Hi, I am new to the kaggle forums and I am still learning and is attempting on competitions that comes up(I am a noob). Would greatly appreciate if someone can guide me for my following doubt.
\n\nMy Doubt: In the dataset, once we flatten the json into columns, the following columns are also produced along with the others.
\n\n1)trafficSource.source \n2)device.browser \n3)geoNetwork.country \n4)geoNetwork.networkDomain \n5)geoNetwork.region
\n\nI have seen in the discussion boards that categorical features are being label encoded for this dataset. However, why are we not attempting for a one hot encoding(OHE) approach. Because I have read somewhere that OHE is better than Label encoding.
\n\nAdditionally, I am not able to try out OHE on the dataset as the factor level is obviously huge for each column mentioned above. Is that the reason why label encoding is done for the dataset? Is there any other way to deal these columns other than label encoding ?
\n\nAny help is much appreciated. Apologising once again for any mistakes I have written in the topic
", "messages": [ { "id": 400506, "postDate": "2018-10-08T12:35:34.683Z", "content": "OHE makes large columns in dataset. because categorical varibles have many unique values ( high cardinality ).\nso YouHan lee makes kernel frequency encoding, means encoding.
\n\nhttps://www.kaggle.com/youhanlee/which-encoding-is-good-for-time-validation-1-4417
", "rawMarkdown": "OHE makes large columns in dataset. because categorical varibles have many unique values ( high cardinality ).\nso YouHan lee makes kernel frequency encoding, means encoding. \n\nhttps://www.kaggle.com/youhanlee/which-encoding-is-good-for-time-validation-1-4417" }, { "id": 400475, "postDate": "2018-10-08T11:40:34.937Z", "content": "Hi, I am new to the kaggle forums and I am still learning and is attempting on competitions that comes up(I am a noob). Would greatly appreciate if someone can guide me for my following doubt.
\n\nMy Doubt: In the dataset, once we flatten the json into columns, the following columns are also produced along with the others.
\n\n1)trafficSource.source \n2)device.browser \n3)geoNetwork.country \n4)geoNetwork.networkDomain \n5)geoNetwork.region
\n\nI have seen in the discussion boards that categorical features are being label encoded for this dataset. However, why are we not attempting for a one hot encoding(OHE) approach. Because I have read somewhere that OHE is better than Label encoding.
\n\nAdditionally, I am not able to try out OHE on the dataset as the factor level is obviously huge for each column mentioned above. Is that the reason why label encoding is done for the dataset? Is there any other way to deal these columns other than label encoding ?
\n\nAny help is much appreciated. Apologising once again for any mistakes I have written in the topic
", "rawMarkdown": "Hi, I am new to the kaggle forums and I am still learning and is attempting on competitions that comes up(I am a noob). Would greatly appreciate if someone can guide me for my following doubt.\n\nMy Doubt: In the dataset, once we flatten the json into columns, the following columns are also produced along with the others.\n\n1)trafficSource.source \n2)device.browser \n3)geoNetwork.country \n4)geoNetwork.networkDomain \n5)geoNetwork.region\n\nI have seen in the discussion boards that categorical features are being label encoded for this dataset. However, why are we not attempting for a one hot encoding(OHE) approach. Because I have read somewhere that OHE is better than Label encoding. \n\nAdditionally, I am not able to try out OHE on the dataset as the factor level is obviously huge for each column mentioned above. Is that the reason why label encoding is done for the dataset? Is there any other way to deal these columns other than label encoding ?\n\nAny help is much appreciated. Apologising once again for any mistakes I have written in the topic " }, { "id": 400508, "postDate": "2018-10-08T12:40:51.453Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 400506, "author_name": "choco", "author_url": "", "post_date": "2018-10-08T12:35:34.683000", "content": "OHE makes large columns in dataset. because categorical varibles have many unique values ( high cardinality ).\nso YouHan lee makes kernel frequency encoding, means encoding.
\n\nhttps://www.kaggle.com/youhanlee/which-encoding-is-good-for-time-validation-1-4417
", "votes": 0, "replies": [] }, { "id": 400508, "author_name": "", "author_url": "", "post_date": "2018-10-08T12:40:51.453000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "400506": "OHE makes large columns in dataset. because categorical varibles have many unique values ( high cardinality ).\nso YouHan lee makes kernel frequency encoding, means encoding. \n\nhttps://www.kaggle.com/youhanlee/which-encoding-is-good-for-time-validation-1-4417", "400475": "Hi, I am new to the kaggle forums and I am still learning and is attempting on competitions that comes up(I am a noob). Would greatly appreciate if someone can guide me for my following doubt.\n\nMy Doubt: In the dataset, once we flatten the json into columns, the following columns are also produced along with the others.\n\n1)trafficSource.source \n2)device.browser \n3)geoNetwork.country \n4)geoNetwork.networkDomain \n5)geoNetwork.region\n\nI have seen in the discussion boards that categorical features are being label encoded for this dataset. However, why are we not attempting for a one hot encoding(OHE) approach. Because I have read somewhere that OHE is better than Label encoding. \n\nAdditionally, I am not able to try out OHE on the dataset as the factor level is obviously huge for each column mentioned above. Is that the reason why label encoding is done for the dataset? Is there any other way to deal these columns other than label encoding ?\n\nAny help is much appreciated. Apologising once again for any mistakes I have written in the topic ", "400508": "" } }