[{"data":1,"prerenderedAt":477},["ShallowReactive",2],{"project-16626374":3},{"id":4,"name":5,"line":6,"line_source":7,"title":8,"url":9,"domain":10,"maker":11,"posted_at":12,"year":13,"points":14,"comments":15,"rank_in_year":16,"shown_in_year":17,"top":18,"tier":19,"category":20,"subcategory":21,"format":22,"keywords":23,"reception":28,"sentiment":29,"stance_mix":30,"scored_comments":35,"aspects":36,"strip":46,"post_text":48,"what":49,"said":70,"maker_in_thread":139,"shape":148,"grid":153,"attempts":438,"same_week":439,"maker_shows":42,"maker_hidden":469,"reader_version":470,"card":471},"16626374","Word2Bits","Quantized Word Vectors","title","Show HN: Word2Bits – Quantized Word Vectors","https:\u002F\u002Fgithub.com\u002Fagnusmaximus\u002FWord2Bits","github.com","maxlam",1521526924,2018,221,110,150,2155,7,"notable","languages","word embeddings library","library",[24,25,26,27],"word vectors","word2vec","quantization","embeddings","curious",0.247,{"positive":31,"neutral":32,"mixed":33,"critical":34},13,25,0,3,41,{"idea":37,"performance":40,"privacy":41,"docs":43,"execution":44,"design":45},{"positive":38,"critical":39},5,2,{"positive":34,"critical":33},{"positive":33,"critical":42},1,{"positive":42,"critical":33},{"positive":42,"critical":33},{"positive":42,"critical":33},[38,34,42,34,33,34,38,38,34,33,33,33,33,33,33,33,33,38,38,34,34,34,38,33,33,33,33,38,33,33,33,33,34,34,47,47,33,38,34,34],4,null,{"problem":50,"audience":48,"pricing":48,"motivation":48,"time_to_build":48,"stack":57,"openness":68,"stage":69,"team_size":69},{"value":51,"source":52,"evidence":53},"Word vectors take large storage; quantizing them to few bits shrinks them","maker",[54],{"id":55,"by":11,"text":56},"16626679","The idea is that you can kind of capture the \"meaning\" of a word with a sequence of numbers (a vector) -- and then you use these vectors for machine learning tasks to do cool…",[58,64],{"value":59,"source":52,"evidence":60},"TensorFlow",[61],{"id":62,"by":11,"text":63},"16627047","Training these quantized word vectors has to be done in full precision (so no memory gains during training the word vectors). But when you save them to disk every value is either…",{"value":65,"source":52,"evidence":66},"PyTorch",[67],{"id":62,"by":11,"text":63},"open_source","unknown",{"praise":71,"objections":91,"requests":105,"questions":117,"alternatives":129},[72,84],{"value":73,"source":74,"evidence":75},"Useful for memory-constrained and client-side applications","commenter",[76,80],{"id":77,"by":78,"text":79},"16626973","dougb5","This is great and echoes a recent fascination for me. One application of compact word embeddings is that if they're small enough you can ship a whole model to the user in a web…",{"id":81,"by":82,"text":83},"16627081","Jack000","can this be done with fasttext as well? Word2bits is definitely great for memory-constrained applications but for server use memory isn't as much a constraint (there's a direct…",{"value":85,"source":74,"evidence":86},"Interesting results beating Word2Vec on SQuAD",[87],{"id":88,"by":89,"text":90},"16626584","rococode","Very cool that this beats Word2Vec on SQuAD! I wonder if the current state of the art models are using the standard GloVe word vecs and might see improvement from this. Tbh I…",[92,99],{"value":93,"source":74,"evidence":94},"Vectors may carry gender and other biases from training data",[95],{"id":96,"by":97,"text":98},"16626666","rspeer","It's interesting and slightly uncomfortable that the illustration for similar words uses the word \"man\" as an example, given the gender biases that result from learning word…",{"value":100,"source":52,"evidence":101},"Dimensions appear to be noise and hard to interpret",[102],{"id":103,"by":11,"text":104},"16626729","Definitely tried to figure out if the dimensions mean anything -- as far as I can tell they don't really mean much :(",[106,113],{"value":107,"source":74,"evidence":108},"Test debiasing methods on quantized vectors",[109,110],{"id":96,"by":97,"text":98},{"id":111,"by":97,"text":112},"16626739","There may be a way to build it into the loss function so that it happens before the quantization, right? (and holy crap, look how fast the HN conservatives are getting to my…",{"value":114,"source":74,"evidence":115},"Add fastText-style out-of-vocabulary handling",[116],{"id":81,"by":82,"text":83},[118,125],{"value":119,"source":74,"evidence":120},"How to read the graphs",[121],{"id":122,"by":123,"text":124},"16626477","wodenokoto","Can someone explain how to read the graph?",{"value":126,"source":74,"evidence":127},"How many quantization values were tried",[128],{"id":88,"by":89,"text":90},[130,135],{"value":131,"source":74,"evidence":132},"word2vec and GloVe",[133,134],{"id":88,"by":89,"text":90},{"id":96,"by":97,"text":98},{"value":136,"source":74,"evidence":137},"fastText",[138],{"id":81,"by":82,"text":83},{"comments":140,"answered":141,"key":142},17,12,[143,144,145],{"id":55,"text":56},{"id":62,"text":63},{"id":146,"text":147},"16628037","Absolutely, please do!",{"commenters":149,"max_depth":150,"maker_replies":140,"first_comment_mins":151,"lifespan_hours":152},48,10,37,34,[154,156,160,164,168,171,174,176,180,184,187,190,194,196,200,203,205,208,212,214,217,218,221,223,226,230,233,235,237,239,241,243,246,248,251,253,255,257,259,261,263,266,269,272,275,278,280,283,286,289,291,293,297,299,301,304,306,309,313,315,317,318,321,324,326,329,331,335,337,339,342,344,346,348,350,352,354,357,360,362,364,366,368,370,372,374,376,379,381,384,385,389,391,393,398,399,402,405,407,410,412,414,417,419,423,425,429,431,434,436],[77,33,38,155,78,33],"project\u002Fpositive\u002Fidea",[157,33,34,158,159,33],"16627115","project\u002Fneutral\u002Fidea","visarga",[161,33,34,162,163,33],"16627612","project\u002Fneutral\u002Fperformance","narrator",[165,33,34,166,167,33],"16630364","project\u002Fneutral\u002Fprivacy","rpedela",[169,33,34,166,170,33],"16630958","darawk",[172,33,42,173,167,33],"16631359","project\u002Fcritical\u002Fprivacy",[175,33,34,166,78,33],"16631474",[177,42,42,178,179,33],"16627922","project\u002Fcritical\u002Fidea","tmalsburg2",[181,42,33,182,183,33],"16628794","commenter\u002Fneutral\u002Fidea","popinman322",[185,42,34,158,186,33],"16629048","morelandjs",[188,42,33,182,189,33],"16629447","loxias",[191,42,33,192,193,33],"16629859","commenter\u002Fneutral\u002Fnone","jdonaldson",[195,42,33,192,189,33],"16631144",[197,39,34,198,199,33],"16626604","project\u002Fneutral\u002Fdocs","Sukotto",[201,39,34,198,202,33],"16626670","knolan",[55,39,33,204,11,42],"project\u002Fpositive\u002Fdocs",[206,39,38,207,199,33],"16626856","maker\u002Fpositive\u002Fdocs",[209,39,38,210,211,33],"16626887","project\u002Fpositive\u002Fperformance","opportune",[62,39,33,213,11,42],"maker\u002Fneutral\u002Fperformance",[215,39,38,216,211,33],"16627121","maker\u002Fpositive\u002Fexecution",[146,39,33,216,11,42],[219,39,34,158,220,33],"16627156","jacobush",[222,39,34,158,211,33],"16627293",[224,39,34,158,225,33],"16627746","avinium",[227,34,33,228,229,33],"16626654","other\u002Fneutral\u002Fidea","creichenbach",[231,34,33,228,232,33],"16626693","isoprophlex",[234,34,33,228,229,33],"16626930",[236,34,33,228,232,33],"16629333",[238,34,33,228,229,33],"16637577",[240,34,33,228,123,33],"16626728",[242,34,33,228,229,33],"16626948",[244,34,33,182,245,33],"16627087","igravious",[247,34,33,182,229,33],"16627954",[249,34,33,182,250,33],"16629396","rjurney",[252,34,33,228,211,33],"16626780",[254,34,33,228,229,33],"16626963",[256,34,33,228,211,33],"16626985",[258,34,33,228,229,33],"16628002",[260,34,33,228,211,33],"16628556",[262,34,33,228,229,33],"16637525",[264,34,33,265,189,33],"16629845","other\u002Fneutral\u002Fnone",[267,34,33,228,268,33],"16632515","sushirain",[270,34,33,228,271,33],"16627187","nl",[273,34,33,274,229,33],"16628049","other\u002Fneutral\u002Fdocs",[276,34,33,274,277,33],"16628582","physicsyogi",[279,34,33,274,271,33],"16631856",[281,34,33,265,282,33],"16629461","patelajay285",[284,47,38,155,285,33],"16627200","Radim",[287,38,38,155,288,33],"16626655","a_bonobo",[88,290,38,210,89,33],6,[292,290,33,213,11,42],"16626597",[294,290,34,295,296,33],"16628903","maker\u002Fneutral\u002Fidea","yorwba",[298,290,33,295,11,42],"16632548",[300,290,34,295,296,33],"16634370",[302,290,33,303,11,42],"16639436","maker\u002Fpositive\u002Fidea",[305,18,34,158,189,33],"16628962",[307,18,33,192,308,33],"16634751","_mhr_",[310,311,38,155,312,33],"16627690",8,"mark_l_watson",[96,314,42,178,97,33],9,[316,314,33,295,11,42],"16626720",[111,314,34,295,97,33],[319,314,33,320,11,42],"16626763","maker\u002Fneutral\u002Fnone",[322,314,38,323,97,33],"16626789","maker\u002Fpositive\u002Fnone",[325,314,33,323,11,42],"16626834",[327,314,33,265,328,33],"16627040","hnuser1234",[330,314,33,192,328,33],"16627054",[332,314,33,333,334,33],"16627141","commenter\u002Fcritical\u002Fnone","strawcomb",[336,314,33,192,328,33],"16627229",[338,314,33,192,97,33],"16630593",[340,314,33,192,341,33],"16627211","fvdessen",[343,314,33,192,97,33],"16629199",[345,314,33,192,341,33],"16629934",[347,314,33,192,97,33],"16630743",[349,314,33,192,341,33],"16631278",[351,314,38,303,97,33],"16632227",[353,314,33,192,341,33],"16633016",[355,314,33,265,356,33],"16627410","BjoernKW",[358,314,33,192,359,33],"16627691","IAmEveryone",[361,314,33,192,341,33],"16627961",[363,314,33,192,359,33],"16628273",[365,314,33,182,296,33],"16629381",[367,314,33,192,97,33],"16629517",[369,314,33,192,296,33],"16629779",[371,314,33,192,341,33],"16630101",[373,314,33,192,97,33],"16630386",[375,314,33,192,97,33],"16630206",[377,314,33,265,378,33],"16627832","empath75",[380,314,33,192,97,33],"16629405",[382,150,34,320,383,33],"16626684","andreyk",[103,150,33,295,11,42],[386,150,34,387,388,33],"16629933","maker\u002Fneutral\u002Fdesign","Maybestring",[390,150,33,295,11,42],"16632560",[392,150,33,228,189,33],"16630952",[394,395,34,396,397,33],"16633165",11,"project\u002Fneutral\u002Fnone","malcook",[81,141,38,210,82,33],[400,141,34,396,401,33],"16627822","manojlds",[403,31,38,323,404,33],"16626544","Dawny33",[122,406,34,198,123,33],14,[408,406,34,198,409,33],"16626494","iampims",[411,406,33,198,11,42],"16626518",[413,406,33,198,11,42],"16626528",[415,406,34,416,123,33],"16626695","project\u002Fneutral\u002Fdesign",[418,406,33,416,11,42],"16626750",[420,406,38,421,422,33],"16626896","project\u002Fpositive\u002Fdesign","s0cket",[424,406,33,396,11,42],"16626976",[426,427,34,396,428,33],"16626547",15,"newman8r",[430,427,33,396,11,42],"16626557",[432,427,34,387,433,33],"16626565","NKCSS",[435,427,33,396,11,42],"16626573",[437,427,34,396,428,33],"16626596",[],[440,445,450,455,460,464],{"id":441,"name":442,"points":443,"category":444},"16619580","URL Canary",273,"security_privacy",{"id":446,"name":447,"points":448,"category":449},"16591962","Where is Sci-Hub",247,"other",{"id":451,"name":452,"points":453,"category":454},"16606408","Makesite",146,"dev_tools",{"id":456,"name":457,"points":458,"category":459},"16636935","Sheet2Site",145,"web_saas",{"id":461,"name":462,"points":463,"category":454},"16594515","Fossa-cli",104,{"id":465,"name":466,"points":467,"category":468},"16591878","Talent vs. Luck",86,"learning_science",false,"v2",{"id":472,"name":5,"line":6,"ai":469,"cat":20,"year":13,"points":14,"comments":15,"rec":28,"s":473,"top":18,"kw":474,"maker":11,"domain":10,"tier":19,"attempts":33,"strip":475,"url":9,"mix":476,"praise":73,"objection":93,"sub":21},16626374,0.25,[24,25,26,27],"5313035530000000055333500005000033440533",[31,32,33,34],1794036564413]