1. GoogleNews-vectors-negative300 - https://code.google.com/archive/p/word2vec/\n2. glove.840B.300d - https://nlp.stanford.edu/projects/glove/\n3. paragram_300_sl999 - https://cogcomp.org/page/resource_view/106\n4. wiki-news-300d-1M - https://fasttext.cc/docs/en/english-vectors.html\nWhy everyone is trying the glove, paragram and wiki not GOOGLE. i have seen majority guys in competition working with 2nd,3rd and 4th but not the 1st one,\nIs there any particular reason for that.
", "messages": [ { "id": "426843", "postDate": "11/24/2018 01:45:47", "content": "1. GoogleNews-vectors-negative300 - https://code.google.com/archive/p/word2vec/\n2. glove.840B.300d - https://nlp.stanford.edu/projects/glove/\n3. paragram_300_sl999 - https://cogcomp.org/page/resource_view/106\n4. wiki-news-300d-1M - https://fasttext.cc/docs/en/english-vectors.html\nWhy everyone is trying the glove, paragram and wiki not GOOGLE. i have seen majority guys in competition working with 2nd,3rd and 4th but not the 1st one,\nIs there any particular reason for that.
", "rawMarkdown": "1. GoogleNews-vectors-negative300 - https://code.google.com/archive/p/word2vec/\n 2. glove.840B.300d - https://nlp.stanford.edu/projects/glove/\n 3. paragram_300_sl999 - https://cogcomp.org/page/resource_view/106\n 4. wiki-news-300d-1M - https://fasttext.cc/docs/en/english-vectors.html\n\nWhy everyone is trying the glove, paragram and wiki not GOOGLE. i have seen majority guys in competition working with 2nd,3rd and 4th but not the 1st one,\nIs there any particular reason for that.", "votes": null }, { "id": "426867", "postDate": "11/24/2018 03:21:12", "content": "I compared the validation score from these two kernels:
\n\nGloVe tends to work best.
", "rawMarkdown": "I compared the validation score from these two kernels:\n\n* https://www.kaggle.com/strideradu/word2vec-and-gensim-go-go-go\n* https://www.kaggle.com/sudalairajkumar/a-look-at-different-embeddings\n\nGloVe tends to work best.", "votes": null }, { "id": "428095", "postDate": "11/26/2018 18:40:47", "content": "thanks finally got the answer
", "rawMarkdown": "thanks finally got the answer", "votes": null }, { "id": "428096", "postDate": "11/26/2018 18:41:56", "content": "I have one more question do pre-processing the data before applying the model will be helpful ??
", "rawMarkdown": "I have one more question do pre-processing the data before applying the model will be helpful ??", "votes": null }, { "id": "428099", "postDate": "11/26/2018 18:48:38", "content": "Not in my case.
", "rawMarkdown": "Not in my case.", "votes": null }, { "id": "428284", "postDate": "11/27/2018 02:46:23", "content": "thankyou @Shujian Liu
", "rawMarkdown": "thankyou @Shujian Liu", "votes": null }, { "id": "435344", "postDate": "12/07/2018 22:02:47", "content": "Personally, I can't even test my code on my own machine if I want to use Google's creation. If there were a .csv version, it'd be fine... but it takes WAY too much memory (I only have 8 gig on my laptop) and I gave up on finding out how much time just to load. For a 'kernels only' competition, just trying to load their darn file would eat up most of the runtime. The .CSV's can be read one line at a time, grabbing words that actually appear in the train and test vocabulary... whereas the \"gensim\" form requires loading the whole darn thing... like I said, I don't know how long that would take because I gave up. So personally I figure the Google creation is a complete non-starter.
", "rawMarkdown": "Personally, I can't even test my code on my own machine if I want to use Google's creation. If there were a .csv version, it'd be fine... but it takes WAY too much memory (I only have 8 gig on my laptop) and I gave up on finding out how much time just to load. For a 'kernels only' competition, just trying to load their darn file would eat up most of the runtime. The .CSV's can be read one line at a time, grabbing words that actually appear in the train and test vocabulary... whereas the \"gensim\" form requires loading the whole darn thing... like I said, I don't know how long that would take because I gave up. So personally I figure the Google creation is a complete non-starter.", "votes": null }, { "id": "435749", "postDate": "12/08/2018 17:21:24", "content": "Got you @Steven \nthankyou
", "rawMarkdown": "Got you @Steven \nthankyou", "votes": null } ], "comments": [ { "id": 426867, "author_name": "shujian", "author_url": "", "post_date": "11/24/2018 03:21:12", "content": "I compared the validation score from these two kernels:
\n\nGloVe tends to work best.
", "votes": null, "replies": [ { "id": 428095, "author_name": "kalyankkr", "author_url": "", "post_date": "11/26/2018 18:40:47", "content": "thanks finally got the answer
", "votes": null, "replies": [] }, { "id": 428096, "author_name": "kalyankkr", "author_url": "", "post_date": "11/26/2018 18:41:56", "content": "I have one more question do pre-processing the data before applying the model will be helpful ??
", "votes": null, "replies": [] }, { "id": 428099, "author_name": "shujian", "author_url": "", "post_date": "11/26/2018 18:48:38", "content": "Not in my case.
", "votes": null, "replies": [] }, { "id": 428284, "author_name": "kalyankkr", "author_url": "", "post_date": "11/27/2018 02:46:23", "content": "thankyou @Shujian Liu
", "votes": null, "replies": [] } ] }, { "id": 435344, "author_name": "stevenaleach", "author_url": "", "post_date": "12/07/2018 22:02:47", "content": "Personally, I can't even test my code on my own machine if I want to use Google's creation. If there were a .csv version, it'd be fine... but it takes WAY too much memory (I only have 8 gig on my laptop) and I gave up on finding out how much time just to load. For a 'kernels only' competition, just trying to load their darn file would eat up most of the runtime. The .CSV's can be read one line at a time, grabbing words that actually appear in the train and test vocabulary... whereas the \"gensim\" form requires loading the whole darn thing... like I said, I don't know how long that would take because I gave up. So personally I figure the Google creation is a complete non-starter.
", "votes": null, "replies": [ { "id": 435749, "author_name": "kalyankkr", "author_url": "", "post_date": "12/08/2018 17:21:24", "content": "Got you @Steven \nthankyou
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "426843": "1. GoogleNews-vectors-negative300 - https://code.google.com/archive/p/word2vec/\n 2. glove.840B.300d - https://nlp.stanford.edu/projects/glove/\n 3. paragram_300_sl999 - https://cogcomp.org/page/resource_view/106\n 4. wiki-news-300d-1M - https://fasttext.cc/docs/en/english-vectors.html\n\nWhy everyone is trying the glove, paragram and wiki not GOOGLE. i have seen majority guys in competition working with 2nd,3rd and 4th but not the 1st one,\nIs there any particular reason for that.", "426867": "I compared the validation score from these two kernels:\n\n* https://www.kaggle.com/strideradu/word2vec-and-gensim-go-go-go\n* https://www.kaggle.com/sudalairajkumar/a-look-at-different-embeddings\n\nGloVe tends to work best.", "428095": "thanks finally got the answer", "428096": "I have one more question do pre-processing the data before applying the model will be helpful ??", "428099": "Not in my case.", "428284": "thankyou @Shujian Liu", "435344": "Personally, I can't even test my code on my own machine if I want to use Google's creation. If there were a .csv version, it'd be fine... but it takes WAY too much memory (I only have 8 gig on my laptop) and I gave up on finding out how much time just to load. For a 'kernels only' competition, just trying to load their darn file would eat up most of the runtime. The .CSV's can be read one line at a time, grabbing words that actually appear in the train and test vocabulary... whereas the \"gensim\" form requires loading the whole darn thing... like I said, I don't know how long that would take because I gave up. So personally I figure the Google creation is a complete non-starter.", "435749": "Got you @Steven \nthankyou" }, "source": "meta" }