Moving along with the competition, just need to know the views which is the best way to tackle the dataset and proceed with modelling!!
", "messages": [ { "id": "650374", "postDate": "10/16/2019 09:59:24", "content": "Moving along with the competition, just need to know the views which is the best way to tackle the dataset and proceed with modelling!!
", "rawMarkdown": "Moving along with the competition, just need to know the views which is the best way to tackle the dataset and proceed with modelling!!", "votes": null }, { "id": "650567", "postDate": "10/16/2019 14:09:45", "content": "Hello Rajat, as we are considering a target that \"moves\" considering a time variable (timestamp), the Time Series behaves better than Regression, but nothing prevents us from testing all approaches because it will always depend on the problem. business to be solved, as well as the data that is available for modeling.
", "rawMarkdown": "Hello Rajat, as we are considering a target that \"moves\" considering a time variable (timestamp), the Time Series behaves better than Regression, but nothing prevents us from testing all approaches because it will always depend on the problem. business to be solved, as well as the data that is available for modeling.", "votes": null }, { "id": "650575", "postDate": "10/16/2019 14:16:14", "content": "IMHO we should start with considering it as a Regression Problem for the following reason:
\n\nThe data doesn't only contain multiple time series features related to weather, but it consists of building meta data as well which is static in nature (don't vary with time).
\n\nI will say this is a regression problem with time as a feature. The problem is we can't use timestamps as it is. We have to break time stamp into day, month, hour, minute etc. Please look into the winning solutions of \"Power Laws: Forecasting Energy Consumption - By DrivenData\" mentioned here. That will give you further idea.
", "rawMarkdown": "IMHO we should start with considering it as a Regression Problem for the following reason:\n\nThe data doesn't only contain multiple time series features related to weather, but it consists of building meta data as well which is static in nature (don't vary with time).\n\nI will say this is a regression problem with time as a feature. The problem is we can't use timestamps as it is. We have to break time stamp into day, month, hour, minute etc. Please look into the winning solutions of \"Power Laws: Forecasting Energy Consumption - By DrivenData\" mentioned [here](https://www.kaggle.com/c/ashrae-energy-prediction/discussion/112855#latest-649925). That will give you further idea.", "votes": null }, { "id": "650585", "postDate": "10/16/2019 14:25:09", "content": "@arnabbiswas1 thanks for the link, i had went through that in your previous thread and tried the details, and its quite inspiring, although the top solutions had more than 2-3 models, which was quite unusual for me as a beginner, could you shed some light in that? and its advantages
", "rawMarkdown": "arnabbiswas1 thanks for the link, i had went through that in your previous thread and tried the details, and its quite inspiring, although the top solutions had more than 2-3 models, which was quite unusual for me as a beginner, could you shed some light in that? and its advantages", "votes": null }, { "id": "650587", "postDate": "10/16/2019 14:25:46", "content": "yes, the field is vast and we should try different approaches as well, will keep it in mind
", "rawMarkdown": "yes, the field is vast and we should try different approaches as well, will keep it in mind", "votes": null }, { "id": "650591", "postDate": "10/16/2019 14:29:20", "content": "Different models could help you predict on different data. \nSuppose model 1 is able to predict a+b = c\nand model 2 is able to predict a + f = g\nbut vice versa isn't, so if you use both it'll be always helpful.
", "rawMarkdown": "Different models could help you predict on different data. \nSuppose model 1 is able to predict a+b = c\nand model 2 is able to predict a + f = g\nbut vice versa isn't, so if you use both it'll be always helpful.", "votes": null }, { "id": "650605", "postDate": "10/16/2019 14:39:59", "content": "Welcome to Kaggle. When you are not sure which algorithm you should try first, go for LightGBM or XGBoost. :-)
", "rawMarkdown": "Welcome to Kaggle. When you are not sure which algorithm you should try first, go for LightGBM or XGBoost. :-)", "votes": null } ], "comments": [ { "id": 650567, "author_name": "josebruno", "author_url": "", "post_date": "10/16/2019 14:09:45", "content": "Hello Rajat, as we are considering a target that \"moves\" considering a time variable (timestamp), the Time Series behaves better than Regression, but nothing prevents us from testing all approaches because it will always depend on the problem. business to be solved, as well as the data that is available for modeling.
", "votes": null, "replies": [ { "id": 650587, "author_name": "rajatranjan", "author_url": "", "post_date": "10/16/2019 14:25:46", "content": "yes, the field is vast and we should try different approaches as well, will keep it in mind
", "votes": null, "replies": [] } ] }, { "id": 650575, "author_name": "arnabbiswas1", "author_url": "", "post_date": "10/16/2019 14:16:14", "content": "IMHO we should start with considering it as a Regression Problem for the following reason:
\n\nThe data doesn't only contain multiple time series features related to weather, but it consists of building meta data as well which is static in nature (don't vary with time).
\n\nI will say this is a regression problem with time as a feature. The problem is we can't use timestamps as it is. We have to break time stamp into day, month, hour, minute etc. Please look into the winning solutions of \"Power Laws: Forecasting Energy Consumption - By DrivenData\" mentioned here. That will give you further idea.
", "votes": null, "replies": [ { "id": 650585, "author_name": "rajatranjan", "author_url": "", "post_date": "10/16/2019 14:25:09", "content": "@arnabbiswas1 thanks for the link, i had went through that in your previous thread and tried the details, and its quite inspiring, although the top solutions had more than 2-3 models, which was quite unusual for me as a beginner, could you shed some light in that? and its advantages
", "votes": null, "replies": [] }, { "id": 650591, "author_name": "themonologue", "author_url": "", "post_date": "10/16/2019 14:29:20", "content": "Different models could help you predict on different data. \nSuppose model 1 is able to predict a+b = c\nand model 2 is able to predict a + f = g\nbut vice versa isn't, so if you use both it'll be always helpful.
", "votes": null, "replies": [] }, { "id": 650605, "author_name": "arnabbiswas1", "author_url": "", "post_date": "10/16/2019 14:39:59", "content": "Welcome to Kaggle. When you are not sure which algorithm you should try first, go for LightGBM or XGBoost. :-)
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "650374": "Moving along with the competition, just need to know the views which is the best way to tackle the dataset and proceed with modelling!!", "650567": "Hello Rajat, as we are considering a target that \"moves\" considering a time variable (timestamp), the Time Series behaves better than Regression, but nothing prevents us from testing all approaches because it will always depend on the problem. business to be solved, as well as the data that is available for modeling.", "650575": "IMHO we should start with considering it as a Regression Problem for the following reason:\n\nThe data doesn't only contain multiple time series features related to weather, but it consists of building meta data as well which is static in nature (don't vary with time).\n\nI will say this is a regression problem with time as a feature. The problem is we can't use timestamps as it is. We have to break time stamp into day, month, hour, minute etc. Please look into the winning solutions of \"Power Laws: Forecasting Energy Consumption - By DrivenData\" mentioned [here](https://www.kaggle.com/c/ashrae-energy-prediction/discussion/112855#latest-649925). That will give you further idea.", "650585": "arnabbiswas1 thanks for the link, i had went through that in your previous thread and tried the details, and its quite inspiring, although the top solutions had more than 2-3 models, which was quite unusual for me as a beginner, could you shed some light in that? and its advantages", "650587": "yes, the field is vast and we should try different approaches as well, will keep it in mind", "650591": "Different models could help you predict on different data. \nSuppose model 1 is able to predict a+b = c\nand model 2 is able to predict a + f = g\nbut vice versa isn't, so if you use both it'll be always helpful.", "650605": "Welcome to Kaggle. When you are not sure which algorithm you should try first, go for LightGBM or XGBoost. :-)" }, "source": "meta" }