tag:blogger.com,1999:blog-5743983044224833668.post6057604512773407916..comments2010-05-15T10:03:20.050-07:00Jeff Fosterhttp://www.blogger.com/profile/08195722595923882332noreply@blogger.comBlogger5125tag:blogger.com,1999:blog-5743983044224833668.post-33071299291049172872010-05-15T10:03:20.042-07:002010-05-15T10:03:20.042-07:00Thanks Paweł - that looks much cleaner! <br /><br />Looking back now, I&#39;m of the opinion anything that uses recur is almost certainly something you can do with a standard function. I guess recursion is a code smell in a functional language since most likely you can express it in terms of a map / fold operation.Jeff Fosterhttp://www.blogger.com/profile/08195722595923882332noreply@blogger.comtag:blogger.com,1999:blog-5743983044224833668.post-86060209511217234632010-05-15T08:52:32.430-07:002010-05-15T08:52:32.430-07:00&quot;train&quot; function does look a lot like reduce so I thought it&#39;d be nice if you made test data &#39;reducable&#39; and refactor &quot;train&quot; into function that may be used by &quot;reduce&quot;.<br /><br />You can find source code at:<br />http://gist.github.com/402257Paweł Badeńskihttp://www.blogger.com/profile/02544441831377477201noreply@blogger.comtag:blogger.com,1999:blog-5743983044224833668.post-38499402568056384622010-01-29T00:57:24.082-08:002010-01-29T00:57:24.082-08:00What is your opinion, the structure of typiclal NN can be best described from the point of view objects or functions?<br /><br />And what structure would be better to use for NN structure, weights, inputs.<br />list<br />set<br />vectors<br />maps<br /><br />or maybe use Java&#39;s structure?Infospacerhttp://www.blogger.com/profile/03490720712678508108noreply@blogger.comtag:blogger.com,1999:blog-5743983044224833668.post-88947666120838969212010-01-28T23:25:04.678-08:002010-01-28T23:25:04.678-08:00I&#39;m definitely not an expert in FP or ANN so take this with a pinch of salt!<br /><br />I think your concerns about many matrix copies are (probably) unfounded. The Clojure data structures are designed to mutate efficiently; they share the structure that stays the same only the new data changes. In addition, when you need to mutate you can do so in a safe way using <i>transient</i>.<br /><br />Even if it is a little slower then you can probably win in the end with multi-processor support!Jeff Fosterhttp://www.blogger.com/profile/08195722595923882332noreply@blogger.comtag:blogger.com,1999:blog-5743983044224833668.post-46934592785427568902010-01-28T18:12:03.623-08:002010-01-28T18:12:03.623-08:00Nice post. I&#39;m novice in FP. Now I&#39;m trying to play with Clojure. It&#39;s seems for me this paradigma is much closer to my way of thinking than OO. I was looking for any implementations of ANN on functional language. I&#39;m planning to implement Restricted Boltzman Machine for humor and jokes extraction from text. <br /><br />I&#39;d like to know your opinion regarding using functional programming vs OO programming for ANN construction. Which one is better suited for this kind of tasks. As soon as I know with FP it is much simpler to programm for multiprocessor hardware. But how about memory consumption. With BP algorithm we should create many matrix copy when adjusting wheights.<br /><br />So anybody using FP languages for ANN?Infospacerhttp://www.blogger.com/profile/03490720712678508108noreply@blogger.com