First of all congrats to the winners! Great job! Furthermore thanks to the organizers for this cool competition!
\nWhat a shakeup! Actually one of my worst models in respect to public scores would have been the best on the private leaderboard…
\nIf you are interested you find the corresponding notebook (fitting an Extra Tree Classifier) here:
\nhttps://www.kaggle.com/docxian/autism-prediction-evalml/notebook
Obviously, I did not pick that one but a blend of two logistic regressions yielding a private score of 0.97402 (public 0.94320).
", "comments": [ { "id": 1718139, "author_name": "Kishore S", "author_url": "", "post_date": "2022-03-10T14:36:16.063000", "content": "I have also not expected the shakeup given that 80% of the testing data was used for calculating the public LB.
\nMy baseline scikit-learn's Logistic regression model with default parameters & same feature engineering as my best solution would still have been the 8th place solution (Private LB: 0.98051).
", "votes": 3, "replies": [] }, { "id": 1718783, "author_name": "Ankan Hore", "author_url": "", "post_date": "2022-03-11T06:46:43.317000", "content": "Hey Chris! I am learning DS concepts, building classification and regression models. I learned quite a few things from your notebook- using EvalML, removing type constraints before training data and also, working with ML pipelines and how you concluded with the final model.
\nBut I want to understand the intuition behind ---
\n# combine train/test into one data frame
\ndf_train['fold'] = 'train' # for easier extraction later
\ndf_test['fold'] = 'test' # for easier extraction later
\ndf_test['target'] = 0 # set dummy target on test set
Was the fold column used to label/differentiate the train and test data , in order to have clearer extraction afterwards?
", "votes": 1, "replies": [ { "id": 1718943, "author_name": "Chris X", "author_url": "", "post_date": "2022-03-11T10:19:29.913000", "content": "Yes, I use the fold column to easily separate training and test data later. The reason why I combine training and test set at all is that EvalML derives the types of the individual features and I want to make sure that this works consistently across training and test set. Otherwise you might get different types for the same column in train vs test respectively which causes an error in EvalML.
\nThe function used for type derivation is the following:
\ndf = infer_feature_types(df)
\nIt is also in general often useful to run things like feature engineering on the combined train/test data frame in order to assure consistency, e. g. when recoding feature categories etc.
", "votes": 0, "replies": [] }, { "id": 1718983, "author_name": "Ankan Hore", "author_url": "", "post_date": "2022-03-11T11:26:07.723000", "content": "Thanks for helping me out! Yes, I faced a similar issue while solving case study for a company but I couldn't resolve the type mismatch issue in test dataset in that alloted timeframe :(.
\nWell, EvalML helps quite a lot in maintaining the sanctity of the data in the train and test sets.
", "votes": 0, "replies": [] } ] } ], "comment_count": 4, "messages": [ { "id": 1717736, "postDate": "2022-03-10T07:11:22.563Z", "content": "First of all congrats to the winners! Great job! Furthermore thanks to the organizers for this cool competition!
\nWhat a shakeup! Actually one of my worst models in respect to public scores would have been the best on the private leaderboard…
\nIf you are interested you find the corresponding notebook (fitting an Extra Tree Classifier) here:
\nhttps://www.kaggle.com/docxian/autism-prediction-evalml/notebook
Obviously, I did not pick that one but a blend of two logistic regressions yielding a private score of 0.97402 (public 0.94320).
", "rawMarkdown": "First of all congrats to the winners! Great job! Furthermore thanks to the organizers for this cool competition!\n\nWhat a shakeup! Actually one of my worst models in respect to public scores would have been the best on the private leaderboard...\n\nIf you are interested you find the corresponding notebook (fitting an Extra Tree Classifier) here:\nhttps://www.kaggle.com/docxian/autism-prediction-evalml/notebook\n\nObviously, I did not pick that one but a blend of two logistic regressions yielding a private score of 0.97402 (public 0.94320).", "votes": 13 }, { "id": 1718139, "postDate": "2022-03-10T14:36:16.063Z", "content": "I have also not expected the shakeup given that 80% of the testing data was used for calculating the public LB.
\nMy baseline scikit-learn's Logistic regression model with default parameters & same feature engineering as my best solution would still have been the 8th place solution (Private LB: 0.98051).
", "rawMarkdown": "I have also not expected the shakeup given that 80% of the testing data was used for calculating the public LB.\n\nMy baseline scikit-learn's Logistic regression model with default parameters & same feature engineering as my best solution would still have been the 8th place solution (Private LB: 0.98051). ", "votes": 3 }, { "id": 1718783, "postDate": "2022-03-11T06:46:43.317Z", "content": "Hey Chris! I am learning DS concepts, building classification and regression models. I learned quite a few things from your notebook- using EvalML, removing type constraints before training data and also, working with ML pipelines and how you concluded with the final model.
\nBut I want to understand the intuition behind ---
\n# combine train/test into one data frame
\ndf_train['fold'] = 'train' # for easier extraction later
\ndf_test['fold'] = 'test' # for easier extraction later
\ndf_test['target'] = 0 # set dummy target on test set
Was the fold column used to label/differentiate the train and test data , in order to have clearer extraction afterwards?
", "rawMarkdown": "Hey Chris! I am learning DS concepts, building classification and regression models. I learned quite a few things from your notebook- using EvalML, removing type constraints before training data and also, working with ML pipelines and how you concluded with the final model. \n\nBut I want to understand the intuition behind ---\n\n*# combine train/test into one data frame*\n*df_train['fold'] = 'train' # for easier extraction later*\n*df_test['fold'] = 'test' # for easier extraction later*\n*df_test['target'] = 0 # set dummy target on test set*\n\nWas the fold column used to label/differentiate the train and test data , in order to have clearer extraction afterwards?\n", "votes": 1, "replies": [ { "id": 1718943, "postDate": "2022-03-11T10:19:29.913Z", "content": "Yes, I use the fold column to easily separate training and test data later. The reason why I combine training and test set at all is that EvalML derives the types of the individual features and I want to make sure that this works consistently across training and test set. Otherwise you might get different types for the same column in train vs test respectively which causes an error in EvalML.
\nThe function used for type derivation is the following:
\ndf = infer_feature_types(df)
\nIt is also in general often useful to run things like feature engineering on the combined train/test data frame in order to assure consistency, e. g. when recoding feature categories etc.
", "rawMarkdown": "Yes, I use the fold column to easily separate training and test data later. The reason why I combine training and test set at all is that EvalML derives the types of the individual features and I want to make sure that this works consistently across training and test set. Otherwise you might get different types for the same column in train vs test respectively which causes an error in EvalML.\n\nThe function used for type derivation is the following:\n\n*df = infer_feature_types(df)*\n\nIt is also in general often useful to run things like feature engineering on the combined train/test data frame in order to assure consistency, e. g. when recoding feature categories etc." }, { "id": 1718983, "postDate": "2022-03-11T11:26:07.723Z", "content": "Thanks for helping me out! Yes, I faced a similar issue while solving case study for a company but I couldn't resolve the type mismatch issue in test dataset in that alloted timeframe :(.
\nWell, EvalML helps quite a lot in maintaining the sanctity of the data in the train and test sets.
", "rawMarkdown": "Thanks for helping me out! Yes, I faced a similar issue while solving case study for a company but I couldn't resolve the type mismatch issue in test dataset in that alloted timeframe :(.\n\nWell, EvalML helps quite a lot in maintaining the sanctity of the data in the train and test sets. \n" } ] } ], "complete": true, "source": "live", "fetched_at": "2026-10-20T08:18:53Z", "competition_id": 34093, "competition_slug": "autism-prediction", "team_id": 8083315, "solution_url": "/competitions/autism-prediction/writeups/chris-x-public-score-0-93430-private-score-0-98701" }