There are single LSTM/GRU models that are beating the baseline. Is it just finding the right hyper parameters and tuning the model? Or is there scope for feature engineering here? \nAnd also how to choose the right models to blend, is it just by hit and trial?\nPlease help Kaggle Gods xD
", "messages": [ { "id": "457960", "postDate": "01/18/2019 12:32:56", "content": "There are single LSTM/GRU models that are beating the baseline. Is it just finding the right hyper parameters and tuning the model? Or is there scope for feature engineering here? \nAnd also how to choose the right models to blend, is it just by hit and trial?\nPlease help Kaggle Gods xD
", "rawMarkdown": "There are single LSTM/GRU models that are beating the baseline. Is it just finding the right hyper parameters and tuning the model? Or is there scope for feature engineering here? \nAnd also how to choose the right models to blend, is it just by hit and trial?\nPlease help Kaggle Gods xD", "votes": null }, { "id": "457994", "postDate": "01/18/2019 13:58:21", "content": "@NiranjanRao I think its huge amount of experience with algorithms. I know a Kaggle Grandmaster very well and as far as I have observed his work I have found that they have a lot of experience and practice of many algorithms. They know algorithms upside down and inside out. They have very interesting approaches to decrease the test error. Another grandmaster I closely follow is famous to present elegant simple solutions to very sophisticated problems. A lot of people were using Deep learning and NN and what not but he won the competition with 1st place and his solution had very simplistic application of linear regression with some advance regularization. Feature engineering is a step in multi step process to get the best model.
\n\nThere is a kernel on feature engineering for this competition see https://www.kaggle.com/shaz13/feature-engineering-for-nlp-classification
\n\nThere is discussion topic for blending models https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/71778
\n\nIf you need to understand the feature extraction have a look at policy of asking questions by quora in this topic https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/77691#456365
\n\nI hope it helps. Good Luck :)
", "rawMarkdown": "NiranjanRao I think its huge amount of experience with algorithms. I know a Kaggle Grandmaster very well and as far as I have observed his work I have found that they have a lot of experience and practice of many algorithms. They know algorithms upside down and inside out. They have very interesting approaches to decrease the test error. Another grandmaster I closely follow is famous to present elegant simple solutions to very sophisticated problems. A lot of people were using Deep learning and NN and what not but he won the competition with 1st place and his solution had very simplistic application of linear regression with some advance regularization. Feature engineering is a step in multi step process to get the best model. \n\nThere is a kernel on feature engineering for this competition see https://www.kaggle.com/shaz13/feature-engineering-for-nlp-classification\n\nThere is discussion topic for blending models https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/71778\n\nIf you need to understand the feature extraction have a look at policy of asking questions by quora in this topic https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/77691#456365\n\nI hope it helps. Good Luck :)", "votes": null }, { "id": "462046", "postDate": "01/27/2019 14:46:25", "content": "@Niranjan Rao , \nPlease can you this kernal :\nhttps://www.kaggle.com/ashishpatel26/ensemble-learning-for-beginner-to-advance
\n\nPlease let me know if have any query .
", "rawMarkdown": "Niranjan Rao , \nPlease can you this kernal :\nhttps://www.kaggle.com/ashishpatel26/ensemble-learning-for-beginner-to-advance\n\nPlease let me know if have any query .", "votes": null } ], "comments": [ { "id": 457994, "author_name": "cyberia", "author_url": "", "post_date": "01/18/2019 13:58:21", "content": "@NiranjanRao I think its huge amount of experience with algorithms. I know a Kaggle Grandmaster very well and as far as I have observed his work I have found that they have a lot of experience and practice of many algorithms. They know algorithms upside down and inside out. They have very interesting approaches to decrease the test error. Another grandmaster I closely follow is famous to present elegant simple solutions to very sophisticated problems. A lot of people were using Deep learning and NN and what not but he won the competition with 1st place and his solution had very simplistic application of linear regression with some advance regularization. Feature engineering is a step in multi step process to get the best model.
\n\nThere is a kernel on feature engineering for this competition see https://www.kaggle.com/shaz13/feature-engineering-for-nlp-classification
\n\nThere is discussion topic for blending models https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/71778
\n\nIf you need to understand the feature extraction have a look at policy of asking questions by quora in this topic https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/77691#456365
\n\nI hope it helps. Good Luck :)
", "votes": null, "replies": [] }, { "id": 462046, "author_name": "sunilcube", "author_url": "", "post_date": "01/27/2019 14:46:25", "content": "@Niranjan Rao , \nPlease can you this kernal :\nhttps://www.kaggle.com/ashishpatel26/ensemble-learning-for-beginner-to-advance
\n\nPlease let me know if have any query .
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "457960": "There are single LSTM/GRU models that are beating the baseline. Is it just finding the right hyper parameters and tuning the model? Or is there scope for feature engineering here? \nAnd also how to choose the right models to blend, is it just by hit and trial?\nPlease help Kaggle Gods xD", "457994": "NiranjanRao I think its huge amount of experience with algorithms. I know a Kaggle Grandmaster very well and as far as I have observed his work I have found that they have a lot of experience and practice of many algorithms. They know algorithms upside down and inside out. They have very interesting approaches to decrease the test error. Another grandmaster I closely follow is famous to present elegant simple solutions to very sophisticated problems. A lot of people were using Deep learning and NN and what not but he won the competition with 1st place and his solution had very simplistic application of linear regression with some advance regularization. Feature engineering is a step in multi step process to get the best model. \n\nThere is a kernel on feature engineering for this competition see https://www.kaggle.com/shaz13/feature-engineering-for-nlp-classification\n\nThere is discussion topic for blending models https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/71778\n\nIf you need to understand the feature extraction have a look at policy of asking questions by quora in this topic https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/77691#456365\n\nI hope it helps. Good Luck :)", "462046": "Niranjan Rao , \nPlease can you this kernal :\nhttps://www.kaggle.com/ashishpatel26/ensemble-learning-for-beginner-to-advance\n\nPlease let me know if have any query ." }, "source": "meta" }