Hi guys!
\n\nThis is certainly a very interesting competition. Already noticing how some stations seem to do fewer but more complicated work, and others many more repetitive yet simpler tasks based on the features measured for each.
\n\nMy question to the administrators is if there is any graphic or image showing the network among stations (with directions). That could lead to some better insights on designing a solution. Maybe BOSCH could give some details for that? Else of course we could make our guesses on how stations relate to each other. :)
\n\nLow chance of getting one I suppose, but worth asking. :)
", "messages": [ { "id": "131577", "postDate": "08/18/2016 10:13:49", "content": "Hi guys!
\n\nThis is certainly a very interesting competition. Already noticing how some stations seem to do fewer but more complicated work, and others many more repetitive yet simpler tasks based on the features measured for each.
\n\nMy question to the administrators is if there is any graphic or image showing the network among stations (with directions). That could lead to some better insights on designing a solution. Maybe BOSCH could give some details for that? Else of course we could make our guesses on how stations relate to each other. :)
\n\nLow chance of getting one I suppose, but worth asking. :)
", "rawMarkdown": "Hi guys!\r\n\r\nThis is certainly a very interesting competition. Already noticing how some stations seem to do fewer but more complicated work, and others many more repetitive yet simpler tasks based on the features measured for each.\r\n\r\nMy question to the administrators is if there is any graphic or image showing the network among stations (with directions). That could lead to some better insights on designing a solution. Maybe BOSCH could give some details for that? Else of course we could make our guesses on how stations relate to each other. :)\r\n\r\nLow chance of getting one I suppose, but worth asking. :)", "votes": null }, { "id": "131586", "postDate": "08/18/2016 11:39:31", "content": "Has anyone managed to figure out the timestamps yet? They seem to be a set of two numbers, rather than a single fraction. Its probably important to understand exactly what those mean. I doubt a difference of 0.6 and 0.8 is exactly 3 times the difference between 0.6 and 1.2.
", "rawMarkdown": "Has anyone managed to figure out the timestamps yet? They seem to be a set of two numbers, rather than a single fraction. Its probably important to understand exactly what those mean. I doubt a difference of 0.6 and 0.8 is exactly 3 times the difference between 0.6 and 1.2.", "votes": null }, { "id": "131743", "postDate": "08/19/2016 11:14:35", "content": "hmm not there yet sankhamukherjee, but if I get any insights into that I'll post again :)
", "rawMarkdown": "hmm not there yet sankhamukherjee, but if I get any insights into that I'll post again :)", "votes": null }, { "id": "131820", "postDate": "08/19/2016 22:14:52", "content": "The final part of each feature name is a number, ranging from 0 to 4263. This allows to arrange the features in order:
\n\nL0_S0_F0 (numeric)\nL0_S0_D1 (date)\nL0_S0_F2 (numeric)\nL0_S0_D3 (date)\n...\nL3_S51_F4260 (numeric)\nL3_S51_D4261 (date)\nL3_S51_F4262 (numeric)\nL3_S51_D4263 (date)\nWhen the columns are arranged in this order, the numeric date values ("time stamps") seem to be increasing within each row. This makes me think that the steps are probably performed sequentially and that the "network among stations" might just be a straight line.
\n\nPS. I don't understand sankhamukherjee's comment about timestamps being two numbers. To me it seems currently that the column L0_S0_D1, for example, is one number which gives the time (in some unspecified units) where L0_S0_F0 was determined. Similarly, I think L3_S51_D4263 is one number which gives the time for L3_S51_F4262.
", "rawMarkdown": "The final part of each feature name is a number, ranging from 0 to 4263. This allows to arrange the features in order:\r\n\r\n L0_S0_F0 (numeric)\r\n L0_S0_D1 (date)\r\n L0_S0_F2 (numeric)\r\n L0_S0_D3 (date)\r\n ...\r\n L3_S51_F4260 (numeric)\r\n L3_S51_D4261 (date)\r\n L3_S51_F4262 (numeric)\r\n L3_S51_D4263 (date)\r\n\r\nWhen the columns are arranged in this order, the numeric date values (\"time stamps\") seem to be increasing within each row. This makes me think that the steps are probably performed sequentially and that the \"network among stations\" might just be a straight line.\r\n\r\nPS. I don't understand sankhamukherjee's comment about timestamps being two numbers. To me it seems currently that the column L0_S0_D1, for example, is one number which gives the time (in some unspecified units) where L0_S0_F0 was determined. Similarly, I think L3_S51_D4263 is one number which gives the time for L3_S51_F4262.", "votes": null }, { "id": "131835", "postDate": "08/20/2016 02:11:49", "content": "Actually you are right! This makes the problem much simpler ...
", "rawMarkdown": "Actually you are right! This makes the problem much simpler ...", "votes": null }, { "id": "131836", "postDate": "08/20/2016 02:37:37", "content": "Awesome Jochen. That makes me feel better. Stations running in parallel doing the same thing may complicate the analysis somewhat. Oh and btw, as people may have already noticed, S0 through S23 wirh a handful of features seem to be exclusively for L0, which makes me think L0 is where the prep work happens. Two stations S24 and S25 are in L1, with almost a thousand features each. This makes me think L1 is for complicated stuff like folding the batter. You wouldn't buy multiple small folding machines.
\n\nAnd L2 has three stations S26 to S28 each with equal number of features measured (67), which makes me think L2 is where the baking occurs. You wouldn't buy more than three ovens I think.
\n\nFinally, L3 has stations 29 to 51. Based on the number of features measured, I'd imagine S29 and S30 does some sort of cooling and settling work. The downstream stations mostly have to do with packing.
\n\nWell, that is my hunch at least, however it may help contextualize the data. :)
", "rawMarkdown": "Awesome Jochen. That makes me feel better. Stations running in parallel doing the same thing may complicate the analysis somewhat. Oh and btw, as people may have already noticed, S0 through S23 wirh a handful of features seem to be exclusively for L0, which makes me think L0 is where the prep work happens. Two stations S24 and S25 are in L1, with almost a thousand features each. This makes me think L1 is for complicated stuff like folding the batter. You wouldn't buy multiple small folding machines.\r\n\r\nAnd L2 has three stations S26 to S28 each with equal number of features measured (67), which makes me think L2 is where the baking occurs. You wouldn't buy more than three ovens I think.\r\n\r\nFinally, L3 has stations 29 to 51. Based on the number of features measured, I'd imagine S29 and S30 does some sort of cooling and settling work. The downstream stations mostly have to do with packing.\r\n\r\nWell, that is my hunch at least, however it may help contextualize the data. :)", "votes": null }, { "id": "131838", "postDate": "08/20/2016 03:40:32", "content": "Actually that makes a lot of sense. I think this information is going to be super-critical in finding an algorithm for getting the best result. It just turns out that the starting numbers for the date field fluctuates randomly. I am guessing that the organizers just randomized the order of the items in terms of when they were manufactured. It would be interesting to see if all the failed items failed around the same time.
", "rawMarkdown": "Actually that makes a lot of sense. I think this information is going to be super-critical in finding an algorithm for getting the best result. It just turns out that the starting numbers for the date field fluctuates randomly. I am guessing that the organizers just randomized the order of the items in terms of when they were manufactured. It would be interesting to see if all the failed items failed around the same time.", "votes": null }, { "id": "131911", "postDate": "08/20/2016 13:51:30", "content": "Yup well here's a bit of summary information about how many features each station (and thus each Line) has, and how many features for each station have dates. This is based largely on available headers. The actual values in the columns may end up being blank so there may actually be fewer features you can use per station.
\n\nAnyway, this is one of the things I'm using to help keep me sane and keep some grasp of what the raw train data looks like. Uploading it in case it may also help anybody else :)
", "rawMarkdown": "Yup well here's a bit of summary information about how many features each station (and thus each Line) has, and how many features for each station have dates. This is based largely on available headers. The actual values in the columns may end up being blank so there may actually be fewer features you can use per station.\r\n\r\nAnyway, this is one of the things I'm using to help keep me sane and keep some grasp of what the raw train data looks like. Uploading it in case it may also help anybody else :)", "votes": null }, { "id": "132153", "postDate": "08/23/2016 02:24:18", "content": "[quote=Jochen Voss;131820]\nWhen the columns are arranged in this order, the numeric date values ("time stamps") seem to be increasing within each row. \n[/quote]
\n\nHow do you sort the columns by the digits after the F and D?
", "rawMarkdown": "[quote=Jochen Voss;131820]\r\nWhen the columns are arranged in this order, the numeric date values (\"time stamps\") seem to be increasing within each row. \r\n[/quote]\r\n\r\nHow do you sort the columns by the digits after the F and D?", "votes": null }, { "id": "132165", "postDate": "08/23/2016 07:40:59", "content": "[quote=spjca;132153]
\n\nHow do you sort the columns by the digits after the F and D?
\n\n[/quote]
\n\nHi spjca,
\n\nThe numeric value after D (Dn) is matched by a numeric value after F (Fn), where
\n\nDn = Fn + 1 \n(or of course Fn = Dn - 1)
\n\nAs such, L0_S0_F0 is matched by L0_S0_D1.
\n\nPersonally I think this provides a bit of a handicap as the mapping of timestamps to features is less direct, and we'll have to either create an intermediary mapping key between numeric/cateogrical to date and vice versa, OR rename headers completely. (I'd prefer to avoid doing the latter.)
\n\nIn the excel file you will notice that there are a total of 3,108 features (both categorical and numeric) but only 905 timestamps. Not all features have timestamps.
", "rawMarkdown": "[quote=spjca;132153]\r\n\r\nHow do you sort the columns by the digits after the F and D? \r\n\r\n[/quote]\r\n\r\nHi spjca,\r\n\r\nThe numeric value after D (Dn) is matched by a numeric value after F (Fn), where\r\n\r\nDn = Fn + 1 \r\n(or of course Fn = Dn - 1)\r\n\r\nAs such, L0_S0_F0 is matched by L0_S0_D1.\r\n\r\nPersonally I think this provides a bit of a handicap as the mapping of timestamps to features is less direct, and we'll have to either create an intermediary mapping key between numeric/cateogrical to date and vice versa, OR rename headers completely. (I'd prefer to avoid doing the latter.)\r\n\r\nIn the excel file you will notice that there are a total of 3,108 features (both categorical and numeric) but only 905 timestamps. Not all features have timestamps.", "votes": null }, { "id": "132229", "postDate": "08/23/2016 18:45:32", "content": "Great stuff - JSD. With your knowledge of the timestamps, you might have some insight to add to the production map I posted here.
\n\nhttps://www.kaggle.com/c/bosch-production-line-performance/forums/t/23042/production-map
\n\nOne question - do we know this is a bakery or are you using it as example in your post above?
", "rawMarkdown": "Great stuff - JSD. With your knowledge of the timestamps, you might have some insight to add to the production map I posted here.\r\n\r\nhttps://www.kaggle.com/c/bosch-production-line-performance/forums/t/23042/production-map \r\n\r\nOne question - do we know this is a bakery or are you using it as example in your post above?", "votes": null }, { "id": "132271", "postDate": "08/24/2016 07:01:06", "content": "[quote=JohnM;132229]
\n\nOne question - do we know this is a bakery or are you using it as example in your post above?
\n\n[/quote]
\n\nHi John,
\n\nDefinitely not a bakery.
\n\nI'm going to fess up here and admit I misunderstood the problem as predicting whether a zero represents a decadent souffle was produced and a one represents a deflated souffle end product. Hence, my previous statements. (To save some face, please just imagine I was speaking figuratively. :P)
\n\nLooking at the same info though, I'd still reach similar guesses.
\n\nLine Zero deals with prep-work, such as probably sorting parts, and mostly is not time-dependent considering the lack of timestamps for all but the first few stations.
\n\nLine One does something very very very very complicated for everything considering the thousand or so features each of its stations have. Stations 24 and 25 are likely highly specialized and extremely expensive stations.
\n\nLine Two with its three stations may run in parallel and/or do the same relatively simple thing, perhaps even serve as substitutes for each other in case one conks out. The count of features both categorical and numeric are standardized across all, and so are the ratios of what are measured with time. I'd have to see the timestamp distributions across these stations though to be more confident about what I say.
\n\nLine Three's first two stations (29 and 30) do something as critical with its count of features. The rest of the downstream are stations do the finishing touches.
\n\nSaw your image in the other post by the way; nice work :) still trying to understand the concept of product family though, and if I get any insights from the timestamps, I'll be sure to post there in that thread. Cheers.
", "rawMarkdown": "[quote=JohnM;132229]\r\n\r\nOne question - do we know this is a bakery or are you using it as example in your post above?\r\n\r\n[/quote]\r\n\r\nHi John,\r\n\r\nDefinitely not a bakery.\r\n\r\nI'm going to fess up here and admit I misunderstood the problem as predicting whether a zero represents a decadent souffle was produced and a one represents a deflated souffle end product. Hence, my previous statements. (To save some face, please just imagine I was speaking figuratively. :P)\r\n\r\nLooking at the same info though, I'd still reach similar guesses.\r\n\r\nLine Zero deals with prep-work, such as probably sorting parts, and mostly is not time-dependent considering the lack of timestamps for all but the first few stations. \r\n\r\nLine One does something very very very very complicated for everything considering the thousand or so features each of its stations have. Stations 24 and 25 are likely highly specialized and extremely expensive stations.\r\n\r\nLine Two with its three stations *may* run in parallel and/or do the same relatively simple thing, perhaps even serve as substitutes for each other in case one conks out. The count of features both categorical and numeric are standardized across all, and so are the ratios of what are measured with time. I'd have to see the timestamp distributions across these stations though to be more confident about what I say.\r\n\r\nLine Three's first two stations (29 and 30) do something as critical with its count of features. The rest of the downstream are stations do the finishing touches.\r\n\r\nSaw your image in the other post by the way; nice work :) still trying to understand the concept of product family though, and if I get any insights from the timestamps, I'll be sure to post there in that thread. Cheers.", "votes": null }, { "id": "132364", "postDate": "08/24/2016 22:30:00", "content": "Thanks for the insights. Also, I like the idea of the mohawk guy in the picture making a highly engineered souffle!
", "rawMarkdown": "Thanks for the insights. Also, I like the idea of the mohawk guy in the picture making a highly engineered souffle!", "votes": null }, { "id": "132555", "postDate": "08/26/2016 18:22:56", "content": "nice, provides a good insight on this large dataset
", "rawMarkdown": "nice, provides a good insight on this large dataset", "votes": null }, { "id": "136995", "postDate": "09/27/2016 08:28:02", "content": "What's the Id column? is it the part number? If so what does each row mean(part number again?) they don't seem to match.
", "rawMarkdown": "What's the Id column? is it the part number? If so what does each row mean(part number again?) they don't seem to match.", "votes": null } ], "comments": [ { "id": 131586, "author_name": "sankhamukherjee", "author_url": "", "post_date": "08/18/2016 11:39:31", "content": "Has anyone managed to figure out the timestamps yet? They seem to be a set of two numbers, rather than a single fraction. Its probably important to understand exactly what those mean. I doubt a difference of 0.6 and 0.8 is exactly 3 times the difference between 0.6 and 1.2.
", "votes": null, "replies": [] }, { "id": 131743, "author_name": "jedsd3285", "author_url": "", "post_date": "08/19/2016 11:14:35", "content": "hmm not there yet sankhamukherjee, but if I get any insights into that I'll post again :)
", "votes": null, "replies": [] }, { "id": 131820, "author_name": "jochenvoss", "author_url": "", "post_date": "08/19/2016 22:14:52", "content": "The final part of each feature name is a number, ranging from 0 to 4263. This allows to arrange the features in order:
\n\nL0_S0_F0 (numeric)\nL0_S0_D1 (date)\nL0_S0_F2 (numeric)\nL0_S0_D3 (date)\n...\nL3_S51_F4260 (numeric)\nL3_S51_D4261 (date)\nL3_S51_F4262 (numeric)\nL3_S51_D4263 (date)\nWhen the columns are arranged in this order, the numeric date values ("time stamps") seem to be increasing within each row. This makes me think that the steps are probably performed sequentially and that the "network among stations" might just be a straight line.
\n\nPS. I don't understand sankhamukherjee's comment about timestamps being two numbers. To me it seems currently that the column L0_S0_D1, for example, is one number which gives the time (in some unspecified units) where L0_S0_F0 was determined. Similarly, I think L3_S51_D4263 is one number which gives the time for L3_S51_F4262.
", "votes": null, "replies": [] }, { "id": 131835, "author_name": "sankhamukherjee", "author_url": "", "post_date": "08/20/2016 02:11:49", "content": "Actually you are right! This makes the problem much simpler ...
", "votes": null, "replies": [] }, { "id": 131836, "author_name": "jedsd3285", "author_url": "", "post_date": "08/20/2016 02:37:37", "content": "Awesome Jochen. That makes me feel better. Stations running in parallel doing the same thing may complicate the analysis somewhat. Oh and btw, as people may have already noticed, S0 through S23 wirh a handful of features seem to be exclusively for L0, which makes me think L0 is where the prep work happens. Two stations S24 and S25 are in L1, with almost a thousand features each. This makes me think L1 is for complicated stuff like folding the batter. You wouldn't buy multiple small folding machines.
\n\nAnd L2 has three stations S26 to S28 each with equal number of features measured (67), which makes me think L2 is where the baking occurs. You wouldn't buy more than three ovens I think.
\n\nFinally, L3 has stations 29 to 51. Based on the number of features measured, I'd imagine S29 and S30 does some sort of cooling and settling work. The downstream stations mostly have to do with packing.
\n\nWell, that is my hunch at least, however it may help contextualize the data. :)
", "votes": null, "replies": [] }, { "id": 131838, "author_name": "sankhamukherjee", "author_url": "", "post_date": "08/20/2016 03:40:32", "content": "Actually that makes a lot of sense. I think this information is going to be super-critical in finding an algorithm for getting the best result. It just turns out that the starting numbers for the date field fluctuates randomly. I am guessing that the organizers just randomized the order of the items in terms of when they were manufactured. It would be interesting to see if all the failed items failed around the same time.
", "votes": null, "replies": [] }, { "id": 131911, "author_name": "jedsd3285", "author_url": "", "post_date": "08/20/2016 13:51:30", "content": "Yup well here's a bit of summary information about how many features each station (and thus each Line) has, and how many features for each station have dates. This is based largely on available headers. The actual values in the columns may end up being blank so there may actually be fewer features you can use per station.
\n\nAnyway, this is one of the things I'm using to help keep me sane and keep some grasp of what the raw train data looks like. Uploading it in case it may also help anybody else :)
", "votes": null, "replies": [] }, { "id": 132153, "author_name": "seanpj", "author_url": "", "post_date": "08/23/2016 02:24:18", "content": "[quote=Jochen Voss;131820]\nWhen the columns are arranged in this order, the numeric date values ("time stamps") seem to be increasing within each row. \n[/quote]
\n\nHow do you sort the columns by the digits after the F and D?
", "votes": null, "replies": [] }, { "id": 132165, "author_name": "jedsd3285", "author_url": "", "post_date": "08/23/2016 07:40:59", "content": "[quote=spjca;132153]
\n\nHow do you sort the columns by the digits after the F and D?
\n\n[/quote]
\n\nHi spjca,
\n\nThe numeric value after D (Dn) is matched by a numeric value after F (Fn), where
\n\nDn = Fn + 1 \n(or of course Fn = Dn - 1)
\n\nAs such, L0_S0_F0 is matched by L0_S0_D1.
\n\nPersonally I think this provides a bit of a handicap as the mapping of timestamps to features is less direct, and we'll have to either create an intermediary mapping key between numeric/cateogrical to date and vice versa, OR rename headers completely. (I'd prefer to avoid doing the latter.)
\n\nIn the excel file you will notice that there are a total of 3,108 features (both categorical and numeric) but only 905 timestamps. Not all features have timestamps.
", "votes": null, "replies": [] }, { "id": 132229, "author_name": "jpmiller", "author_url": "", "post_date": "08/23/2016 18:45:32", "content": "Great stuff - JSD. With your knowledge of the timestamps, you might have some insight to add to the production map I posted here.
\n\nhttps://www.kaggle.com/c/bosch-production-line-performance/forums/t/23042/production-map
\n\nOne question - do we know this is a bakery or are you using it as example in your post above?
", "votes": null, "replies": [] }, { "id": 132271, "author_name": "jedsd3285", "author_url": "", "post_date": "08/24/2016 07:01:06", "content": "[quote=JohnM;132229]
\n\nOne question - do we know this is a bakery or are you using it as example in your post above?
\n\n[/quote]
\n\nHi John,
\n\nDefinitely not a bakery.
\n\nI'm going to fess up here and admit I misunderstood the problem as predicting whether a zero represents a decadent souffle was produced and a one represents a deflated souffle end product. Hence, my previous statements. (To save some face, please just imagine I was speaking figuratively. :P)
\n\nLooking at the same info though, I'd still reach similar guesses.
\n\nLine Zero deals with prep-work, such as probably sorting parts, and mostly is not time-dependent considering the lack of timestamps for all but the first few stations.
\n\nLine One does something very very very very complicated for everything considering the thousand or so features each of its stations have. Stations 24 and 25 are likely highly specialized and extremely expensive stations.
\n\nLine Two with its three stations may run in parallel and/or do the same relatively simple thing, perhaps even serve as substitutes for each other in case one conks out. The count of features both categorical and numeric are standardized across all, and so are the ratios of what are measured with time. I'd have to see the timestamp distributions across these stations though to be more confident about what I say.
\n\nLine Three's first two stations (29 and 30) do something as critical with its count of features. The rest of the downstream are stations do the finishing touches.
\n\nSaw your image in the other post by the way; nice work :) still trying to understand the concept of product family though, and if I get any insights from the timestamps, I'll be sure to post there in that thread. Cheers.
", "votes": null, "replies": [] }, { "id": 132364, "author_name": "jpmiller", "author_url": "", "post_date": "08/24/2016 22:30:00", "content": "Thanks for the insights. Also, I like the idea of the mohawk guy in the picture making a highly engineered souffle!
", "votes": null, "replies": [] }, { "id": 132555, "author_name": "tarunparmar", "author_url": "", "post_date": "08/26/2016 18:22:56", "content": "nice, provides a good insight on this large dataset
", "votes": null, "replies": [] }, { "id": 136995, "author_name": "anandjeyahar", "author_url": "", "post_date": "09/27/2016 08:28:02", "content": "What's the Id column? is it the part number? If so what does each row mean(part number again?) they don't seem to match.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "131577": "Hi guys!\r\n\r\nThis is certainly a very interesting competition. Already noticing how some stations seem to do fewer but more complicated work, and others many more repetitive yet simpler tasks based on the features measured for each.\r\n\r\nMy question to the administrators is if there is any graphic or image showing the network among stations (with directions). That could lead to some better insights on designing a solution. Maybe BOSCH could give some details for that? Else of course we could make our guesses on how stations relate to each other. :)\r\n\r\nLow chance of getting one I suppose, but worth asking. :)", "131586": "Has anyone managed to figure out the timestamps yet? They seem to be a set of two numbers, rather than a single fraction. Its probably important to understand exactly what those mean. I doubt a difference of 0.6 and 0.8 is exactly 3 times the difference between 0.6 and 1.2.", "131743": "hmm not there yet sankhamukherjee, but if I get any insights into that I'll post again :)", "131820": "The final part of each feature name is a number, ranging from 0 to 4263. This allows to arrange the features in order:\r\n\r\n L0_S0_F0 (numeric)\r\n L0_S0_D1 (date)\r\n L0_S0_F2 (numeric)\r\n L0_S0_D3 (date)\r\n ...\r\n L3_S51_F4260 (numeric)\r\n L3_S51_D4261 (date)\r\n L3_S51_F4262 (numeric)\r\n L3_S51_D4263 (date)\r\n\r\nWhen the columns are arranged in this order, the numeric date values (\"time stamps\") seem to be increasing within each row. This makes me think that the steps are probably performed sequentially and that the \"network among stations\" might just be a straight line.\r\n\r\nPS. I don't understand sankhamukherjee's comment about timestamps being two numbers. To me it seems currently that the column L0_S0_D1, for example, is one number which gives the time (in some unspecified units) where L0_S0_F0 was determined. Similarly, I think L3_S51_D4263 is one number which gives the time for L3_S51_F4262.", "131835": "Actually you are right! This makes the problem much simpler ...", "131836": "Awesome Jochen. That makes me feel better. Stations running in parallel doing the same thing may complicate the analysis somewhat. Oh and btw, as people may have already noticed, S0 through S23 wirh a handful of features seem to be exclusively for L0, which makes me think L0 is where the prep work happens. Two stations S24 and S25 are in L1, with almost a thousand features each. This makes me think L1 is for complicated stuff like folding the batter. You wouldn't buy multiple small folding machines.\r\n\r\nAnd L2 has three stations S26 to S28 each with equal number of features measured (67), which makes me think L2 is where the baking occurs. You wouldn't buy more than three ovens I think.\r\n\r\nFinally, L3 has stations 29 to 51. Based on the number of features measured, I'd imagine S29 and S30 does some sort of cooling and settling work. The downstream stations mostly have to do with packing.\r\n\r\nWell, that is my hunch at least, however it may help contextualize the data. :)", "131838": "Actually that makes a lot of sense. I think this information is going to be super-critical in finding an algorithm for getting the best result. It just turns out that the starting numbers for the date field fluctuates randomly. I am guessing that the organizers just randomized the order of the items in terms of when they were manufactured. It would be interesting to see if all the failed items failed around the same time.", "131911": "Yup well here's a bit of summary information about how many features each station (and thus each Line) has, and how many features for each station have dates. This is based largely on available headers. The actual values in the columns may end up being blank so there may actually be fewer features you can use per station.\r\n\r\nAnyway, this is one of the things I'm using to help keep me sane and keep some grasp of what the raw train data looks like. Uploading it in case it may also help anybody else :)", "132153": "[quote=Jochen Voss;131820]\r\nWhen the columns are arranged in this order, the numeric date values (\"time stamps\") seem to be increasing within each row. \r\n[/quote]\r\n\r\nHow do you sort the columns by the digits after the F and D?", "132165": "[quote=spjca;132153]\r\n\r\nHow do you sort the columns by the digits after the F and D? \r\n\r\n[/quote]\r\n\r\nHi spjca,\r\n\r\nThe numeric value after D (Dn) is matched by a numeric value after F (Fn), where\r\n\r\nDn = Fn + 1 \r\n(or of course Fn = Dn - 1)\r\n\r\nAs such, L0_S0_F0 is matched by L0_S0_D1.\r\n\r\nPersonally I think this provides a bit of a handicap as the mapping of timestamps to features is less direct, and we'll have to either create an intermediary mapping key between numeric/cateogrical to date and vice versa, OR rename headers completely. (I'd prefer to avoid doing the latter.)\r\n\r\nIn the excel file you will notice that there are a total of 3,108 features (both categorical and numeric) but only 905 timestamps. Not all features have timestamps.", "132229": "Great stuff - JSD. With your knowledge of the timestamps, you might have some insight to add to the production map I posted here.\r\n\r\nhttps://www.kaggle.com/c/bosch-production-line-performance/forums/t/23042/production-map \r\n\r\nOne question - do we know this is a bakery or are you using it as example in your post above?", "132271": "[quote=JohnM;132229]\r\n\r\nOne question - do we know this is a bakery or are you using it as example in your post above?\r\n\r\n[/quote]\r\n\r\nHi John,\r\n\r\nDefinitely not a bakery.\r\n\r\nI'm going to fess up here and admit I misunderstood the problem as predicting whether a zero represents a decadent souffle was produced and a one represents a deflated souffle end product. Hence, my previous statements. (To save some face, please just imagine I was speaking figuratively. :P)\r\n\r\nLooking at the same info though, I'd still reach similar guesses.\r\n\r\nLine Zero deals with prep-work, such as probably sorting parts, and mostly is not time-dependent considering the lack of timestamps for all but the first few stations. \r\n\r\nLine One does something very very very very complicated for everything considering the thousand or so features each of its stations have. Stations 24 and 25 are likely highly specialized and extremely expensive stations.\r\n\r\nLine Two with its three stations *may* run in parallel and/or do the same relatively simple thing, perhaps even serve as substitutes for each other in case one conks out. The count of features both categorical and numeric are standardized across all, and so are the ratios of what are measured with time. I'd have to see the timestamp distributions across these stations though to be more confident about what I say.\r\n\r\nLine Three's first two stations (29 and 30) do something as critical with its count of features. The rest of the downstream are stations do the finishing touches.\r\n\r\nSaw your image in the other post by the way; nice work :) still trying to understand the concept of product family though, and if I get any insights from the timestamps, I'll be sure to post there in that thread. Cheers.", "132364": "Thanks for the insights. Also, I like the idea of the mohawk guy in the picture making a highly engineered souffle!", "132555": "nice, provides a good insight on this large dataset", "136995": "What's the Id column? is it the part number? If so what does each row mean(part number again?) they don't seem to match." }, "source": "meta" }