matlab - calculation gives me NaN -

-

I am trying to implement multilateral logistics regression using the gradient lineage, but my cost job is NaN Does the value of the weight tell me what I am doing? Calculates the cost for the gradient lineage, assumes there is a% extra feature in inputX, continuous and weights a class X. Function [cost] = costFunctionMultiNominal (inputX, resultY, weight)% UNTITLED8 Features matrix [rows, columns] = size (inputX); NumOfClasses = Size (weight, 1); Yoga = 0; I = for 1: rows class level academic = 0; For Jammu = 1: numOfClasses if resultY (i) == j denominatorSum = 0; For L = 1: numOfClasses denominatorSum = Denominator + XP ((InputX (I, :) * Weight (L, :) ') - 4444); End ** Class Level End = Class Level End + Log (XP (IPX (I, :) * Weight (Jay, :) '- 4444) / DinominatorSam); ** End End End = Equation + Class Level Easy; Cost of the end = equation / (- lines); Weight update function is: 

  function [weight] = getWeightsUsingGradientDescentMultiNominal (training X, resulting in, iterMax, alpha, weight 0,0, lambda) Weighted weight, weight 0 diagrams are random weight by% return shield dynasty. Detailed description here is found in rows = size (training 1X, 1); cols = size (trainingX, 2) +1; Weight = weight 0; NumOfClasses = Size (weight, 1); Adding one to the input data for constant rules = one (rows, 1); X = [A training X]; % Each column corresponds to one weight, weight is updated according to the column:% together with the CTS function together tempCost = 0; Display (costFunctionMultiNominal (x, resultY, watt)); Plot (1, costFunctionMultiNominal (x, resultY, watt), 'r'); Wait; For N = 1: iterMax% has to do this for all classes, for i = rows in weighths = 1: numOfClasses% first calculates sigma on rows for all x x = sum (1, cols); For I = 1: rows p = -1 * calculation of multiples noamal (x (i, :), weight, j); If the result (i) == jp = 1 + p; End equation = Equation + X (i, :) * p; End weight (j, :) = weight (j, :) - (alpha) * (closing / (- lines) + lambda * weight (jay, :)); End cost = costfunctionmulti nonimal (x, result, weight); Performance (cost); costDiff = tempCost - cost; If I ~ = 0 & amp; Amp; Stomach (costdiff) / cost & lt; = 0.0001 display ('breaking down due to cost!'); break; End tempCost = cost; Plot (i, cost, 'r'); End stop; Finally   
 As far as I understand,  Nayen  is coming in a lot of exponential terms. I tried to reduce the number of losses (-4444), but there was no benefit.  
 If I try NaN and she tells me that this cost is closed on the line in the ceremony (bold in the above code):  
  classLevelSummation = classLevelSummation + Log (XP (InputX (I, :) * Weight (J., '' - 4444) / Negotiator);   
 classLevelSummation becomes NaN, even if I remove the large constant value -4444   
 
 
 in the  log (XP (blah) / two)  This is unnecessary for exponent and then log in, these two actions undo each other, and this The call may be more than the floating point range if you remember that  is equal to log (a / b)   log (a) - Log (B)  Code>.  
  exp  has a simple overflow probability that you do an inf instead of a nan, though. You should check the value of  denominatorSum  because it also comes from an exponential term.

Comments

Post a Comment

Popular posts from this blog

php - PDO bindParam() fatal error -

-
I am trying to learn PDO, now I have written this small code, but it gives me a serious error: < P> Fatal error: Call a member function at a non-object dot () ... $ con = new mysqli ("127.0.0.1", "root", "", "csvdangercheck"); $ Query = $ con- & gt; ("Id,` var1`, `var2`) values ​​(" id,: var1,: var2) ";); $ query- & gt; bypass pattern (': id', $ id); $ query- & gt; BindParam (': var1', $ VAL1); $ Query- & gt; bindParam (': var2', $ val2); $ Query-> Executed (); I Trying to use print_r ($ con-> errorInfo ()) but Fatal error: Undefined method mysqli :: errorInfo ( ) ... can someone tell me what am I missing? Flaffeh said, you have PDO Missing Mysqy, try it: $ con = new PDO ('mysql: host = 127.0.0.1; dbname = csvdangercheck', 'root', '');
Read more

logging - How can I log both the Request.InputStream and Response.OutputStream traffic in my ASP.NET MVC3 Application for specific Actions? -

-
For a specific set of tasks, I need to log in to the incoming request inputstream as well as outgoing response. is. I am using an ActionFilterAttribute for this and overriding the OnActionExecuted and OnResultExecuted methods. So I'm starting with this idea ... Public category ActionLoggerAttribute: ActionFilterArrbit {Public Override Zero On-Offed Execution (ActionActioned Contact Filter Filter) {base.OnActionAxecuted (FilterContext); HttpRequestBase request = filterContext.HttpContext.Request; // TODO: Log the request. InputStream} Public Override Zero OnResultExecuted (ResultExecutedContext FilterContext) {base.OnResultExecuted (filterContext); HTTPRPSPointBase Response = FilterContacts Hpptex response; // TODO: Log on Response.OutputStream}} Ideally I'll hook it up with enterprise library logging because I'm already using it for error logging. Am I reaching the input and output stream at the right time? Can I easily use the Enterprise Library to log i
Read more

java - Why my included JSP file won't get processed correctly? -

-
I am trying to create Java web framework (and learning), and on the basis of 'Code Generator In the process of making the material view of the database in the process of development, I stumbled into a difficulty, which I do not know how to solve it. First of all, I want the following index.jsp : & lt; Body & gt; Use all pages to be created. & Lt;% @ include file = "header.jsp"% & gt; & Lt; Hours / & gt; & Lt;% @ included file = "body.jsp"%> & Lt; Hours / & gt; & Lt;% @ include file = "footer.jsp"% & gt; & Lt; / Body & gt; And, in body.jsp , I should have it like this: & lt; Jsp: include page = "$ {Application_modul}"%> Where application_modul is an attribute defined in its controller: request.setAttribute ("application_modul", "USER_ACCOUNT \\ ' View_user_account.jsp "); This file can be found properly, but the processed JSP does
Read more