Jesse; that's great, thanks for sharing! Using the averages/counting/grouping scripts as inputs for ML solutions seem like the way forward!
\n\n/mattias
", "rawMarkdown": "Jesse; that's great, thanks for sharing! Using the averages/counting/grouping scripts as inputs for ML solutions seem like the way forward!\r\n\r\n/mattias" }, { "id": 124328, "postDate": "2016-06-17T07:37:49.840Z", "content": "[quote=Jesse Burström;124306]
\n\nHave idea to use the previous post of combining festures giving 0.52 or so on lb with xgboost. Doing raw xgboost gave me 0.67. Also to get the time series one could put the ladt weeks demand as a variable. Two problems 1 the beginning have to omit 1 week. 2 problem test set have 2 weeks so need to first predict the first week to have variable for next. It seems very possible though.
\n\n[/quote]
\n\nBy raw xgboost do you mean using only past weeks as features? I've been trying to add client and product features but could only manage ~0.62 so far.
", "rawMarkdown": "[quote=Jesse Burström;124306]\r\n\r\nHave idea to use the previous post of combining festures giving 0.52 or so on lb with xgboost. Doing raw xgboost gave me 0.67. Also to get the time series one could put the ladt weeks demand as a variable. Two problems 1 the beginning have to omit 1 week. 2 problem test set have 2 weeks so need to first predict the first week to have variable for next. It seems very possible though.\r\n\r\n[/quote]\r\n\r\nBy raw xgboost do you mean using only past weeks as features? I've been trying to add client and product features but could only manage ~0.62 so far." }, { "id": 124274, "postDate": "2016-06-16T17:36:48.837Z", "content": "Thanks for your response. I'm new to this and got confused by the train vs test formats. Part of this is what you mention - the two weeks in the test file. Cheers.
", "rawMarkdown": "Thanks for your response. I'm new to this and got confused by the train vs test formats. Part of this is what you mention - the two weeks in the test file. Cheers." }, { "id": 124262, "postDate": "2016-06-16T14:56:42.503Z", "content": "This may have been answered somewhere, but do we have confirmation that the weeks are numbered sequentially? That is, week 11 is the most recent week and week 3 happened 8 weeks earlier? Just trying to be sure that it isn't the other way 'round.
", "rawMarkdown": "This may have been answered somewhere, but do we have confirmation that the weeks are numbered sequentially? That is, week 11 is the most recent week and week 3 happened 8 weeks earlier? Just trying to be sure that it isn't the other way 'round." }, { "id": 122994, "postDate": "2016-06-09T03:01:02.050Z", "content": "¿Se necesita saber de series de tiempo para concursar? Es mi primera vez en Kaggle y no lo tengo muy claro \nIs need to know of series of time? This is my first time in Kaggle
", "rawMarkdown": "¿Se necesita saber de series de tiempo para concursar? Es mi primera vez en Kaggle y no lo tengo muy claro \r\nIs need to know of series of time? This is my first time in Kaggle" }, { "id": 122940, "postDate": "2016-06-08T17:24:35.710Z", "content": "Gert is indeed the king of forecasting...
", "rawMarkdown": "Gert is indeed the king of forecasting..." }, { "id": 124891, "postDate": "2016-06-23T06:32:29.400Z", "rawMarkdown": "", "isDeleted": true }, { "id": 124826, "postDate": "2016-06-22T16:11:35.687Z", "rawMarkdown": "", "isDeleted": true }, { "id": 124421, "postDate": "2016-06-18T09:30:57.717Z", "rawMarkdown": "", "isDeleted": true }, { "id": 124420, "postDate": "2016-06-18T09:28:20.373Z", "rawMarkdown": "", "isDeleted": true }, { "id": 124306, "postDate": "2016-06-17T01:24:45.383Z", "rawMarkdown": "", "isDeleted": true }, { "id": 123247, "postDate": "2016-06-10T16:29:32.393Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 122918, "author_name": "Čeduljko", "author_url": "", "post_date": "2016-06-08T09:46:27.450000", "content": "And then Gert will go to a camping site in Belgium and destroy us all!
", "votes": 15, "replies": [] }, { "id": 122916, "author_name": "TheFOX", "author_url": "", "post_date": "2016-06-08T09:36:11.017000", "content": "All of us will also be waiting for you to share great and insightful information about various aspects of the data. You are simply astounding and awesome in your analysis of data.
\n\nRegards\nSayantan
", "votes": 4, "replies": [] }, { "id": 124265, "author_name": "CarrDelling", "author_url": "", "post_date": "2016-06-16T15:50:14.933000", "content": "It should be working like that - or else we would be using data from the future to predict the past.
\n\nSomething interesting, however, is that we have data of 7 weeks (3,4,5,6,7,8,9) for predicting the two following ones (10 and 11). So we shouldn't be dealing with all the test instances in the same way... ;)
", "votes": 1, "replies": [] }, { "id": 123220, "author_name": "CarrDelling", "author_url": "", "post_date": "2016-06-10T12:28:03.560000", "content": "[quote=GonzaloAndrésMorenoGómez;122994]
\n\n¿Se necesita saber de series de tiempo para concursar? Es mi primera vez en Kaggle y no lo tengo muy claro \nIs need to know of series of time? This is my first time in Kaggle
\n\n[/quote]
\n\nNo necesariamente, pero necesitaras saber como trabajar con problemas de regresión.
\n\nUna forma sencilla de ver este problema seria: Para cada producto, dadas las unidades vendidas en la semana 1, en la semana 2, ... en la semana N, puedes predecir las ventas en la siguiente semana? (semana N+1). Y así puedes hacer lo mismo con cada producto.
\n\nSuerte!
\n\n=============================================
\n\nNot necessarily, but you would need to know your way with regression problems.
\n\nAn easy way of formulating the problem would be: For each product, given the unit sales in week 1, week 2, ... week N, can you predict the unit sales in the next week? (week N+1). And the you can do the same with each other product.
\n\nGood luck!
", "votes": 1, "replies": [] }, { "id": 124886, "author_name": "Kevin Kirchhoff", "author_url": "", "post_date": "2016-06-23T04:48:15.517000", "content": "[quote=Jesse Burström;124826]
\n\nIt worked!!! I got without any hassle 0.49 on LB. Thanks to @Misfyre for the averages script which i used as input to xgboost. I tried to use custom objective function creating gradient and hessian for the competition loss function. It worked not due to the fact it assumes only positive values. But if the evaluation function is custom with only change to alway put predictions to absolute value it worked nicely(objectie standard rmse): R feval function i used:
\n\nevalerror <- function(preds, dtrain) {\n labels <- getinfo(dtrain, "label")\n #preds[preds<0] <- 0\n preds <- abs(preds)\n err <- rmsell(preds, labels)\n return(list(metric = "rmsell", value = err))\n}\n[/quote]
\n\nAre you using using linear regression with xgboost?
", "votes": 0, "replies": [] }, { "id": 124848, "author_name": "Mattias Fagerlund", "author_url": "", "post_date": "2016-06-22T18:50:23.887000", "content": "Jesse; that's great, thanks for sharing! Using the averages/counting/grouping scripts as inputs for ML solutions seem like the way forward!
\n\n/mattias
", "votes": 0, "replies": [] }, { "id": 124328, "author_name": "Sangxia", "author_url": "", "post_date": "2016-06-17T07:37:49.840000", "content": "[quote=Jesse Burström;124306]
\n\nHave idea to use the previous post of combining festures giving 0.52 or so on lb with xgboost. Doing raw xgboost gave me 0.67. Also to get the time series one could put the ladt weeks demand as a variable. Two problems 1 the beginning have to omit 1 week. 2 problem test set have 2 weeks so need to first predict the first week to have variable for next. It seems very possible though.
\n\n[/quote]
\n\nBy raw xgboost do you mean using only past weeks as features? I've been trying to add client and product features but could only manage ~0.62 so far.
", "votes": 0, "replies": [] }, { "id": 124274, "author_name": "MarkEvans", "author_url": "", "post_date": "2016-06-16T17:36:48.837000", "content": "Thanks for your response. I'm new to this and got confused by the train vs test formats. Part of this is what you mention - the two weeks in the test file. Cheers.
", "votes": 0, "replies": [] }, { "id": 124262, "author_name": "MarkEvans", "author_url": "", "post_date": "2016-06-16T14:56:42.503000", "content": "This may have been answered somewhere, but do we have confirmation that the weeks are numbered sequentially? That is, week 11 is the most recent week and week 3 happened 8 weeks earlier? Just trying to be sure that it isn't the other way 'round.
", "votes": 0, "replies": [] }, { "id": 122994, "author_name": "GonzaloAndrésMorenoGómez", "author_url": "", "post_date": "2016-06-09T03:01:02.050000", "content": "¿Se necesita saber de series de tiempo para concursar? Es mi primera vez en Kaggle y no lo tengo muy claro \nIs need to know of series of time? This is my first time in Kaggle
", "votes": 0, "replies": [] }, { "id": 122940, "author_name": "TheFOX", "author_url": "", "post_date": "2016-06-08T17:24:35.710000", "content": "Gert is indeed the king of forecasting...
", "votes": 0, "replies": [] }, { "id": 124891, "author_name": "", "author_url": "", "post_date": "2016-06-23T06:32:29.400000", "content": "", "votes": 0, "replies": [] }, { "id": 124826, "author_name": "", "author_url": "", "post_date": "2016-06-22T16:11:35.687000", "content": "", "votes": 0, "replies": [] }, { "id": 124421, "author_name": "", "author_url": "", "post_date": "2016-06-18T09:30:57.717000", "content": "", "votes": 0, "replies": [] }, { "id": 124420, "author_name": "", "author_url": "", "post_date": "2016-06-18T09:28:20.373000", "content": "", "votes": 0, "replies": [] }, { "id": 124306, "author_name": "", "author_url": "", "post_date": "2016-06-17T01:24:45.383000", "content": "", "votes": 0, "replies": [] }, { "id": 123247, "author_name": "", "author_url": "", "post_date": "2016-06-10T16:29:32.393000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "122918": "And then Gert will go to a camping site in Belgium and destroy us all!", "122878": "It changes a lot from the current competitions. I hope this one has an interesting data set, it looks very nice due to the topic and the data column descriptions provided.\r\n\r\nSounds like a lot of features to engineer, and that will be probably really great to work in this (unusual?) competition!!", "122916": "All of us will also be waiting for you to share great and insightful information about various aspects of the data. You are simply astounding and awesome in your analysis of data. \r\n\r\nRegards\r\nSayantan", "124265": "It should be working like that - or else we would be using data from the future to predict the past.\r\n\r\nSomething interesting, however, is that we have data of 7 weeks (3,4,5,6,7,8,9) for predicting the two following ones (10 and 11). So we shouldn't be dealing with all the test instances in the same way... ;)", "123220": "[quote=GonzaloAndrésMorenoGómez;122994]\r\n\r\n¿Se necesita saber de series de tiempo para concursar? Es mi primera vez en Kaggle y no lo tengo muy claro \r\nIs need to know of series of time? This is my first time in Kaggle\r\n\r\n[/quote]\r\n\r\nNo necesariamente, pero necesitaras saber como trabajar con problemas de regresión.\r\n\r\nUna forma sencilla de ver este problema seria: Para cada producto, dadas las unidades vendidas en la semana 1, en la semana 2, ... en la semana N, puedes predecir las ventas en la siguiente semana? (semana N+1). Y así puedes hacer lo mismo con cada producto.\r\n\r\nSuerte!\r\n\r\n=============================================\r\n\r\nNot necessarily, but you would need to know your way with regression problems.\r\n\r\nAn easy way of formulating the problem would be: For each product, given the unit sales in week 1, week 2, ... week N, can you predict the unit sales in the next week? (week N+1). And the you can do the same with each other product.\r\n\r\nGood luck!\r\n\r\n", "124886": "[quote=Jesse Burström;124826]\r\n\r\nIt worked!!! I got without any hassle 0.49 on LB. Thanks to @Misfyre for the averages script which i used as input to xgboost. I tried to use custom objective function creating gradient and hessian for the competition loss function. It worked not due to the fact it assumes only positive values. But if the evaluation function is custom with only change to alway put predictions to absolute value it worked nicely(objectie standard rmse): R feval function i used:\r\n\r\n evalerror <- function(preds, dtrain) {\r\n labels <- getinfo(dtrain, \"label\")\r\n #preds[preds<0] <- 0\r\n preds <- abs(preds)\r\n err <- rmsell(preds, labels)\r\n return(list(metric = \"rmsell\", value = err))\r\n }\r\n\r\n\r\n\r\n[/quote]\r\n\r\n\r\nAre you using using linear regression with xgboost?", "124848": "Jesse; that's great, thanks for sharing! Using the averages/counting/grouping scripts as inputs for ML solutions seem like the way forward!\r\n\r\n/mattias", "124328": "[quote=Jesse Burström;124306]\r\n\r\nHave idea to use the previous post of combining festures giving 0.52 or so on lb with xgboost. Doing raw xgboost gave me 0.67. Also to get the time series one could put the ladt weeks demand as a variable. Two problems 1 the beginning have to omit 1 week. 2 problem test set have 2 weeks so need to first predict the first week to have variable for next. It seems very possible though.\r\n\r\n[/quote]\r\n\r\nBy raw xgboost do you mean using only past weeks as features? I've been trying to add client and product features but could only manage ~0.62 so far.", "124274": "Thanks for your response. I'm new to this and got confused by the train vs test formats. Part of this is what you mention - the two weeks in the test file. Cheers.", "124262": "This may have been answered somewhere, but do we have confirmation that the weeks are numbered sequentially? That is, week 11 is the most recent week and week 3 happened 8 weeks earlier? Just trying to be sure that it isn't the other way 'round.", "122994": "¿Se necesita saber de series de tiempo para concursar? Es mi primera vez en Kaggle y no lo tengo muy claro \r\nIs need to know of series of time? This is my first time in Kaggle", "122940": "Gert is indeed the king of forecasting...", "124891": "", "124826": "", "124421": "", "124420": "", "124306": "", "123247": "" } }