I have a doubt concerning the exact meaning of external data. Suppose\nthat I want to use the following feature: is_sweet, that feature is True\nwhen the Product Name contains "chocolate" or "vanilla" for example, \nof course the definition per se of the feature depends only of the dataset\nbut the insight that "chocolate" and "vanilla" is sweet isn't intrinsic of the\ndata but external knowledge of the meaning of the words.\nThanks and sorry in advance if the question is too silly!
", "messages": [ { "id": "124751", "postDate": "06/21/2016 23:55:34", "content": "I have a doubt concerning the exact meaning of external data. Suppose\nthat I want to use the following feature: is_sweet, that feature is True\nwhen the Product Name contains "chocolate" or "vanilla" for example, \nof course the definition per se of the feature depends only of the dataset\nbut the insight that "chocolate" and "vanilla" is sweet isn't intrinsic of the\ndata but external knowledge of the meaning of the words.\nThanks and sorry in advance if the question is too silly!
", "rawMarkdown": "I have a doubt concerning the exact meaning of external data. Suppose\r\nthat I want to use the following feature: is_sweet, that feature is True\r\nwhen the Product Name contains \"chocolate\" or \"vanilla\" for example, \r\nof course the definition per se of the feature depends only of the dataset\r\nbut the insight that \"chocolate\" and \"vanilla\" is sweet isn't intrinsic of the\r\ndata but external knowledge of the meaning of the words.\r\nThanks and sorry in advance if the question is too silly!", "votes": null }, { "id": "124770", "postDate": "06/22/2016 06:51:55", "content": "I would imagine the rules on external data apply to stuff that actually gets used by your algorithm, for instance if you used the numbers from some external study on the volume of chocolate sales in different Mexican states, that would be a rule violation. Using your pre-existing knowledge of what you know about chocolate and vanilla solely for the purpose of labeling your feature doesn't seem like a rule violation to me. After all, you're free to call your feature whatever you want. If you're that worried though, instead of is_sweet, call the feature contains_chocolate_or_vanilla and your problem is solved.
", "rawMarkdown": "I would imagine the rules on external data apply to stuff that actually gets used by your algorithm, for instance if you used the numbers from some external study on the volume of chocolate sales in different Mexican states, that would be a rule violation. Using your pre-existing knowledge of what you know about chocolate and vanilla solely for the purpose of labeling your feature doesn't seem like a rule violation to me. After all, you're free to call your feature whatever you want. If you're that worried though, instead of is_sweet, call the feature contains_chocolate_or_vanilla and your problem is solved.", "votes": null }, { "id": "124777", "postDate": "06/22/2016 09:17:48", "content": "As a general rule, it helps to think that the competition host might want to implement parts of the best solutions found in the contest. If those parts include external information that someone else created, there might be a gazillion problems to implement them - due to licensing, ...
\n\nBut if you just parse the information in the data file, and assign a meaning to some tokens of them ('vainilla', 'chocolate', 'megustaelcolordelproductoycreoquepuedeserunafeaturemaravillosa',...), the host won't have any problem in adding that into their implementation.
", "rawMarkdown": "As a general rule, it helps to think that the competition host might want to implement parts of the best solutions found in the contest. If those parts include external information that someone else created, there might be a gazillion problems to implement them - due to licensing, ... \r\n\r\nBut if you just parse the information in the data file, and assign a meaning to some tokens of them ('vainilla', 'chocolate', 'megustaelcolordelproductoycreoquepuedeserunafeaturemaravillosa',...), the host won't have any problem in adding that into their implementation.", "votes": null } ], "comments": [ { "id": 124770, "author_name": "lopezgd", "author_url": "", "post_date": "06/22/2016 06:51:55", "content": "I would imagine the rules on external data apply to stuff that actually gets used by your algorithm, for instance if you used the numbers from some external study on the volume of chocolate sales in different Mexican states, that would be a rule violation. Using your pre-existing knowledge of what you know about chocolate and vanilla solely for the purpose of labeling your feature doesn't seem like a rule violation to me. After all, you're free to call your feature whatever you want. If you're that worried though, instead of is_sweet, call the feature contains_chocolate_or_vanilla and your problem is solved.
", "votes": null, "replies": [] }, { "id": 124777, "author_name": "carrdelling", "author_url": "", "post_date": "06/22/2016 09:17:48", "content": "As a general rule, it helps to think that the competition host might want to implement parts of the best solutions found in the contest. If those parts include external information that someone else created, there might be a gazillion problems to implement them - due to licensing, ...
\n\nBut if you just parse the information in the data file, and assign a meaning to some tokens of them ('vainilla', 'chocolate', 'megustaelcolordelproductoycreoquepuedeserunafeaturemaravillosa',...), the host won't have any problem in adding that into their implementation.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "124751": "I have a doubt concerning the exact meaning of external data. Suppose\r\nthat I want to use the following feature: is_sweet, that feature is True\r\nwhen the Product Name contains \"chocolate\" or \"vanilla\" for example, \r\nof course the definition per se of the feature depends only of the dataset\r\nbut the insight that \"chocolate\" and \"vanilla\" is sweet isn't intrinsic of the\r\ndata but external knowledge of the meaning of the words.\r\nThanks and sorry in advance if the question is too silly!", "124770": "I would imagine the rules on external data apply to stuff that actually gets used by your algorithm, for instance if you used the numbers from some external study on the volume of chocolate sales in different Mexican states, that would be a rule violation. Using your pre-existing knowledge of what you know about chocolate and vanilla solely for the purpose of labeling your feature doesn't seem like a rule violation to me. After all, you're free to call your feature whatever you want. If you're that worried though, instead of is_sweet, call the feature contains_chocolate_or_vanilla and your problem is solved.", "124777": "As a general rule, it helps to think that the competition host might want to implement parts of the best solutions found in the contest. If those parts include external information that someone else created, there might be a gazillion problems to implement them - due to licensing, ... \r\n\r\nBut if you just parse the information in the data file, and assign a meaning to some tokens of them ('vainilla', 'chocolate', 'megustaelcolordelproductoycreoquepuedeserunafeaturemaravillosa',...), the host won't have any problem in adding that into their implementation." }, "source": "meta" }