{
  "id": 11208,
  "title": "R ff package tutorial",
  "url": "/competitions/avazu-ctr-prediction/discussion/11208",
  "author_name": "",
  "post_date": "2014-12-11T15:05:49.520Z",
  "votes": null,
  "comment_count": 19,
  "views": 13916,
  "content": "<p>Hallo everybody,</p>\n<p>I'm looking for R ff package tutorial for my master&nbsp;thesis. I have a little bit experience with R but I never use it for very big data set. I've heard that &nbsp;it's possible to solve big data problem&nbsp;with ff package, therefore I'm looking for some corresponding tutorials. In google I haven't found useful informations for that, maybe anybody of you can help me. thank you very much...</p>\n<p>nan</p>",
  "messages": [
    {
      "id": "60001",
      "postDate": "12/11/2014 15:05:49",
      "content": "<p>Hallo everybody,</p>\n<p>I'm looking for R ff package tutorial for my master&nbsp;thesis. I have a little bit experience with R but I never use it for very big data set. I've heard that &nbsp;it's possible to solve big data problem&nbsp;with ff package, therefore I'm looking for some corresponding tutorials. In google I haven't found useful informations for that, maybe anybody of you can help me. thank you very much...</p>\n<p>nan</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "60008",
      "postDate": "12/11/2014 16:17:46",
      "content": "<p>Offtopic: Seriously, why would people try to use R for this kind of data? It's just impractical. Python is just sooo easier to work with large data. And it's pretty easy to learn, it's pretty much the same with R.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "60017",
      "postDate": "12/11/2014 16:33:47",
      "content": "<p>[quote=Ivan Lobov;60008]</p>\n<p>Offtopic: Seriously, why would people try to use R for this kind of data? It's just impractical. Python is just sooo easier to work with large data. And it's pretty easy to learn, it's pretty much the same with R.</p>\n<p>[/quote]</p>\n<p>would you mind talk about which kind of package to use in Python ? I tried many thing but just not work well.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "60018",
      "postDate": "12/11/2014 16:34:40",
      "content": "<p>Hallo Ivan,</p>\n<p>thanks a lot for your answer, I'm trying now also with Python and your are right, &nbsp;I think Python words better than R for big data problem. I just writing my master paper and I'm afraid that I won't have enough time to learn Python bevor i finish my master paper, therefore I'm looking a compromise with R.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "60026",
      "postDate": "12/11/2014 17:17:28",
      "content": "<p>[quote=hey_world1293;60017]</p>\n<p>[quote=Ivan Lobov;60008]</p>\n<p>Offtopic: Seriously, why would people try to use R for this kind of data? It's just impractical. Python is just sooo easier to work with large data. And it's pretty easy to learn, it's pretty much the same with R.</p>\n<p>[/quote]</p>\n<p>would you mind talk about which kind of package to use in Python ? I tried many thing but just not work well.</p>\n<p>[/quote]</p>\n<p>I'm not sure if you're asking about libraries or algorithms.</p>\n<p>If libraries then you should try <a href=\"http://scikit-learn.org/stable/\">Sklearn</a>&nbsp;if you haven't, especially <a href=\"http://scikit-learn.org/stable/modules/sgd.html\">SGD</a>&nbsp;algos, if the data does not fit into your main memory.</p>\n<p>Another way to go is to try out <a href=\"https://www.kaggle.com/c/avazu-ctr-prediction/forums/t/10927/beat-the-benchmark-with-less-than-1mb-of-memory\">benchmark code</a>&nbsp;and tinker with it. I would also recommend to try a <a href=\"http://www.kaggle.com/c/criteo-display-ad-challenge/forums/t/10322/beat-the-benchmark-with-less-then-200mb-of-memory\">simpler version </a>from another competition, it's much easier to understand at first. But of course it's not as advanced as the ftrl-proximal (first link).</p>\n<p>If you're asking about algorithms then I wouldn't know how to answer it, since you should find the one that gives better results. But the starting point should definitely be a simple Logistic Regressions.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "60046",
      "postDate": "12/11/2014 21:38:44",
      "content": "<p>[quote=Shengnan;60001]</p>\n<p>Hallo everybody,</p>\n<p>I'm looking for R ff package tutorial for my master&nbsp;thesis.</p>\n<p>[/quote]</p>\n<p>If you are looking for a specific tutorial on <em><strong>ff</strong></em> this might be the entering point.</p>\n<p><a href=\"http://www.r-project.org/conferences/useR-2008/slides/Oehlschlaegel+Adler+Nenadic+Zucchini.pdf\">http://www.r-project.org/conferences/useR-2008/slides/Oehlschlaegel+Adler+Nenadic+Zucchini.pdf</a></p>\n<p>I have used and still use the <em><strong>ff</strong></em> package. Nowadays it feels a little dated but it's still a good option for data manipulation and exploration.&nbsp;</p>\n<p>Other alternatives for big data in R are:</p>\n<ul>\n<li><a href=\"http://cran.r-project.org/web/views/HighPerformanceComputing.html\">http://cran.r-project.org/web/views/HighPerformanceComputing.html</a>&nbsp;(full list of big data and high performance computing packages in R)</li>\n<li><span style=\"line-height: 1.4\">RHadoop developed by Revolution Analytics that is a great interface to Hadoop;<a href=\"http://blog.revolutionanalytics.com/2012/03/r-and-hadoop-step-by-step-tutorials.html\">&nbsp;http://blog.revolutionanalytics.com/2012/03/r-and-hadoop-step-by-step-tutorials.html</a></span></li>\n<li><span style=\"line-height: 1.4\"><a href=\"h2o.ai\">h2o.ai</a>&nbsp;although new, has proven to be very powerful specially when used in large clusters. h2o also has a great R interface.</span></li>\n</ul>\n<p><span style=\"line-height: 1.4\">Of course you should explore other alternatives in CRAN, libraries in Python and Julia until you find whatever tool/s fit/s you best.</span></p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "60048",
      "postDate": "12/11/2014 21:44:51",
      "content": "<p>[quote=wacax;60046]</p>\n<p>[/quote]</p>\n<p>Hallo wacax,</p>\n<p>thanks a lot for your informations. I have now 5 months time to finish my master thesis. And I have a little bit R experience but not with ff or Rhadoop and so on. For my master thesis, the biggest data set is 1.48GB. Maybe you can tell me what should I do in this situation, continue my job with R or I have time to lernen and solve the problem with Python?&nbsp;</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "60051",
      "postDate": "12/11/2014 22:33:09",
      "content": "<p>R should be totally fine with 1.5 GB data.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "60116",
      "postDate": "12/12/2014 18:36:29",
      "content": "<p>[quote=Ivan Lobov;60008]</p>\n<p>Offtopic: Seriously, why would people try to use R for this kind of data? It's just impractical. Python is just sooo easier to work with large data. And it's pretty easy to learn, it's pretty much the same with R.</p>\n<p>[/quote]</p>\n<p>My promise to you to beat the competition with R instead of python.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "60118",
      "postDate": "12/12/2014 20:07:06",
      "content": "<p>[quote=Nim J;60116]</p>\n<p>My promise to you to beat the competition with R instead of python.</p>\n<p>[/quote]</p>\n<p>NEVER!</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "60119",
      "postDate": "12/12/2014 20:15:17",
      "content": "<p>Hallo all,</p>\n<p>I'm really appreciate all your helpful&nbsp;suggests. I'm &nbsp;very happy to meet so many kind boys here. Thank you</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "60134",
      "postDate": "12/13/2014 06:31:51",
      "content": "<p>[quote=Ivan Lobov;60026]</p>\n<p>[quote=hey_world1293;60017]</p>\n<p>[quote=Ivan Lobov;60008]</p>\n<p>Offtopic: Seriously, why would people try to use R for this kind of data? It's just impractical. Python is just sooo easier to work with large data. And it's pretty easy to learn, it's pretty much the same with R.</p>\n<p>[/quote]</p>\n<p>would you mind talk about which kind of package to use in Python ? I tried many thing but just not work well.</p>\n<p>[/quote]</p>\n<p>I'm not sure if you're asking about libraries or algorithms.</p>\n<p>If libraries then you should try <a href=\"http://scikit-learn.org/stable/\">Sklearn</a>&nbsp;if you haven't, especially <a href=\"http://scikit-learn.org/stable/modules/sgd.html\">SGD</a>&nbsp;algos, if the data does not fit into your main memory.</p>\n<p>Another way to go is to try out <a href=\"https://www.kaggle.com/c/avazu-ctr-prediction/forums/t/10927/beat-the-benchmark-with-less-than-1mb-of-memory\">benchmark code</a>&nbsp;and tinker with it. I would also recommend to try a <a href=\"http://www.kaggle.com/c/criteo-display-ad-challenge/forums/t/10322/beat-the-benchmark-with-less-then-200mb-of-memory\">simpler version </a>from another competition, it's much easier to understand at first. But of course it's not as advanced as the ftrl-proximal (first link).</p>\n<p>If you're asking about algorithms then I wouldn't know how to answer it, since you should find the one that gives better results. But the starting point should definitely be a simple Logistic Regressions.</p>\n<p>[/quote]</p>\n<p>Thank you!</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "61868",
      "postDate": "12/25/2014 16:42:25",
      "content": "<p>I am also wondering about R... with the memory usage...</p>\n<p>I&nbsp;load all attributes with colClasse == &quot;factor&quot;</p>\n<p>just tried to build a monel using decision tree but failed due to memory usage (&gt; 15gb). with glm the same</p>\n<p>...my purpose is just to get known better with R so. has anybody from an idea which package I could use</p>\n\n<p>my next try is is weight of evidence (klaR)</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "61869",
      "postDate": "12/25/2014 16:47:38",
      "content": "<p>Usually when this happens the best option in R is to just sample the data until you get a small enough set you can work with. Although it's not ideal, you can get a pretty reproducible model with as little as 100k - 1M data points.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "61871",
      "postDate": "12/25/2014 18:14:50",
      "content": "<p>ahja... i had just 100k but it made a matrix with size &gt; 15gb</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "61973",
      "postDate": "12/28/2014 04:03:13",
      "content": "<p>You may have a surprise at the end, # 1 is back :)</p>\n<p>[quote=inversion;60118]</p>\n<p>[quote=Nim J;60116]</p>\n<p>My promise to you to beat the competition with R instead of python.</p>\n<p>[/quote]</p>\n<p>NEVER!</p>\n<p>[/quote]</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "62283",
      "postDate": "01/05/2015 02:59:36",
      "content": "<p>R cant beat python I guess also</p>\n\n<p>just implemented ftrl in R... it is so slow!!!<br><br>well learned R programming much better in such a way :-)</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "62284",
      "postDate": "01/05/2015 03:03:30",
      "content": "<p>Could you share your implementation?</p>\n<p>I also planed to implement it in R using Rcpp but I'm too lazy for now :)</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "62427",
      "postDate": "01/07/2015 23:11:43",
      "content": "<p>here you are</p>\n<p>https://github.com/philipus/Simple-FTRL-in-R.git</p>\n<p>post scriptum:&nbsp;it is for sure not that good implented as tinrtgu did in the python case. in fact I copied the code more or less. well... with the oop stuff in R i am not familiar with :-) but the code works at least and I got some more experiences in R.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "62722",
      "postDate": "01/14/2015 16:11:27",
      "content": "<p>[quote=Filip Floegel;62427]</p>\n<p>here you are</p>\n<p>https://github.com/philipus/Simple-FTRL-in-R.git</p>\n<p>post scriptum:&nbsp;it is for sure not that good implented as tinrtgu did in the python case. in fact I copied the code more or less. well... with the oop stuff in R i am not familiar with :-) but the code works at least and I got some more experiences in R.</p>\n<p>[/quote]</p>\n<p>Hi Filip,</p>\n<p>Thank you for the code. I appreciate having it in R.</p>\n<p>The size of the hash dictionary is&nbsp;D &lt;- 2^20 which is about&nbsp;1 million.</p>\n<p>What are the chances of hash collision?</p>\n<p>I&nbsp;tried playing a bit with alpha and L1, but I don't get a stable error rate even after 1 million rows of training.</p>\n<p>What procedure would&nbsp;you recommend to optimize the different parameters?</p>\n<p>Tim</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 60008,
      "author_name": "ivanlobov",
      "author_url": "",
      "post_date": "12/11/2014 16:17:46",
      "content": "<p>Offtopic: Seriously, why would people try to use R for this kind of data? It's just impractical. Python is just sooo easier to work with large data. And it's pretty easy to learn, it's pretty much the same with R.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 60017,
      "author_name": "gangbeng",
      "author_url": "",
      "post_date": "12/11/2014 16:33:47",
      "content": "<p>[quote=Ivan Lobov;60008]</p>\n<p>Offtopic: Seriously, why would people try to use R for this kind of data? It's just impractical. Python is just sooo easier to work with large data. And it's pretty easy to learn, it's pretty much the same with R.</p>\n<p>[/quote]</p>\n<p>would you mind talk about which kind of package to use in Python ? I tried many thing but just not work well.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 60018,
      "author_name": "emilyl",
      "author_url": "",
      "post_date": "12/11/2014 16:34:40",
      "content": "<p>Hallo Ivan,</p>\n<p>thanks a lot for your answer, I'm trying now also with Python and your are right, &nbsp;I think Python words better than R for big data problem. I just writing my master paper and I'm afraid that I won't have enough time to learn Python bevor i finish my master paper, therefore I'm looking a compromise with R.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 60026,
      "author_name": "ivanlobov",
      "author_url": "",
      "post_date": "12/11/2014 17:17:28",
      "content": "<p>[quote=hey_world1293;60017]</p>\n<p>[quote=Ivan Lobov;60008]</p>\n<p>Offtopic: Seriously, why would people try to use R for this kind of data? It's just impractical. Python is just sooo easier to work with large data. And it's pretty easy to learn, it's pretty much the same with R.</p>\n<p>[/quote]</p>\n<p>would you mind talk about which kind of package to use in Python ? I tried many thing but just not work well.</p>\n<p>[/quote]</p>\n<p>I'm not sure if you're asking about libraries or algorithms.</p>\n<p>If libraries then you should try <a href=\"http://scikit-learn.org/stable/\">Sklearn</a>&nbsp;if you haven't, especially <a href=\"http://scikit-learn.org/stable/modules/sgd.html\">SGD</a>&nbsp;algos, if the data does not fit into your main memory.</p>\n<p>Another way to go is to try out <a href=\"https://www.kaggle.com/c/avazu-ctr-prediction/forums/t/10927/beat-the-benchmark-with-less-than-1mb-of-memory\">benchmark code</a>&nbsp;and tinker with it. I would also recommend to try a <a href=\"http://www.kaggle.com/c/criteo-display-ad-challenge/forums/t/10322/beat-the-benchmark-with-less-then-200mb-of-memory\">simpler version </a>from another competition, it's much easier to understand at first. But of course it's not as advanced as the ftrl-proximal (first link).</p>\n<p>If you're asking about algorithms then I wouldn't know how to answer it, since you should find the one that gives better results. But the starting point should definitely be a simple Logistic Regressions.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 60046,
      "author_name": "wacaxx",
      "author_url": "",
      "post_date": "12/11/2014 21:38:44",
      "content": "<p>[quote=Shengnan;60001]</p>\n<p>Hallo everybody,</p>\n<p>I'm looking for R ff package tutorial for my master&nbsp;thesis.</p>\n<p>[/quote]</p>\n<p>If you are looking for a specific tutorial on <em><strong>ff</strong></em> this might be the entering point.</p>\n<p><a href=\"http://www.r-project.org/conferences/useR-2008/slides/Oehlschlaegel+Adler+Nenadic+Zucchini.pdf\">http://www.r-project.org/conferences/useR-2008/slides/Oehlschlaegel+Adler+Nenadic+Zucchini.pdf</a></p>\n<p>I have used and still use the <em><strong>ff</strong></em> package. Nowadays it feels a little dated but it's still a good option for data manipulation and exploration.&nbsp;</p>\n<p>Other alternatives for big data in R are:</p>\n<ul>\n<li><a href=\"http://cran.r-project.org/web/views/HighPerformanceComputing.html\">http://cran.r-project.org/web/views/HighPerformanceComputing.html</a>&nbsp;(full list of big data and high performance computing packages in R)</li>\n<li><span style=\"line-height: 1.4\">RHadoop developed by Revolution Analytics that is a great interface to Hadoop;<a href=\"http://blog.revolutionanalytics.com/2012/03/r-and-hadoop-step-by-step-tutorials.html\">&nbsp;http://blog.revolutionanalytics.com/2012/03/r-and-hadoop-step-by-step-tutorials.html</a></span></li>\n<li><span style=\"line-height: 1.4\"><a href=\"h2o.ai\">h2o.ai</a>&nbsp;although new, has proven to be very powerful specially when used in large clusters. h2o also has a great R interface.</span></li>\n</ul>\n<p><span style=\"line-height: 1.4\">Of course you should explore other alternatives in CRAN, libraries in Python and Julia until you find whatever tool/s fit/s you best.</span></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 60048,
      "author_name": "emilyl",
      "author_url": "",
      "post_date": "12/11/2014 21:44:51",
      "content": "<p>[quote=wacax;60046]</p>\n<p>[/quote]</p>\n<p>Hallo wacax,</p>\n<p>thanks a lot for your informations. I have now 5 months time to finish my master thesis. And I have a little bit R experience but not with ff or Rhadoop and so on. For my master thesis, the biggest data set is 1.48GB. Maybe you can tell me what should I do in this situation, continue my job with R or I have time to lernen and solve the problem with Python?&nbsp;</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 60051,
      "author_name": "jiweiliu",
      "author_url": "",
      "post_date": "12/11/2014 22:33:09",
      "content": "<p>R should be totally fine with 1.5 GB data.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 60116,
      "author_name": "namesj",
      "author_url": "",
      "post_date": "12/12/2014 18:36:29",
      "content": "<p>[quote=Ivan Lobov;60008]</p>\n<p>Offtopic: Seriously, why would people try to use R for this kind of data? It's just impractical. Python is just sooo easier to work with large data. And it's pretty easy to learn, it's pretty much the same with R.</p>\n<p>[/quote]</p>\n<p>My promise to you to beat the competition with R instead of python.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 60118,
      "author_name": "inversion",
      "author_url": "",
      "post_date": "12/12/2014 20:07:06",
      "content": "<p>[quote=Nim J;60116]</p>\n<p>My promise to you to beat the competition with R instead of python.</p>\n<p>[/quote]</p>\n<p>NEVER!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 60119,
      "author_name": "emilyl",
      "author_url": "",
      "post_date": "12/12/2014 20:15:17",
      "content": "<p>Hallo all,</p>\n<p>I'm really appreciate all your helpful&nbsp;suggests. I'm &nbsp;very happy to meet so many kind boys here. Thank you</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 60134,
      "author_name": "gangbeng",
      "author_url": "",
      "post_date": "12/13/2014 06:31:51",
      "content": "<p>[quote=Ivan Lobov;60026]</p>\n<p>[quote=hey_world1293;60017]</p>\n<p>[quote=Ivan Lobov;60008]</p>\n<p>Offtopic: Seriously, why would people try to use R for this kind of data? It's just impractical. Python is just sooo easier to work with large data. And it's pretty easy to learn, it's pretty much the same with R.</p>\n<p>[/quote]</p>\n<p>would you mind talk about which kind of package to use in Python ? I tried many thing but just not work well.</p>\n<p>[/quote]</p>\n<p>I'm not sure if you're asking about libraries or algorithms.</p>\n<p>If libraries then you should try <a href=\"http://scikit-learn.org/stable/\">Sklearn</a>&nbsp;if you haven't, especially <a href=\"http://scikit-learn.org/stable/modules/sgd.html\">SGD</a>&nbsp;algos, if the data does not fit into your main memory.</p>\n<p>Another way to go is to try out <a href=\"https://www.kaggle.com/c/avazu-ctr-prediction/forums/t/10927/beat-the-benchmark-with-less-than-1mb-of-memory\">benchmark code</a>&nbsp;and tinker with it. I would also recommend to try a <a href=\"http://www.kaggle.com/c/criteo-display-ad-challenge/forums/t/10322/beat-the-benchmark-with-less-then-200mb-of-memory\">simpler version </a>from another competition, it's much easier to understand at first. But of course it's not as advanced as the ftrl-proximal (first link).</p>\n<p>If you're asking about algorithms then I wouldn't know how to answer it, since you should find the one that gives better results. But the starting point should definitely be a simple Logistic Regressions.</p>\n<p>[/quote]</p>\n<p>Thank you!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 61868,
      "author_name": "filipus",
      "author_url": "",
      "post_date": "12/25/2014 16:42:25",
      "content": "<p>I am also wondering about R... with the memory usage...</p>\n<p>I&nbsp;load all attributes with colClasse == &quot;factor&quot;</p>\n<p>just tried to build a monel using decision tree but failed due to memory usage (&gt; 15gb). with glm the same</p>\n<p>...my purpose is just to get known better with R so. has anybody from an idea which package I could use</p>\n\n<p>my next try is is weight of evidence (klaR)</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 61869,
      "author_name": "piinformatics",
      "author_url": "",
      "post_date": "12/25/2014 16:47:38",
      "content": "<p>Usually when this happens the best option in R is to just sample the data until you get a small enough set you can work with. Although it's not ideal, you can get a pretty reproducible model with as little as 100k - 1M data points.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 61871,
      "author_name": "filipus",
      "author_url": "",
      "post_date": "12/25/2014 18:14:50",
      "content": "<p>ahja... i had just 100k but it made a matrix with size &gt; 15gb</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 61973,
      "author_name": "herimanitra",
      "author_url": "",
      "post_date": "12/28/2014 04:03:13",
      "content": "<p>You may have a surprise at the end, # 1 is back :)</p>\n<p>[quote=inversion;60118]</p>\n<p>[quote=Nim J;60116]</p>\n<p>My promise to you to beat the competition with R instead of python.</p>\n<p>[/quote]</p>\n<p>NEVER!</p>\n<p>[/quote]</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 62283,
      "author_name": "filipus",
      "author_url": "",
      "post_date": "01/05/2015 02:59:36",
      "content": "<p>R cant beat python I guess also</p>\n\n<p>just implemented ftrl in R... it is so slow!!!<br><br>well learned R programming much better in such a way :-)</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 62284,
      "author_name": "herimanitra",
      "author_url": "",
      "post_date": "01/05/2015 03:03:30",
      "content": "<p>Could you share your implementation?</p>\n<p>I also planed to implement it in R using Rcpp but I'm too lazy for now :)</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 62427,
      "author_name": "filipus",
      "author_url": "",
      "post_date": "01/07/2015 23:11:43",
      "content": "<p>here you are</p>\n<p>https://github.com/philipus/Simple-FTRL-in-R.git</p>\n<p>post scriptum:&nbsp;it is for sure not that good implented as tinrtgu did in the python case. in fact I copied the code more or less. well... with the oop stuff in R i am not familiar with :-) but the code works at least and I got some more experiences in R.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 62722,
      "author_name": "timhenry",
      "author_url": "",
      "post_date": "01/14/2015 16:11:27",
      "content": "<p>[quote=Filip Floegel;62427]</p>\n<p>here you are</p>\n<p>https://github.com/philipus/Simple-FTRL-in-R.git</p>\n<p>post scriptum:&nbsp;it is for sure not that good implented as tinrtgu did in the python case. in fact I copied the code more or less. well... with the oop stuff in R i am not familiar with :-) but the code works at least and I got some more experiences in R.</p>\n<p>[/quote]</p>\n<p>Hi Filip,</p>\n<p>Thank you for the code. I appreciate having it in R.</p>\n<p>The size of the hash dictionary is&nbsp;D &lt;- 2^20 which is about&nbsp;1 million.</p>\n<p>What are the chances of hash collision?</p>\n<p>I&nbsp;tried playing a bit with alpha and L1, but I don't get a stable error rate even after 1 million rows of training.</p>\n<p>What procedure would&nbsp;you recommend to optimize the different parameters?</p>\n<p>Tim</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "60001": "",
    "60008": "",
    "60017": "",
    "60018": "",
    "60026": "",
    "60046": "",
    "60048": "",
    "60051": "",
    "60116": "",
    "60118": "",
    "60119": "",
    "60134": "",
    "61868": "",
    "61869": "",
    "61871": "",
    "61973": "",
    "62283": "",
    "62284": "",
    "62427": "",
    "62722": ""
  },
  "source": "meta"
}