I first trained an XGB model on train.csv and got a score of .44, when using both train.csv and supplement_train.csv, my score went up to .55! Don't know how to use it? Just load both in and concatenate them! Keep in mind this worked for me on XGB, it may not work for you but it's definitely worth a try.
\ntrain= '/kaggle/input/g-research-crypto-forecasting/train.csv'
\nsupplemental_train= '/kaggle/input/g-research-crypto-forecasting/supplemental_train.csv'
\ntrain1 = pd.read_csv(train)
\ntrain2 = pd.read_csv(supplemental_train)
\ntrain3 = pd.concat([train1,train2],ignore_index=True)
I first trained an XGB model on train.csv and got a score of .44, when using both train.csv and supplement_train.csv, my score went up to .55! Don't know how to use it? Just load both in and concatenate them! Keep in mind this worked for me on XGB, it may not work for you but it's definitely worth a try.
\ntrain= '/kaggle/input/g-research-crypto-forecasting/train.csv'
\nsupplemental_train= '/kaggle/input/g-research-crypto-forecasting/supplemental_train.csv'
\ntrain1 = pd.read_csv(train)
\ntrain2 = pd.read_csv(supplemental_train)
\ntrain3 = pd.concat([train1,train2],ignore_index=True)
1) The data in supplement_train is included in train. You're just duplicating the data when you append it.
\n2) The Targets used for the Public LB are included in the training data, so your Public LB score doesn't mean anything unless you remove all of the rows that are used for scoring the Public LB first from your training set.
Thanks, I knew I should have ran a check to see if the data was duplicated, but I assumed that it was different.
", "rawMarkdown": "Thanks, I knew I should have ran a check to see if the data was duplicated, but I assumed that it was different.", "votes": null }, { "id": "1578523", "postDate": "11/11/2021 06:24:50", "content": "Hi ,
\nI would just like to add that there are significant amount of difference between the both in the PCA plot and the distribution i found it in my eda which is available public
That should be the case since train.csv covers a much longer period of time than supplement_train.csv.
Thanks for that information
", "rawMarkdown": "Thanks for that information", "votes": null }, { "id": "1579623", "postDate": "11/12/2021 04:37:15", "content": "@brandenkmurray can you please tell me if the train_supplemental is the csv file with public lb targets only and not the other samples because I want to remove the public lb samples from training
", "rawMarkdown": "brandenkmurray can you please tell me if the train_supplemental is the csv file with public lb targets only and not the other samples because I want to remove the public lb samples from training", "votes": null } ], "comments": [ { "id": 1578352, "author_name": "brandenkmurray", "author_url": "", "post_date": "11/10/2021 23:34:39", "content": "1) The data in supplement_train is included in train. You're just duplicating the data when you append it.
\n2) The Targets used for the Public LB are included in the training data, so your Public LB score doesn't mean anything unless you remove all of the rows that are used for scoring the Public LB first from your training set.
Thanks, I knew I should have ran a check to see if the data was duplicated, but I assumed that it was different.
", "votes": null, "replies": [] }, { "id": 1579623, "author_name": "swaralipibose", "author_url": "", "post_date": "11/12/2021 04:37:15", "content": "@brandenkmurray can you please tell me if the train_supplemental is the csv file with public lb targets only and not the other samples because I want to remove the public lb samples from training
", "votes": null, "replies": [] } ] }, { "id": 1578523, "author_name": "swaralipibose", "author_url": "", "post_date": "11/11/2021 06:24:50", "content": "Hi ,
\nI would just like to add that there are significant amount of difference between the both in the PCA plot and the distribution i found it in my eda which is available public
That should be the case since train.csv covers a much longer period of time than supplement_train.csv.
Thanks for that information
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1578338": "I first trained an XGB model on train.csv and got a score of .44, when using both train.csv and supplement_train.csv, my score went up to .55! Don't know how to use it? Just load both in and concatenate them! Keep in mind this worked for me on XGB, it may not work for you but it's definitely worth a try.\n\n`train= '/kaggle/input/g-research-crypto-forecasting/train.csv'`\n`supplemental_train= '/kaggle/input/g-research-crypto-forecasting/supplemental_train.csv'`\n`train1 = pd.read_csv(train)`\n`train2 = pd.read_csv(supplemental_train)`\n`train3 = pd.concat([train1,train2],ignore_index=True)`", "1578352": "1) The data in `supplement_train` is included in `train`. You're just duplicating the data when you append it.\n2) The Targets used for the Public LB are included in the training data, so your Public LB score doesn't mean anything unless you remove all of the rows that are used for scoring the Public LB first from your training set.", "1578361": "Thanks, I knew I should have ran a check to see if the data was duplicated, but I assumed that it was different.", "1578523": "Hi ,\nI would just like to add that there are significant amount of difference between the both in the PCA plot and the distribution i found it in my eda which is available public", "1579229": "That should be the case since `train.csv` covers a much longer period of time than `supplement_train.csv`.", "1579620": "Thanks for that information", "1579623": "brandenkmurray can you please tell me if the train_supplemental is the csv file with public lb targets only and not the other samples because I want to remove the public lb samples from training" }, "source": "meta" }