\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4011140%2Ff1738186ed280f76a342579774163f1d%2FCoverPhoto.png?generation=1580338211001143&alt=media\")
Hello,
\n\nI found this project/competition really interesting and fun. I didn't get to see it until after it was over, but spent the last couple weeks working through it. I came up with a method by Site ID that might be of interest. I wrote an article on Medium.com that was published by Towards Data Science today for it.
\n
I realize I could have done a better job cleaning the data, training the ML model's hyperparameters, and also allocating the time series data better (it appears I overfit some), but all in all, this is a pretty good framework to go from if you want to have a modular approach to the project.
\n\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4011140%2Fd593a86880a558429e32529c79c1e61b%2FASHRAE_FlowChart.png?generation=1580337918091243&alt=media\")
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4011140%2Fb1a3eb76097fa871a631df5e29298972%2FASHRAE_Part1.png?generation=1580337941211682&alt=media\")
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4011140%2Ff056c80d54cfc593c5b49d957e585c7f%2FASHRAE_Part2.png?generation=1580338299428909&alt=media\")
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4011140%2F2c04f1ace9650245cf9359945678494e%2FASHRAE_Part3.png?generation=1580337951460252&alt=media\")
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4011140%2Ff4a0d697a6aaa0537da8b506c0b1873b%2FASHRAE_Part4.png?generation=1580337961529898&alt=media\")
The Jupyter Notebooks I made can be found here:\nhttps://github.com/stevensmiley1989/ASHRAE-for-ML
\n\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4011140%2Ff11d2aaaf2789d5abafc89495d6d46d5%2FFigure_4-0-0-Site%20ID%200.png?generation=1580340379991933&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4011140%2Fd3e4481d20a27ac4d3bfe08091e08e3e%2FFigure_4-1-0-Site%20ID%201.png?generation=1580340391139859&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4011140%2F7976d197a15115ebdb862f94ea5f8dce%2FFigure_4-2-0-Site%20ID%202.png?generation=1580340387309357&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4011140%2F0398ecc65ae6e8329e37c60681c70a47%2FFigure_4-3-0-Site%20ID%203.png?generation=1580340393348954&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4011140%2F15b30974de6a8f162a3eafc2814a9424%2FFigure_4-4-0-Site%20ID%204.png?generation=1580340393477833&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4011140%2F1ee24373766e273f9a29f3442f438456%2FFigure_4-5-0-Site%20ID%205.png?generation=1580340392792566&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4011140%2F9382206de35d90b4b3f79e771dfd74e9%2FFigure_4-6-0-Site%20ID%206.png?generation=1580340387842502&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4011140%2Fe2ca937f40c55b78c731ba0c4a54d1f9%2FFigure_4-7-0-Site%20ID%207.png?generation=1580340392250103&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4011140%2F953d84c63426fdcf6b67e72a343a759a%2FFigure_4-8-0-Site%20ID%208.png?generation=1580340386059371&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4011140%2F3360940a84b917b7c7676c5ae3a6f6a0%2FFigure_4-9-0-Site%20ID%209.png?generation=1580340376874069&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4011140%2F4c7177cc59124f0560d90b9b3ad7b14c%2FFigure_4-10-0-Site%20ID%2010.png?generation=1580340391242109&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4011140%2Fe2b809f083a634423e679057cf828913%2FFigure_4-11-0-Site%20ID%2011.png?generation=1580340391616666&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4011140%2Fdd66cefeec78cf6fc4459d3720d80471%2FFigure_4-12-0-Site%20ID%2012.png?generation=1580340393846696&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4011140%2F75301d02b28f3557aa2cc93d39959734%2FFigure_4-13-0-Site%20ID%2013.png?generation=1580340389723397&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4011140%2F3db861b140296e12762898bdd9254c76%2FFigure_4-14-0-Site%20ID%2014.png?generation=1580340390914676&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4011140%2F77850758b0158cf4192416dd13ed984e%2FFigure_4-15-0-Site%20ID%2015.png?generation=1580340388315981&alt=media\")
Hello,
\n\nI found this project/competition really interesting and fun. I didn't get to see it until after it was over, but spent the last couple weeks working through it. I came up with a method by Site ID that might be of interest. I wrote an article on Medium.com that was published by Towards Data Science today for it.
\n
I realize I could have done a better job cleaning the data, training the ML model's hyperparameters, and also allocating the time series data better (it appears I overfit some), but all in all, this is a pretty good framework to go from if you want to have a modular approach to the project.
\n\n
The Jupyter Notebooks I made can be found here:\nhttps://github.com/stevensmiley1989/ASHRAE-for-ML
\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
Great\nAmazing Work\nThanks for Sharing!! @stevensmiley
", "rawMarkdown": "Great\nAmazing Work\nThanks for Sharing!! @stevensmiley", "votes": null }, { "id": "732754", "postDate": "01/30/2020 08:07:06", "content": "Thank you š
", "rawMarkdown": "Thank you š", "votes": null }, { "id": "732777", "postDate": "01/30/2020 08:54:58", "content": "This is great! I still have a query. Shouldnt one even segregate the values by sensorID, and then run TimeSeriesModeling on each. Each sensor is different rt.
", "rawMarkdown": "This is great! I still have a query. Shouldnt one even segregate the values by sensorID, and then run TimeSeriesModeling on each. Each sensor is different rt.", "votes": null }, { "id": "732789", "postDate": "01/30/2020 09:08:58", "content": "Thank you! Iām assuming you mean meter ID? Like electric, steam, hot water or chilled water? Yes, that probably would work. I actually think it would be best to perform kmeans cluster with a silhouette value to determine the optimal clusters for grouping. I just skipped that step and went with the Site ID.
", "rawMarkdown": "Thank you! Iām assuming you mean meter ID? Like electric, steam, hot water or chilled water? Yes, that probably would work. I actually think it would be best to perform kmeans cluster with a silhouette value to determine the optimal clusters for grouping. I just skipped that step and went with the Site ID.", "votes": null }, { "id": "747294", "postDate": "02/16/2020 08:11:02", "content": "Excellent documentation @stevensmiley! We are planning to add this solution to our directory of solutions. thanks again!
", "rawMarkdown": "Excellent documentation @stevensmiley! We are planning to add this solution to our directory of solutions. thanks again!", "votes": null }, { "id": "747531", "postDate": "02/16/2020 14:51:42", "content": "Thank you! I really enjoyed doing it!
", "rawMarkdown": "Thank you! I really enjoyed doing it!", "votes": null } ], "comments": [ { "id": 732708, "author_name": "veeralakrishna", "author_url": "", "post_date": "01/30/2020 06:28:27", "content": "Great\nAmazing Work\nThanks for Sharing!! @stevensmiley
", "votes": null, "replies": [ { "id": 732754, "author_name": "stevensmiley", "author_url": "", "post_date": "01/30/2020 08:07:06", "content": "Thank you š
", "votes": null, "replies": [] } ] }, { "id": 732777, "author_name": "anirbank", "author_url": "", "post_date": "01/30/2020 08:54:58", "content": "This is great! I still have a query. Shouldnt one even segregate the values by sensorID, and then run TimeSeriesModeling on each. Each sensor is different rt.
", "votes": null, "replies": [ { "id": 732789, "author_name": "stevensmiley", "author_url": "", "post_date": "01/30/2020 09:08:58", "content": "Thank you! Iām assuming you mean meter ID? Like electric, steam, hot water or chilled water? Yes, that probably would work. I actually think it would be best to perform kmeans cluster with a silhouette value to determine the optimal clusters for grouping. I just skipped that step and went with the Site ID.
", "votes": null, "replies": [] } ] }, { "id": 747294, "author_name": "claytonmiller", "author_url": "", "post_date": "02/16/2020 08:11:02", "content": "Excellent documentation @stevensmiley! We are planning to add this solution to our directory of solutions. thanks again!
", "votes": null, "replies": [ { "id": 747531, "author_name": "stevensmiley", "author_url": "", "post_date": "02/16/2020 14:51:42", "content": "Thank you! I really enjoyed doing it!
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "732518": "Hello,\n\nI found this project/competition really interesting and fun. I didn't get to see it until after it was over, but spent the last couple weeks working through it. I came up with a method by Site ID that might be of interest. I wrote an article on Medium.com that was published by [Towards Data Science](https://towardsdatascience.com/estimating-counterfactual-energy-usage-of-buildings-with-machine-learning-8ca91ec66c08) today for it. \n\n\nI realize I could have done a better job cleaning the data, training the ML model's hyperparameters, and also allocating the time series data better (it appears I overfit some), but all in all, this is a pretty good framework to go from if you want to have a modular approach to the project. \n\n\n\n\n\n\n\n\n\n\n\n\nThe Jupyter Notebooks I made can be found here:\nhttps://github.com/stevensmiley1989/ASHRAE-for-ML\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n", "732708": "Great\nAmazing Work\nThanks for Sharing!! @stevensmiley", "732754": "Thank you š", "732777": "This is great! I still have a query. Shouldnt one even segregate the values by sensorID, and then run TimeSeriesModeling on each. Each sensor is different rt.", "732789": "Thank you! Iām assuming you mean meter ID? Like electric, steam, hot water or chilled water? Yes, that probably would work. I actually think it would be best to perform kmeans cluster with a silhouette value to determine the optimal clusters for grouping. I just skipped that step and went with the Site ID.", "747294": "Excellent documentation @stevensmiley! We are planning to add this solution to our directory of solutions. thanks again!", "747531": "Thank you! I really enjoyed doing it!" }, "source": "meta" }