[{"data":1,"prerenderedAt":315},["ShallowReactive",2],{"project-25829757":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":29,"sentiment":30,"stance_mix":31,"scored_comments":35,"aspects":36,"strip":43,"post_text":45,"what":46,"said":88,"maker_in_thread":180,"shape":183,"grid":188,"attempts":277,"same_week":278,"maker_shows":39,"maker_hidden":307,"reader_version":308,"card":309},"25829757","tf-quant-finance","A high-performance TensorFlow library for quantitative finance","title","Show HN: A high-performance TensorFlow library for quantitative finance","https:\u002F\u002Fgithub.com\u002Fgoogle\u002Ftf-quant-finance","github.com","partingshots",1611027788,2021,153,38,319,2533,13,"notable","finance_crypto","quantitative finance library","library",[24,25,26,27,28],"tensorflow","quantitative finance","option pricing","gpu","monte carlo","curious",-0.048,{"positive":32,"neutral":33,"mixed":34,"critical":32},4,12,0,20,{"idea":37,"performance":40,"execution":42},{"positive":38,"critical":39},3,2,{"positive":41,"critical":41},1,{"positive":34,"critical":41},[38,44,44,44,44,38,34,38,34,34,34,38,34,34,34,34,34,34,34,34,34,38,41,38,41,34,41,34,34,34,38,38,34,41,38,38,38,38],5,null,{"problem":47,"audience":63,"pricing":45,"motivation":45,"time_to_build":45,"stack":71,"openness":85,"stage":86,"team_size":87},{"value":48,"source":49,"evidence":50},"Running quantitative finance computations such as derivatives pricing and risk on TensorFlow for speed","inferred",[51,55,59],{"id":52,"by":53,"text":54},"25830484","jabberwcky","> but does anyone actually do that? Backtesting a regular single symbol model on CPU is often not cheap, especially at higher frequencies. Now multiply that by 50 correlated…",{"id":56,"by":57,"text":58},"25830640","adament","If you are pricing and risk-managing complex derivatives books then Pandas is not the right tool for the job. Pandas does not include derivatives pricing models, daycount…",{"id":60,"by":61,"text":62},"25830903","mynegation","> I get that calculating a zillion option prices on a GPU is faster, but does anyone actually do that? Yes. Counterparty credit risk calculations are done as Monte Carlo…",{"value":64,"source":49,"evidence":65},"Quants and derivatives pricing and risk teams",[66,70],{"id":67,"by":68,"text":69},"25830145","cliftonk","Agreed this is less convenient than pandas for many tasks, however large market makers (citadel securities, Susquehanna, Jane st, Hudson River trading) employ abstractions like…",{"id":56,"by":57,"text":58},[72,78],{"value":73,"source":74,"evidence":75},"TensorFlow","maker",[76],{"id":77},"post",{"value":79,"source":49,"evidence":80},"Python",[81],{"id":82,"by":83,"text":84},"25832540","secondcoming","We tried to use Tensorflow's native C++ library to run some predictive model in real time on CPU-only. The performance sucked, even though the model was not particularly complex.…","open_source","unknown","company",{"praise":89,"objections":101,"requests":130,"questions":137,"alternatives":164},[90,97],{"value":91,"source":92,"evidence":93},"Large market makers use similar abstractions, so GPU-accelerated quant tooling has real value","commenter",[94,95,96],{"id":67,"by":68,"text":69},{"id":52,"by":53,"text":54},{"id":60,"by":61,"text":62},{"value":98,"source":92,"evidence":99},"Deep learning libraries could match C++ performance with less code",[100],{"id":56,"by":57,"text":58},[102,109,119],{"value":103,"source":92,"evidence":104},"Pandas is simpler and more mature for most finance data work",[105],{"id":106,"by":107,"text":108},"25830055","dvt","Is there any actual value in doing quantitative finance via TF as opposed to just Pandas? I get that calculating a zillion option prices on a GPU is faster, but does anyone…",{"value":110,"source":92,"evidence":111},"Floating point precision may be unsuitable for finance",[112,116],{"id":113,"by":114,"text":115},"25831348","proverbialbunny","Is this using floating point numbers under the hood? There is a lot you can do with floating point, but for some of what I've worked on you really do want fixed precision types.",{"id":117,"by":114,"text":118},"25835543",">Always ask yourself - do you need to know this number to better than 1 in 10^14? Yes, because over time thousands of multiplies and divides means you will get more than a penny…",{"value":120,"source":92,"evidence":121},"Project appears to be by people without market experience",[122,126],{"id":123,"by":124,"text":125},"25832426","pvitz","Right now, it looks like a fun project of students who got their hands on Andersen\u002FPiterbarg, Brigo\u002FMercurio and the QMC part of Glasserman. I wonder what their real goal is with…",{"id":127,"by":128,"text":129},"25834290","andyxor","This was not done by people who make their living in markets",[131],{"value":132,"source":92,"evidence":133},"Explain the issues with existing work",[134],{"id":135,"by":11,"text":136},"25841787","Can you expand? Would love to hear what issues you have so far with the existing work.",[138,153],{"value":139,"source":92,"evidence":140},"Why Google is building this and whether it targets finance firms or spins off a fund",[141,145,149],{"id":142,"by":143,"text":144},"25834150","pantsforbirds","I was wondering if it was developed to help promote TF more to finance firms.",{"id":146,"by":147,"text":148},"25841737","person_of_color","Are they trying to spin off a hedge fund? wtf.",{"id":150,"by":151,"text":152},"25831704","neolog","Was this developed for Google internal finance projects?",{"value":154,"source":92,"evidence":155},"Whether TensorFlow or PyTorch is better for this use",[156,160],{"id":157,"by":158,"text":159},"25830441","reallymental","I understand that this is google's library, and they're promoting TensorFlow's usage, and subsequently the usage of TPU's. But this can just as easily be done in Pytorch right?…",{"id":161,"by":162,"text":163},"25831415","ipsum2","TFX\u002FTensorFlow Serving doesn't have a PyTorch equivalent. TF has better mobile mobile support. Jax also works on TPUs and is closer (with other libraries on top like Haiku) to…",[165,169,173],{"value":166,"source":92,"evidence":167},"Pandas",[168],{"id":106,"by":107,"text":108},{"value":170,"source":92,"evidence":171},"PyTorch",[172],{"id":157,"by":158,"text":159},{"value":174,"source":92,"evidence":175},"C++ pricing frameworks",[176],{"id":177,"by":178,"text":179},"25831534","jsinai","More likely production systems will be written in C++ than Python.",{"comments":41,"answered":34,"key":181},[182],{"id":135,"text":136},{"commenters":184,"max_depth":185,"maker_replies":41,"first_comment_mins":186,"lifespan_hours":187},29,9,56,23,[189,191,193,195,196,197,198,202,206,208,211,212,214,218,221,222,225,227,229,231,233,235,238,240,244,247,250,252,254,256,259,262,265,268,269,270,271,275],[106,34,38,190,107,34],"project\u002Fneutral\u002Fidea",[67,34,44,192,68,34],"project\u002Fpositive\u002Fidea",[52,34,44,194,53,34],"project\u002Fpositive\u002Fperformance",[56,34,44,192,57,34],[60,34,44,192,61,34],[177,34,38,190,178,34],[199,34,34,200,201,34],"25837902","commenter\u002Fneutral\u002Fnone","chewbaxxa",[203,41,38,204,205,34],"25830799","project\u002Fneutral\u002Fnone","akhilcacharya",[157,39,34,207,158,34],"other\u002Fneutral\u002Fnone",[209,39,34,207,210,34],"25830536","atty",[161,39,34,207,162,34],[113,38,38,213,114,34],"project\u002Fneutral\u002Fexecution",[215,38,34,216,217,34],"25831969","commenter\u002Fneutral\u002Fexecution","pimbrah",[219,38,34,216,220,34],"25835270","bfrink",[117,38,34,216,114,34],[223,38,34,216,224,34],"25836564","hansvm",[226,38,34,216,114,34],"25836633",[228,38,34,216,220,34],"25837699",[230,38,34,216,224,34],"25838064",[232,38,34,216,220,34],"25838453",[234,38,34,216,114,34],"25838898",[236,32,38,213,237,34],"25831003","sandGorgon",[82,44,41,239,83,34],"project\u002Fcritical\u002Fperformance",[241,242,38,190,243,34],"25836751",6,"vmception",[123,245,41,246,124,34],7,"project\u002Fcritical\u002Fidea",[248,245,34,200,249,34],"25834138","olalonde",[127,245,41,251,128,34],"project\u002Fcritical\u002Fexecution",[253,245,34,200,124,34],"25838080",[135,245,34,255,11,41],"maker\u002Fneutral\u002Fnone",[257,245,34,207,258,34],"25840132","JustFinishedBSG",[260,261,38,255,147,34],"25831937",8,[263,261,38,255,264,34],"25831962","tazjin",[266,261,34,207,267,34],"25832157","monkeydust",[146,261,41,246,147,34],[150,185,38,204,151,34],[142,185,38,204,143,34],[272,273,38,190,274,34],"25831594",10,"leoplct",[276,273,38,190,57,34],"25831701",[],[279,284,289,294,298,302],{"id":280,"name":281,"points":282,"category":283},"25848278","Beeper",962,"consumer_social",{"id":285,"name":286,"points":287,"category":288},"25846347","Vim",667,"learning_science",{"id":290,"name":291,"points":292,"category":293},"25821873","Turquoise Health",447,"data",{"id":295,"name":296,"points":297,"category":288},"25836395","Tejotron",409,{"id":299,"name":300,"points":301,"category":283},"25848793","Sundayy",215,{"id":303,"name":304,"points":305,"category":306},"25786640","Vmctl\u002FVmcli",198,"dev_tools",false,"v2",{"id":310,"name":5,"line":6,"ai":307,"cat":20,"year":13,"points":14,"comments":15,"rec":29,"s":311,"top":18,"kw":312,"maker":11,"domain":10,"tier":19,"attempts":34,"strip":313,"url":9,"mix":314,"praise":91,"objection":103,"sub":21},25829757,-0.05,[24,25,26,27,28],"35555303000300000000031310100033013333",[32,33,34,32],1793991022404]