Sorry for the beginner question but:
\nHow can I use a single model for all the sites?
Sorry for the beginner question but:
\nHow can I use a single model for all the sites?
What do you mean by single model . Can you explain more details plz🙏
", "rawMarkdown": "What do you mean by single model . Can you explain more details plz🙏", "votes": null }, { "id": "1219418", "postDate": "02/26/2021 19:49:55", "content": "I am not 100% sure but I think currently most of us using (site specific) 24 different models. However, there is a way to train only one model to handle this problem instead of 24.
", "rawMarkdown": "I am not 100% sure but I think currently most of us using (site specific) 24 different models. However, there is a way to train only one model to handle this problem instead of 24.", "votes": null }, { "id": "1219456", "postDate": "02/26/2021 20:53:30", "content": "You could combine everything into one model by treating the issue as a spatial statistics problem and then working to better understand how covariation changes with distance. From there you could add predictors to better understand what the result looks like given a different location
", "rawMarkdown": "You could combine everything into one model by treating the issue as a spatial statistics problem and then working to better understand how covariation changes with distance. From there you could add predictors to better understand what the result looks like given a different location", "votes": null }, { "id": "1219587", "postDate": "02/27/2021 01:10:17", "content": "Hi, @bayartsogtya ! Nice to see you here as well. If you remember me in MoA comp. 😄
\nI actually achieved the current score with so called \"single model\", non-site-specific model.
\nFor now, I'm using only wifi features. To do this, we have to change the shape of dataset. I believe there are bunch of ways to do this.
I still trying to find better ways.
", "rawMarkdown": "Hi, @bayartsogtya ! Nice to see you here as well. If you remember me in MoA comp. 😄\n\nI actually achieved the current score with so called \"single model\", non-site-specific model. \nFor now, I'm using only wifi features. To do this, we have to change the shape of dataset. I believe there are bunch of ways to do this.\n\nI still trying to find better ways.", "votes": null }, { "id": "1219758", "postDate": "02/27/2021 08:02:43", "content": "Hi @kokitanisaka , I dont see a need to reshape the data. Putting everything into one \"master\" frame should be enough, but maybe I am missing something. Are you using a sequence model? (if you are willing to disclose 🙂)
", "rawMarkdown": "Hi @kokitanisaka , I dont see a need to reshape the data. Putting everything into one \"master\" frame should be enough, but maybe I am missing something. Are you using a sequence model? (if you are willing to disclose 🙂)", "votes": null }, { "id": "1219847", "postDate": "02/27/2021 09:46:48", "content": "I assumed that people who building site specific models are using @devinanzelmo 's dataset or @hiro5299834 's dataset.
\nThese are really useful but if we want to make prediction with one model with these data, we need to make extra effort, am I right? Because every sites have different number of features. That means, we can't just concatenate all the date without thinking.
So the meaning of \"reshape\" here is, somehow combine every sites data into one \"master\" frame. To do this, I think there should be many ways.
", "rawMarkdown": "I assumed that people who building site specific models are using @devinanzelmo 's [dataset](https://www.kaggle.com/devinanzelmo/indoor-navigation-and-location-wifi-features) or @hiro5299834 's [dataset](https://www.kaggle.com/hiro5299834/indoor-navigation-and-location-wifi-features). \nThese are really useful but if we want to make prediction with one model with these data, we need to make extra effort, am I right? Because every sites have different number of features. That means, we can't just concatenate all the date without thinking. \n\nSo the meaning of \"reshape\" here is, somehow combine every sites data into one \"master\" frame. To do this, I think there should be many ways.", "votes": null }, { "id": "1220327", "postDate": "02/27/2021 21:10:31", "content": "\n\n\nNice to see you here as well. If you remember me in MoA comp. 😄
\n
Of course I remember your helpful resources and great solo performance!
\n\n\nI believe there are bunch of ways to do this.
\n
Thank you so much for your detailed answer. It feels like I need to work on data processing instead of modeling!
", "rawMarkdown": "kokitanisaka \n\n> Nice to see you here as well. If you remember me in MoA comp. 😄\n\nOf course I remember your helpful resources and great solo performance!\n\n> I believe there are bunch of ways to do this.\n\nThank you so much for your detailed answer. It feels like I need to work on data processing instead of modeling!", "votes": null }, { "id": "1220378", "postDate": "02/27/2021 22:48:03", "content": "\n\nIt feels like I need to work on data processing instead of modeling!
\n
I think so too!
\nIt's still in early stage, we can put effort on understanding data and making data for modeling. 😄
What do you mean by single model . Can you explain more details plz🙏
", "votes": null, "replies": [ { "id": 1219418, "author_name": "bayartsogtya", "author_url": "", "post_date": "02/26/2021 19:49:55", "content": "I am not 100% sure but I think currently most of us using (site specific) 24 different models. However, there is a way to train only one model to handle this problem instead of 24.
", "votes": null, "replies": [] } ] }, { "id": 1219456, "author_name": "nosnibor", "author_url": "", "post_date": "02/26/2021 20:53:30", "content": "You could combine everything into one model by treating the issue as a spatial statistics problem and then working to better understand how covariation changes with distance. From there you could add predictors to better understand what the result looks like given a different location
", "votes": null, "replies": [] }, { "id": 1219587, "author_name": "kokitanisaka", "author_url": "", "post_date": "02/27/2021 01:10:17", "content": "Hi, @bayartsogtya ! Nice to see you here as well. If you remember me in MoA comp. 😄
\nI actually achieved the current score with so called \"single model\", non-site-specific model.
\nFor now, I'm using only wifi features. To do this, we have to change the shape of dataset. I believe there are bunch of ways to do this.
I still trying to find better ways.
", "votes": null, "replies": [ { "id": 1219758, "author_name": "nicohrubec", "author_url": "", "post_date": "02/27/2021 08:02:43", "content": "Hi @kokitanisaka , I dont see a need to reshape the data. Putting everything into one \"master\" frame should be enough, but maybe I am missing something. Are you using a sequence model? (if you are willing to disclose 🙂)
", "votes": null, "replies": [] }, { "id": 1219847, "author_name": "kokitanisaka", "author_url": "", "post_date": "02/27/2021 09:46:48", "content": "I assumed that people who building site specific models are using @devinanzelmo 's dataset or @hiro5299834 's dataset.
\nThese are really useful but if we want to make prediction with one model with these data, we need to make extra effort, am I right? Because every sites have different number of features. That means, we can't just concatenate all the date without thinking.
So the meaning of \"reshape\" here is, somehow combine every sites data into one \"master\" frame. To do this, I think there should be many ways.
", "votes": null, "replies": [] }, { "id": 1220327, "author_name": "bayartsogtya", "author_url": "", "post_date": "02/27/2021 21:10:31", "content": "\n\n\nNice to see you here as well. If you remember me in MoA comp. 😄
\n
Of course I remember your helpful resources and great solo performance!
\n\n\nI believe there are bunch of ways to do this.
\n
Thank you so much for your detailed answer. It feels like I need to work on data processing instead of modeling!
", "votes": null, "replies": [] }, { "id": 1220378, "author_name": "kokitanisaka", "author_url": "", "post_date": "02/27/2021 22:48:03", "content": "\n\nIt feels like I need to work on data processing instead of modeling!
\n
I think so too!
\nIt's still in early stage, we can put effort on understanding data and making data for modeling. 😄