Welcome to the ASHRAE Great Energy Predictor III Competition! I represent the lead member of the ASHRAE Technical Team for this competition.
\n\nThis challenge provides the opportunity to see how buildings actually use energy in the wild and come up with new ways to model their hourly consumption in a scalable way. Please check out this publication for an understanding of why this effort is necessary and some of the basics of energy prediction for buildings!
\n\nPlease feel free to ask your questions in this thread.
\n\nGood luck!
", "messages": [ { "id": 649970, "postDate": "2019-10-16T01:04:51.967Z", "content": "Welcome to the ASHRAE Great Energy Predictor III Competition! I represent the lead member of the ASHRAE Technical Team for this competition.
\n\nThis challenge provides the opportunity to see how buildings actually use energy in the wild and come up with new ways to model their hourly consumption in a scalable way. Please check out this publication for an understanding of why this effort is necessary and some of the basics of energy prediction for buildings!
\n\nPlease feel free to ask your questions in this thread.
\n\nGood luck!
", "rawMarkdown": "Welcome to the ASHRAE Great Energy Predictor III Competition! I represent the lead member of the ASHRAE Technical Team for this competition.\n\nThis challenge provides the opportunity to see how buildings actually use energy in the wild and come up with new ways to model their hourly consumption in a scalable way. Please check out [this publication](https://www.mdpi.com/2504-4990/1/3/56) for an understanding of why this effort is necessary and some of the basics of energy prediction for buildings!\n\nPlease feel free to ask your questions in this thread.\n\nGood luck!", "votes": 35 }, { "id": 656125, "postDate": "2019-10-23T23:53:19.667Z", "content": "Questions concerning weather data and timestamp.\nCan you validate the weather data for all sites? (site_ids 0 through 15)
\n\nThe weather data appears inconsistent with the timestamp. For example, at site 4, the peak daily temperature occurs between 9PM and midnight, while the daily low temperatures occur between 1PM and 3 PM. This is not an occasional occurrence, but a pattern that is evident throughout the year. Actual daily lows occur around the hour before sunrise, not 2PM. Actual daily highs occur in the mid-afternoon, not at night time. As the data is at odds with data from the national weather service (I can't find any data on the NWS site that supports temperature profiles like the ones given), it appears to be in error.
\n\nThis calls into question weather at other sites. A glance reveals some patterns that are once again at odds with actual weather patterns. For example, at site 0 the temperature decreases well past sunrise, while at site 1, the temperature begins to increase well before sunrise.
\n\nThe meter reading profiles appear consistent with consumption patterns from other data sources. Perhaps the times for the weather data are not local times, but then the weather profile and the meter reading profile are not in synch. Sometimes weather is provided not in local time but by GMT (London time). If this is the case, then the sites are not all in the USA, but lie in other countries. (That would create another problem because we would not know which holiday schedule to apply.)
\n\nOne minor point, hourly meter readings normally indicate consumption at the end of the hour. I.e. a reading for 0200 indicates the consumption from 0100 to 0200. In this case, the annual data should begin with timestamp 2016-01-01 0100 and end with timestamp 2017-0101 0000. But your training sets all end at 2016-12-31 2300 which does not account for the final hour of the year.
\n\nPlease give some insight on the weather and timestamps so that we can synchronize the meter reading and weather profiles.
\n\nThank you
", "rawMarkdown": "Questions concerning weather data and timestamp.\nCan you validate the weather data for all sites? (site_ids 0 through 15)\n\nThe weather data appears inconsistent with the timestamp. For example, at site 4, the peak daily temperature occurs between 9PM and midnight, while the daily low temperatures occur between 1PM and 3 PM. This is not an occasional occurrence, but a pattern that is evident throughout the year. Actual daily lows occur around the hour before sunrise, not 2PM. Actual daily highs occur in the mid-afternoon, not at night time. As the data is at odds with data from the national weather service (I can't find any data on the NWS site that supports temperature profiles like the ones given), it appears to be in error.\n\nThis calls into question weather at other sites. A glance reveals some patterns that are once again at odds with actual weather patterns. For example, at site 0 the temperature decreases well past sunrise, while at site 1, the temperature begins to increase well before sunrise.\n\nThe meter reading profiles appear consistent with consumption patterns from other data sources. Perhaps the times for the weather data are not local times, but then the weather profile and the meter reading profile are not in synch. Sometimes weather is provided not in local time but by GMT (London time). If this is the case, then the sites are not all in the USA, but lie in other countries. (That would create another problem because we would not know which holiday schedule to apply.)\n\nOne minor point, hourly meter readings normally indicate consumption at the end of the hour. I.e. a reading for 0200 indicates the consumption from 0100 to 0200. In this case, the annual data should begin with timestamp 2016-01-01 0100 and end with timestamp 2017-0101 0000. But your training sets all end at 2016-12-31 2300 which does not account for the final hour of the year.\n\nPlease give some insight on the weather and timestamps so that we can synchronize the meter reading and weather profiles.\n\nThank you", "votes": 19, "replies": [ { "id": 656187, "postDate": "2019-10-24T01:48:38.463Z", "content": "Interesting findings!
", "rawMarkdown": "Interesting findings!\n", "votes": 4 }, { "id": 658651, "postDate": "2019-10-26T10:05:24.007Z", "content": "@amazer @claytonmiller yes I found similar issues in this Notebook and opened a discussion topic for that. Maybe we can hear from you/the organizing team, the best course of action?
", "rawMarkdown": "@amazer @claytonmiller yes I found similar issues in this [Notebook](https://www.kaggle.com/jimeno/ashrae-weather-features-are-not-in-local-time) and opened a [discussion topic ](https://www.kaggle.com/c/ashrae-energy-prediction/discussion/114447#latest-658637)for that. Maybe we can hear from you/the organizing team, the best course of action?", "votes": 3 }, { "id": 717281, "postDate": "2020-01-13T01:05:31.730Z", "content": "Hi @amazer -- great investigative work! The community was able to fully identify this issue and I didn't want to throw a wrench in the mix. Thanks for your patience.
", "rawMarkdown": "Hi @amazer -- great investigative work! The community was able to fully identify this issue and I didn't want to throw a wrench in the mix. Thanks for your patience." } ] }, { "id": 651418, "postDate": "2019-10-17T13:25:44.963Z", "content": "\n\nas you can see at my kernel Starter ⚡ Great Energy Predictor \nI explain this.
\n\nHow do you interpret meter_reading=0? no consume?
\n\nIf you check the builing 0 (building_id = 0) you will see:\n- there is no information between January - May 2016 (months 1 to 5). meter_reading==0.\n- From June 2016 (month=6) we have information!
In my opinion you started to measure the consume of that specific building in June 2016. \nIn order to complete the database from 2016, you filled the other months with 0.
\n\nas @dmitriyguller said:
\n\n\n\n\nIgnoring 0-values won't take care of the problem, it will make the problem different. Instead of under-estimating the energy use for a building with a stuttering meter, it will over-estimate it. Another problem is that we don't actually know whether zeroes are ignored or not, so how are we supposed to build the model intelligently?\n If I were to guess, I would say that the big CV-LB differences are due to either different behavior of zeroes in the test set, or different treatment of them. This is why I'm staying out of this competition: the bad choice of metric is going to turn it into nonsense. You're going to get a much bigger bang for the buck figuring out how to deal with zeroes than you will figuring out how to build an intelligent model.
\n
probably you can't answer, but I had to try :)
", "rawMarkdown": "Hi @claytonmiller \n\nas you can see at my kernel [Starter ⚡ Great Energy Predictor ](https://www.kaggle.com/jesucristo/starter-great-energy-predictor)\nI explain this.\n\n**How do you interpret meter_reading=0?** no consume?\n\nIf you check the builing ```0``` (building_id = 0) you will see:\n- there is no information between January - May 2016 (months 1 to 5). meter_reading==0.\n- From June 2016 (month=6) we have information! \n\nIn my opinion you started to measure the consume of that specific building in June 2016. \nIn order to complete the database from 2016, you filled the other months with 0.\n\nas @dmitriyguller said:\n> Ignoring 0-values won't take care of the problem, it will make the problem different. Instead of under-estimating the energy use for a building with a stuttering meter, it will over-estimate it. Another problem is that we don't actually know whether zeroes are ignored or not, so how are we supposed to build the model intelligently?\n>If I were to guess, I would say that the big CV-LB differences are due to either different behavior of zeroes in the test set, or different treatment of them. This is why I'm staying out of this competition: the bad choice of metric is going to turn it into nonsense. **You're going to get a much bigger bang for the buck figuring out how to deal with zeroes than you will figuring out how to build an intelligent model.**\n\nprobably you can't answer, but I had to try :)", "votes": 21, "replies": [ { "id": 717285, "postDate": "2020-01-13T01:08:03.400Z", "content": "@jesucristo -- great question! In the field, a 0 reading could mean a lot of things. Often it simply means no consumption, but also common is that the meter has flat-lined and it is an error.
@claytonmiller Could you double check that the timestamps for weather data and meter reading are using the same timezone? My impression is that meter readings are using local time while weather data is using some other time zone.
\n\nSee the thread here: https://www.kaggle.com/c/ashrae-energy-prediction/discussion/114447#latest-658783
", "rawMarkdown": "@claytonmiller Could you double check that the timestamps for weather data and meter reading are using the same timezone? My impression is that meter readings are using local time while weather data is using some other time zone.\n\nSee the thread here: https://www.kaggle.com/c/ashrae-energy-prediction/discussion/114447#latest-658783", "votes": 12, "replies": [ { "id": 662419, "postDate": "2019-10-31T14:31:28.457Z", "content": "@sohier Maybe you could look into this?
", "rawMarkdown": "@sohier Maybe you could look into this?" }, { "id": 664438, "postDate": "2019-11-03T17:04:06.353Z", "content": "I am assuming that this additional complexity has been introduced to make the challenge interesting.
", "rawMarkdown": "I am assuming that this additional complexity has been introduced to make the challenge interesting." }, { "id": 717284, "postDate": "2020-01-13T01:06:33.177Z", "content": "@sdoria -- See previous posting and my reply
", "rawMarkdown": "@sdoria -- See previous posting and my reply" } ] }, { "id": 652783, "postDate": "2019-10-19T12:06:45.447Z", "content": "The goal of this competition (i.e. building a counterfactual model) is very reminiscent of Google's CausalImpact.
", "rawMarkdown": "The goal of this competition (i.e. building a counterfactual model) is very reminiscent of Google's [CausalImpact](https://google.github.io/CausalImpact/).", "votes": 4 }, { "id": 650791, "postDate": "2019-10-16T17:22:25.777Z", "content": "Just out of curiosity , if it is ASHRAE - Great Energy Predictor III , where are Predictor I and Predictor II ?? 🙄
", "rawMarkdown": "Just out of curiosity , if it is ASHRAE - Great Energy Predictor III , where are Predictor I and Predictor II ?? 🙄 ", "votes": 4, "replies": [ { "id": 651025, "postDate": "2019-10-17T01:12:10.647Z", "content": "Check out this overview of the previous competitions: https://www.kaggle.com/c/ashrae-energy-prediction/overview/prior-competitions
\n\nThey were held in the 90's and the first one even had the contestants receiving floppy disks in the mail to compete :)
", "rawMarkdown": "Check out this overview of the previous competitions: https://www.kaggle.com/c/ashrae-energy-prediction/overview/prior-competitions\n\nThey were held in the 90's and the first one even had the contestants receiving floppy disks in the mail to compete :)", "votes": 6 }, { "id": 653603, "postDate": "2019-10-20T17:48:35.023Z", "content": "What Clayton has mentioned elsewhere - Kaggle has the floppy disk data available for you to see if todays methods can beat the 90's.
", "rawMarkdown": "What Clayton has mentioned elsewhere - Kaggle has the floppy disk data available for you to see if todays methods can beat the 90's. ", "votes": 2 }, { "id": 654203, "postDate": "2019-10-21T15:33:50.277Z", "content": "", "rawMarkdown": "[Right here](https://www.kaggle.com/c/great-energy-predictor-shootout-i/data)!", "votes": 1 } ] }, { "id": 650131, "postDate": "2019-10-16T06:17:19.153Z", "content": "Looking forward to an interesting competition, @claytonmiller!
\n\nSome kagglers in another thread expressed concern regards to how 'production usable' models out of this competition would be if they incorporated weather information—in reality, only forecasts would be known at inference, yet we're seemingly given actual accurate weather information for up to two years into the future.
\n\nHow do we reconcile this? If weather truly impacts energy usage, then model output should correlate to weather and without an accurate future two-year forecast, ASHRAE won't be able to do much with the results.
", "rawMarkdown": "Looking forward to an interesting competition, @claytonmiller!\n\nSome kagglers in another thread expressed concern regards to how 'production usable' models out of this competition would be if they incorporated weather information—in reality, only forecasts would be known at inference, yet we're seemingly given actual accurate weather information for up to two years into the future.\n\nHow do we reconcile this? If weather truly impacts energy usage, then model output should correlate to weather and without an accurate future two-year forecast, ASHRAE won't be able to do much with the results.", "votes": 4, "replies": [ { "id": 650256, "postDate": "2019-10-16T08:23:43.817Z", "content": "Ideally, the goal of the competition should have been to predict the energy usage 'X' days or 'W' weeks in advance, i.e. standing today, my model should be able to predict the energy consumption of the Xth day in future. That way, I don't the need the weather on the Xth day.
\n\nWill be willing to hear your thoughts @claytonmiller
", "rawMarkdown": "Ideally, the goal of the competition should have been to predict the energy usage 'X' days or 'W' weeks in advance, i.e. standing today, my model should be able to predict the energy consumption of the Xth day in future. That way, I don't the need the weather on the Xth day. \n\nWill be willing to hear your thoughts @claytonmiller ", "votes": -1 }, { "id": 650372, "postDate": "2019-10-16T09:58:46.047Z", "content": "I think I kind of follow the purpose:
\n\nDeveloping energy savings has two key elements: Forecasting future energy usage without improvements, and forecasting energy use after a specific set of improvements have been implemented, like the installation and purchase of investment-grade meters, whose prices continue to fall.
So, here we are trying to \"forecast energy use after a specific set of improvements have been implemented\" and that's why the weather data is available.
", "rawMarkdown": "I think I kind of follow the purpose:\n\n`Developing energy savings has two key elements: Forecasting future energy usage without improvements, and forecasting energy use after a specific set of improvements have been implemented, like the installation and purchase of investment-grade meters, whose prices continue to fall.`\n\nSo, here we are trying to \"forecast energy use after a specific set of improvements have been implemented\" and that's why the weather data is available. ", "votes": 2 }, { "id": 650949, "postDate": "2019-10-16T21:43:36.933Z", "content": "As per publication provided by @claytonmiller
\n\n\"For example, long-term forecasting is usually retroactive as the energy savings prediction is focused on what energy the building would have consumed if the intervention had not taken place. This situation means that input variables such as weather are known and not forecasts themselves. \"
\n\nwhich I think answers your question?\nBr,\nMartin
", "rawMarkdown": "As per publication provided by @claytonmiller \n\n\"For example, long-term forecasting is usually retroactive as the energy savings prediction is focused on what energy the building would have consumed if the intervention had not taken place. This situation means that input variables such as weather are known and not forecasts themselves. \"\n\nwhich I think answers your question?\nBr,\nMartin", "votes": 4 }, { "id": 651024, "postDate": "2019-10-17T01:10:31.890Z", "content": "Hey @authman and @arnabbiswas1 -- @diginfo hit the nail on the head with his explanation of the goal of long term prediction in this context -- you can find more information from this publication: https://www.mdpi.com/2504-4990/1/3/56
\n\nThere is a whole domain of performance contracting which uses retroactive long term prediction to quantify what energy consumption would have potentially occurred without an energy savings intervention
", "rawMarkdown": "Hey @authman and @arnabbiswas1 -- @diginfo hit the nail on the head with his explanation of the goal of long term prediction in this context -- you can find more information from this publication: https://www.mdpi.com/2504-4990/1/3/56\n\nThere is a whole domain of performance contracting which uses retroactive long term prediction to quantify what energy consumption would have potentially occurred without an energy savings intervention", "votes": 3 }, { "id": 653607, "postDate": "2019-10-20T17:54:10.163Z", "content": "In 2009 I retired and was thinking about a new city/state to live in. Being a data driver xx?% kind of guy I downloaded all the weather data for USA for a decade. Than looked for city where it never got too hot or too cold.
\n\nWhen you plot a decade worth of temperatures for a city, the curve is very repeating. So making a model of the next decade of weather for a city is pretty much only an issue of getting the data set. Can't of course model climate change, but otherwise not that big of a task for design. I would assume that such a data set can in fact be purchased so folks designing buildings know what cooling and heating load will occur.
", "rawMarkdown": "In 2009 I retired and was thinking about a new city/state to live in. Being a data driver xx?% kind of guy I downloaded all the weather data for USA for a decade. Than looked for city where it never got too hot or too cold.\n\nWhen you plot a decade worth of temperatures for a city, the curve is very repeating. So making a model of the next decade of weather for a city is pretty much only an issue of getting the data set. Can't of course model climate change, but otherwise not that big of a task for design. I would assume that such a data set can in fact be purchased so folks designing buildings know what cooling and heating load will occur.", "votes": 1 } ] }, { "id": 692066, "postDate": "2019-12-10T20:28:22.147Z", "content": "Hi! I ve a question about metering. Does every building include every meter that it orginaly have? Can we assume that if building has only electricity meter in the competition heating and cooling services are also provided by electricity?\nWith Regards
", "rawMarkdown": "Hi! I ve a question about metering. Does every building include every meter that it orginaly have? Can we assume that if building has only electricity meter in the competition heating and cooling services are also provided by electricity?\nWith Regards", "votes": 1 }, { "id": 651115, "postDate": "2019-10-17T04:04:21.130Z", "content": "Predictor III in the wild. Its called calculus, I have worked in construction for years, The Idea that its colder at night then in the day seems to blow these peoples minds away.
", "rawMarkdown": "Predictor III in the wild. Its called calculus, I have worked in construction for years, The Idea that its colder at night then in the day seems to blow these peoples minds away.", "votes": 1 }, { "id": 653529, "postDate": "2019-10-20T15:36:42.363Z", "content": "Do some of the buildings use a source of heat that is not accounted for in the meters?
", "rawMarkdown": "Do some of the buildings use a source of heat that is not accounted for in the meters?", "votes": 2, "replies": [ { "id": 717291, "postDate": "2020-01-13T01:10:55.890Z", "content": "yes, that is possible
", "rawMarkdown": "yes, that is possible" } ] }, { "id": 656293, "postDate": "2019-10-24T05:55:28.383Z", "content": "To those considering the best algorithm to start with, consider starting with LightGBM.
", "rawMarkdown": "To those considering the best algorithm to start with, consider starting with LightGBM.", "votes": -1 }, { "id": 664792, "postDate": "2019-11-04T08:25:30.747Z", "rawMarkdown": "", "isDeleted": true }, { "id": 658445, "postDate": "2019-10-26T02:32:59.570Z", "content": "Hi @claytonmiller, could you provide some details around the meter time measurements and how they correspond to the weather measurements on a site-by-site basis? There could be a consistent measurement methodology that you folks have employed (and are assuming), but in the wild, all that stuff depends on the utility and can vary greatly. Similarly, I'm assuming the readings are in kWh and temperature appears to be in celsius, but I don't really know. What I'm afraid of is that there's an industrial site or big data center mixed in this dataset that has a substation level MWh meter. This would throw off the models if one is trying to view information in aggregate.
\n\nThanks!
", "rawMarkdown": "Hi @claytonmiller, could you provide some details around the meter time measurements and how they correspond to the weather measurements on a site-by-site basis? There could be a consistent measurement methodology that you folks have employed (and are assuming), but in the wild, all that stuff depends on the utility and can vary greatly. Similarly, I'm assuming the readings are in kWh and temperature appears to be in celsius, but I don't really know. What I'm afraid of is that there's an industrial site or big data center mixed in this dataset that has a substation level MWh meter. This would throw off the models if one is trying to view information in aggregate.\n\nThanks!\n\n", "votes": 5, "isDeleted": true, "replies": [ { "id": 717290, "postDate": "2020-01-13T01:10:30.637Z", "content": "We tried to make all of the reading the same unit (kWh) to prevent mix-ups as you mention and the weather units are posted. We made a mistake in which one of the site's wasn't converted correctly though -- it was in the updated documentation.
", "rawMarkdown": "We tried to make all of the reading the same unit (kWh) to prevent mix-ups as you mention and the weather units are posted. We made a mistake in which one of the site's wasn't converted correctly though -- it was in the updated documentation." } ] }, { "id": 653889, "postDate": "2019-10-21T05:52:48.413Z", "rawMarkdown": "", "isDeleted": true }, { "id": 653306, "postDate": "2019-10-20T08:07:10.773Z", "content": "Thanks
", "rawMarkdown": "Thanks", "votes": 1 }, { "id": 650660, "postDate": "2019-10-16T15:22:10.777Z", "content": "Thanks for the hosting.
", "rawMarkdown": "Thanks for the hosting.", "votes": -1 } ], "comments": [ { "id": 656125, "author_name": "Arthur Mazer", "author_url": "", "post_date": "2019-10-23T23:53:19.667000", "content": "Questions concerning weather data and timestamp.\nCan you validate the weather data for all sites? (site_ids 0 through 15)
\n\nThe weather data appears inconsistent with the timestamp. For example, at site 4, the peak daily temperature occurs between 9PM and midnight, while the daily low temperatures occur between 1PM and 3 PM. This is not an occasional occurrence, but a pattern that is evident throughout the year. Actual daily lows occur around the hour before sunrise, not 2PM. Actual daily highs occur in the mid-afternoon, not at night time. As the data is at odds with data from the national weather service (I can't find any data on the NWS site that supports temperature profiles like the ones given), it appears to be in error.
\n\nThis calls into question weather at other sites. A glance reveals some patterns that are once again at odds with actual weather patterns. For example, at site 0 the temperature decreases well past sunrise, while at site 1, the temperature begins to increase well before sunrise.
\n\nThe meter reading profiles appear consistent with consumption patterns from other data sources. Perhaps the times for the weather data are not local times, but then the weather profile and the meter reading profile are not in synch. Sometimes weather is provided not in local time but by GMT (London time). If this is the case, then the sites are not all in the USA, but lie in other countries. (That would create another problem because we would not know which holiday schedule to apply.)
\n\nOne minor point, hourly meter readings normally indicate consumption at the end of the hour. I.e. a reading for 0200 indicates the consumption from 0100 to 0200. In this case, the annual data should begin with timestamp 2016-01-01 0100 and end with timestamp 2017-0101 0000. But your training sets all end at 2016-12-31 2300 which does not account for the final hour of the year.
\n\nPlease give some insight on the weather and timestamps so that we can synchronize the meter reading and weather profiles.
\n\nThank you
", "votes": 19, "replies": [ { "id": 656187, "author_name": "S D", "author_url": "", "post_date": "2019-10-24T01:48:38.463000", "content": "Interesting findings!
", "votes": 4, "replies": [] }, { "id": 658651, "author_name": "JimFonseca", "author_url": "", "post_date": "2019-10-26T10:05:24.007000", "content": "@amazer @claytonmiller yes I found similar issues in this Notebook and opened a discussion topic for that. Maybe we can hear from you/the organizing team, the best course of action?
", "votes": 3, "replies": [] }, { "id": 717281, "author_name": "Clayton Miller", "author_url": "", "post_date": "2020-01-13T01:05:31.730000", "content": "Hi @amazer -- great investigative work! The community was able to fully identify this issue and I didn't want to throw a wrench in the mix. Thanks for your patience.
", "votes": 0, "replies": [] } ] }, { "id": 651418, "author_name": "Nanashi", "author_url": "", "post_date": "2019-10-17T13:25:44.963000", "content": "\n\nas you can see at my kernel Starter ⚡ Great Energy Predictor \nI explain this.
\n\nHow do you interpret meter_reading=0? no consume?
\n\nIf you check the builing 0 (building_id = 0) you will see:\n- there is no information between January - May 2016 (months 1 to 5). meter_reading==0.\n- From June 2016 (month=6) we have information!
In my opinion you started to measure the consume of that specific building in June 2016. \nIn order to complete the database from 2016, you filled the other months with 0.
\n\nas @dmitriyguller said:
\n\n\n\n\nIgnoring 0-values won't take care of the problem, it will make the problem different. Instead of under-estimating the energy use for a building with a stuttering meter, it will over-estimate it. Another problem is that we don't actually know whether zeroes are ignored or not, so how are we supposed to build the model intelligently?\n If I were to guess, I would say that the big CV-LB differences are due to either different behavior of zeroes in the test set, or different treatment of them. This is why I'm staying out of this competition: the bad choice of metric is going to turn it into nonsense. You're going to get a much bigger bang for the buck figuring out how to deal with zeroes than you will figuring out how to build an intelligent model.
\n
probably you can't answer, but I had to try :)
", "votes": 21, "replies": [ { "id": 717285, "author_name": "Clayton Miller", "author_url": "", "post_date": "2020-01-13T01:08:03.400000", "content": "@jesucristo -- great question! In the field, a 0 reading could mean a lot of things. Often it simply means no consumption, but also common is that the meter has flat-lined and it is an error.
@claytonmiller Could you double check that the timestamps for weather data and meter reading are using the same timezone? My impression is that meter readings are using local time while weather data is using some other time zone.
\n\nSee the thread here: https://www.kaggle.com/c/ashrae-energy-prediction/discussion/114447#latest-658783
", "votes": 12, "replies": [ { "id": 662419, "author_name": "S D", "author_url": "", "post_date": "2019-10-31T14:31:28.457000", "content": "@sohier Maybe you could look into this?
", "votes": 0, "replies": [] }, { "id": 664438, "author_name": "arnab", "author_url": "", "post_date": "2019-11-03T17:04:06.353000", "content": "I am assuming that this additional complexity has been introduced to make the challenge interesting.
", "votes": 0, "replies": [] }, { "id": 717284, "author_name": "Clayton Miller", "author_url": "", "post_date": "2020-01-13T01:06:33.177000", "content": "@sdoria -- See previous posting and my reply
", "votes": 0, "replies": [] } ] }, { "id": 652783, "author_name": "Max Halford", "author_url": "", "post_date": "2019-10-19T12:06:45.447000", "content": "The goal of this competition (i.e. building a counterfactual model) is very reminiscent of Google's CausalImpact.
", "votes": 4, "replies": [] }, { "id": 650791, "author_name": "Md Khairul Islam", "author_url": "", "post_date": "2019-10-16T17:22:25.777000", "content": "Just out of curiosity , if it is ASHRAE - Great Energy Predictor III , where are Predictor I and Predictor II ?? 🙄
", "votes": 4, "replies": [ { "id": 651025, "author_name": "Clayton Miller", "author_url": "", "post_date": "2019-10-17T01:12:10.647000", "content": "Check out this overview of the previous competitions: https://www.kaggle.com/c/ashrae-energy-prediction/overview/prior-competitions
\n\nThey were held in the 90's and the first one even had the contestants receiving floppy disks in the mail to compete :)
", "votes": 6, "replies": [] }, { "id": 653603, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2019-10-20T17:48:35.023000", "content": "What Clayton has mentioned elsewhere - Kaggle has the floppy disk data available for you to see if todays methods can beat the 90's.
", "votes": 2, "replies": [] }, { "id": 654203, "author_name": "Sohier Dane", "author_url": "", "post_date": "2019-10-21T15:33:50.277000", "content": "", "votes": 1, "replies": [] } ] }, { "id": 650131, "author_name": "عثمان", "author_url": "", "post_date": "2019-10-16T06:17:19.153000", "content": "Looking forward to an interesting competition, @claytonmiller!
\n\nSome kagglers in another thread expressed concern regards to how 'production usable' models out of this competition would be if they incorporated weather information—in reality, only forecasts would be known at inference, yet we're seemingly given actual accurate weather information for up to two years into the future.
\n\nHow do we reconcile this? If weather truly impacts energy usage, then model output should correlate to weather and without an accurate future two-year forecast, ASHRAE won't be able to do much with the results.
", "votes": 4, "replies": [ { "id": 650256, "author_name": "arnab", "author_url": "", "post_date": "2019-10-16T08:23:43.817000", "content": "Ideally, the goal of the competition should have been to predict the energy usage 'X' days or 'W' weeks in advance, i.e. standing today, my model should be able to predict the energy consumption of the Xth day in future. That way, I don't the need the weather on the Xth day.
\n\nWill be willing to hear your thoughts @claytonmiller
", "votes": -1, "replies": [] }, { "id": 650372, "author_name": "arnab", "author_url": "", "post_date": "2019-10-16T09:58:46.047000", "content": "I think I kind of follow the purpose:
\n\nDeveloping energy savings has two key elements: Forecasting future energy usage without improvements, and forecasting energy use after a specific set of improvements have been implemented, like the installation and purchase of investment-grade meters, whose prices continue to fall.
So, here we are trying to \"forecast energy use after a specific set of improvements have been implemented\" and that's why the weather data is available.
", "votes": 2, "replies": [] }, { "id": 650949, "author_name": "Martin Elfstadius", "author_url": "", "post_date": "2019-10-16T21:43:36.933000", "content": "As per publication provided by @claytonmiller
\n\n\"For example, long-term forecasting is usually retroactive as the energy savings prediction is focused on what energy the building would have consumed if the intervention had not taken place. This situation means that input variables such as weather are known and not forecasts themselves. \"
\n\nwhich I think answers your question?\nBr,\nMartin
", "votes": 4, "replies": [] }, { "id": 651024, "author_name": "Clayton Miller", "author_url": "", "post_date": "2019-10-17T01:10:31.890000", "content": "Hey @authman and @arnabbiswas1 -- @diginfo hit the nail on the head with his explanation of the goal of long term prediction in this context -- you can find more information from this publication: https://www.mdpi.com/2504-4990/1/3/56
\n\nThere is a whole domain of performance contracting which uses retroactive long term prediction to quantify what energy consumption would have potentially occurred without an energy savings intervention
", "votes": 3, "replies": [] }, { "id": 653607, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2019-10-20T17:54:10.163000", "content": "In 2009 I retired and was thinking about a new city/state to live in. Being a data driver xx?% kind of guy I downloaded all the weather data for USA for a decade. Than looked for city where it never got too hot or too cold.
\n\nWhen you plot a decade worth of temperatures for a city, the curve is very repeating. So making a model of the next decade of weather for a city is pretty much only an issue of getting the data set. Can't of course model climate change, but otherwise not that big of a task for design. I would assume that such a data set can in fact be purchased so folks designing buildings know what cooling and heating load will occur.
", "votes": 1, "replies": [] } ] }, { "id": 692066, "author_name": "Przemysław Rejczak", "author_url": "", "post_date": "2019-12-10T20:28:22.147000", "content": "Hi! I ve a question about metering. Does every building include every meter that it orginaly have? Can we assume that if building has only electricity meter in the competition heating and cooling services are also provided by electricity?\nWith Regards
", "votes": 1, "replies": [] }, { "id": 651115, "author_name": "Craig William", "author_url": "", "post_date": "2019-10-17T04:04:21.130000", "content": "Predictor III in the wild. Its called calculus, I have worked in construction for years, The Idea that its colder at night then in the day seems to blow these peoples minds away.
", "votes": 1, "replies": [] }, { "id": 653529, "author_name": "S D", "author_url": "", "post_date": "2019-10-20T15:36:42.363000", "content": "Do some of the buildings use a source of heat that is not accounted for in the meters?
", "votes": 2, "replies": [ { "id": 717291, "author_name": "Clayton Miller", "author_url": "", "post_date": "2020-01-13T01:10:55.890000", "content": "yes, that is possible
", "votes": 0, "replies": [] } ] }, { "id": 656293, "author_name": "Atharva Patel", "author_url": "", "post_date": "2019-10-24T05:55:28.383000", "content": "To those considering the best algorithm to start with, consider starting with LightGBM.
", "votes": -1, "replies": [] }, { "id": 664792, "author_name": "", "author_url": "", "post_date": "2019-11-04T08:25:30.747000", "content": "", "votes": 0, "replies": [] }, { "id": 658445, "author_name": "", "author_url": "", "post_date": "2019-10-26T02:32:59.570000", "content": "Hi @claytonmiller, could you provide some details around the meter time measurements and how they correspond to the weather measurements on a site-by-site basis? There could be a consistent measurement methodology that you folks have employed (and are assuming), but in the wild, all that stuff depends on the utility and can vary greatly. Similarly, I'm assuming the readings are in kWh and temperature appears to be in celsius, but I don't really know. What I'm afraid of is that there's an industrial site or big data center mixed in this dataset that has a substation level MWh meter. This would throw off the models if one is trying to view information in aggregate.
\n\nThanks!
", "votes": 5, "replies": [ { "id": 717290, "author_name": "Clayton Miller", "author_url": "", "post_date": "2020-01-13T01:10:30.637000", "content": "We tried to make all of the reading the same unit (kWh) to prevent mix-ups as you mention and the weather units are posted. We made a mistake in which one of the site's wasn't converted correctly though -- it was in the updated documentation.
", "votes": 0, "replies": [] } ] }, { "id": 653889, "author_name": "", "author_url": "", "post_date": "2019-10-21T05:52:48.413000", "content": "", "votes": 0, "replies": [] }, { "id": 653306, "author_name": "TianBaojie", "author_url": "", "post_date": "2019-10-20T08:07:10.773000", "content": "Thanks
", "votes": 1, "replies": [] }, { "id": 650660, "author_name": "Naruhiko Nakanishi", "author_url": "", "post_date": "2019-10-16T15:22:10.777000", "content": "Thanks for the hosting.
", "votes": -1, "replies": [] } ], "raw_markdown_by_id": { "649970": "Welcome to the ASHRAE Great Energy Predictor III Competition! I represent the lead member of the ASHRAE Technical Team for this competition.\n\nThis challenge provides the opportunity to see how buildings actually use energy in the wild and come up with new ways to model their hourly consumption in a scalable way. Please check out [this publication](https://www.mdpi.com/2504-4990/1/3/56) for an understanding of why this effort is necessary and some of the basics of energy prediction for buildings!\n\nPlease feel free to ask your questions in this thread.\n\nGood luck!", "656125": "Questions concerning weather data and timestamp.\nCan you validate the weather data for all sites? (site_ids 0 through 15)\n\nThe weather data appears inconsistent with the timestamp. For example, at site 4, the peak daily temperature occurs between 9PM and midnight, while the daily low temperatures occur between 1PM and 3 PM. This is not an occasional occurrence, but a pattern that is evident throughout the year. Actual daily lows occur around the hour before sunrise, not 2PM. Actual daily highs occur in the mid-afternoon, not at night time. As the data is at odds with data from the national weather service (I can't find any data on the NWS site that supports temperature profiles like the ones given), it appears to be in error.\n\nThis calls into question weather at other sites. A glance reveals some patterns that are once again at odds with actual weather patterns. For example, at site 0 the temperature decreases well past sunrise, while at site 1, the temperature begins to increase well before sunrise.\n\nThe meter reading profiles appear consistent with consumption patterns from other data sources. Perhaps the times for the weather data are not local times, but then the weather profile and the meter reading profile are not in synch. Sometimes weather is provided not in local time but by GMT (London time). If this is the case, then the sites are not all in the USA, but lie in other countries. (That would create another problem because we would not know which holiday schedule to apply.)\n\nOne minor point, hourly meter readings normally indicate consumption at the end of the hour. I.e. a reading for 0200 indicates the consumption from 0100 to 0200. In this case, the annual data should begin with timestamp 2016-01-01 0100 and end with timestamp 2017-0101 0000. But your training sets all end at 2016-12-31 2300 which does not account for the final hour of the year.\n\nPlease give some insight on the weather and timestamps so that we can synchronize the meter reading and weather profiles.\n\nThank you", "651418": "Hi @claytonmiller \n\nas you can see at my kernel [Starter ⚡ Great Energy Predictor ](https://www.kaggle.com/jesucristo/starter-great-energy-predictor)\nI explain this.\n\n**How do you interpret meter_reading=0?** no consume?\n\nIf you check the builing ```0``` (building_id = 0) you will see:\n- there is no information between January - May 2016 (months 1 to 5). meter_reading==0.\n- From June 2016 (month=6) we have information! \n\nIn my opinion you started to measure the consume of that specific building in June 2016. \nIn order to complete the database from 2016, you filled the other months with 0.\n\nas @dmitriyguller said:\n> Ignoring 0-values won't take care of the problem, it will make the problem different. Instead of under-estimating the energy use for a building with a stuttering meter, it will over-estimate it. Another problem is that we don't actually know whether zeroes are ignored or not, so how are we supposed to build the model intelligently?\n>If I were to guess, I would say that the big CV-LB differences are due to either different behavior of zeroes in the test set, or different treatment of them. This is why I'm staying out of this competition: the bad choice of metric is going to turn it into nonsense. **You're going to get a much bigger bang for the buck figuring out how to deal with zeroes than you will figuring out how to build an intelligent model.**\n\nprobably you can't answer, but I had to try :)", "658787": "@claytonmiller Could you double check that the timestamps for weather data and meter reading are using the same timezone? My impression is that meter readings are using local time while weather data is using some other time zone.\n\nSee the thread here: https://www.kaggle.com/c/ashrae-energy-prediction/discussion/114447#latest-658783", "652783": "The goal of this competition (i.e. building a counterfactual model) is very reminiscent of Google's [CausalImpact](https://google.github.io/CausalImpact/).", "650791": "Just out of curiosity , if it is ASHRAE - Great Energy Predictor III , where are Predictor I and Predictor II ?? 🙄 ", "650131": "Looking forward to an interesting competition, @claytonmiller!\n\nSome kagglers in another thread expressed concern regards to how 'production usable' models out of this competition would be if they incorporated weather information—in reality, only forecasts would be known at inference, yet we're seemingly given actual accurate weather information for up to two years into the future.\n\nHow do we reconcile this? If weather truly impacts energy usage, then model output should correlate to weather and without an accurate future two-year forecast, ASHRAE won't be able to do much with the results.", "692066": "Hi! I ve a question about metering. Does every building include every meter that it orginaly have? Can we assume that if building has only electricity meter in the competition heating and cooling services are also provided by electricity?\nWith Regards", "651115": "Predictor III in the wild. Its called calculus, I have worked in construction for years, The Idea that its colder at night then in the day seems to blow these peoples minds away.", "653529": "Do some of the buildings use a source of heat that is not accounted for in the meters?", "656293": "To those considering the best algorithm to start with, consider starting with LightGBM.", "664792": "", "658445": "Hi @claytonmiller, could you provide some details around the meter time measurements and how they correspond to the weather measurements on a site-by-site basis? There could be a consistent measurement methodology that you folks have employed (and are assuming), but in the wild, all that stuff depends on the utility and can vary greatly. Similarly, I'm assuming the readings are in kWh and temperature appears to be in celsius, but I don't really know. What I'm afraid of is that there's an industrial site or big data center mixed in this dataset that has a substation level MWh meter. This would throw off the models if one is trying to view information in aggregate.\n\nThanks!\n\n", "653889": "", "653306": "Thanks", "650660": "Thanks for the hosting." } }