Do you do parameter tuning using the whole data set for this competition? \nI am trying to use GridSearchCV(sklearn) with the whole data set and with the data corresponding to each type, but this takes a very long time. Are there any quicker approaches?
", "messages": [ { "id": "579318", "postDate": "07/18/2019 18:26:24", "content": "Do you do parameter tuning using the whole data set for this competition? \nI am trying to use GridSearchCV(sklearn) with the whole data set and with the data corresponding to each type, but this takes a very long time. Are there any quicker approaches?
", "rawMarkdown": "Do you do parameter tuning using the whole data set for this competition? \nI am trying to use GridSearchCV(sklearn) with the whole data set and with the data corresponding to each type, but this takes a very long time. Are there any quicker approaches?", "votes": null }, { "id": "579562", "postDate": "07/19/2019 01:13:35", "content": "In parameter tuning, I use a part of data set (ex. 10% of data).
", "rawMarkdown": "In parameter tuning, I use a part of data set (ex. 10% of data).", "votes": null }, { "id": "579575", "postDate": "07/19/2019 01:53:35", "content": "This is a comment from @hengck23 in one of his discussions. It seems you can tune the parameter with small data set + overfitting
\n\n\n", "rawMarkdown": "This is a comment from @hengck23 in one of his discussions. It seems you can tune the parameter with small data set + overfitting\n\n> let me try to maximize the performance of MPNN first. I will try other network later (which will be better results). Till now, i still cannot overfit a set of small train data (i.e. overfit until kaggle loss is -3). Some of my training parameters/feature representation are not correct yet. Generally, my suggested approached is:\n1. Try to over fit small train set (i.e. loss is very low, preferably close to zero. this is the lowest bound of error. Since kaggler can get to LB -2.3, i expect overfitting to give -3 to -4)\n2. Try to generalize. See how to make sure results stays the same on validation set (and also LB test set)\n3. Try to stabilize. Estimate shakeup and prevent difference of LB public and private set.", "votes": null } ], "comments": [ { "id": 579562, "author_name": "oshidori", "author_url": "", "post_date": "07/19/2019 01:13:35", "content": "let me try to maximize the performance of MPNN first. I will try other network later (which will be better results). Till now, i still cannot overfit a set of small train data (i.e. overfit until kaggle loss is -3). Some of my training parameters/feature representation are not correct yet. Generally, my suggested approached is:\n 1. Try to over fit small train set (i.e. loss is very low, preferably close to zero. this is the lowest bound of error. Since kaggler can get to LB -2.3, i expect overfitting to give -3 to -4)\n 2. Try to generalize. See how to make sure results stays the same on validation set (and also LB test set)\n 3. Try to stabilize. Estimate shakeup and prevent difference of LB public and private set.
\n
In parameter tuning, I use a part of data set (ex. 10% of data).
", "votes": null, "replies": [] }, { "id": 579575, "author_name": "leptony", "author_url": "", "post_date": "07/19/2019 01:53:35", "content": "This is a comment from @hengck23 in one of his discussions. It seems you can tune the parameter with small data set + overfitting
\n\n\n", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "579318": "Do you do parameter tuning using the whole data set for this competition? \nI am trying to use GridSearchCV(sklearn) with the whole data set and with the data corresponding to each type, but this takes a very long time. Are there any quicker approaches?", "579562": "In parameter tuning, I use a part of data set (ex. 10% of data).", "579575": "This is a comment from @hengck23 in one of his discussions. It seems you can tune the parameter with small data set + overfitting\n\n> let me try to maximize the performance of MPNN first. I will try other network later (which will be better results). Till now, i still cannot overfit a set of small train data (i.e. overfit until kaggle loss is -3). Some of my training parameters/feature representation are not correct yet. Generally, my suggested approached is:\n1. Try to over fit small train set (i.e. loss is very low, preferably close to zero. this is the lowest bound of error. Since kaggler can get to LB -2.3, i expect overfitting to give -3 to -4)\n2. Try to generalize. See how to make sure results stays the same on validation set (and also LB test set)\n3. Try to stabilize. Estimate shakeup and prevent difference of LB public and private set." }, "source": "meta" }let me try to maximize the performance of MPNN first. I will try other network later (which will be better results). Till now, i still cannot overfit a set of small train data (i.e. overfit until kaggle loss is -3). Some of my training parameters/feature representation are not correct yet. Generally, my suggested approached is:\n 1. Try to over fit small train set (i.e. loss is very low, preferably close to zero. this is the lowest bound of error. Since kaggler can get to LB -2.3, i expect overfitting to give -3 to -4)\n 2. Try to generalize. See how to make sure results stays the same on validation set (and also LB test set)\n 3. Try to stabilize. Estimate shakeup and prevent difference of LB public and private set.
\n