svm - LibSVM one class classification nu parameter is not a fraction of outliers? -


Please correct me if I am wrong, but a class SVM principle states that nu parameter upper bound (UB) training Outliers in Datasets and LBs of the number of SVs say that I am using RBF Gaussian Kernel, so considering the nu parameter, no matter what the value of gamma I choose, the result of the model Be able to generate Is there a UB outlayer dataset in parameter NU training? However, it is not what I have seen that I have seen using a few simple examples with LbsM in Little: 

  [heart_scale_label, heart_scale_inst] = libsvmread ('.. / heart_scale') ; Ind_good = (heart_scale_label == 1); Heart_scale_label = heart_scale_label (ind_good); Heart_scale_inst = heart_scale_inst (ind_good); Train_data = heart_scale_inst; Train_label = heart_scale_label; Gamma = 0.01; New = 0.01; Model = svmtrain (train_label, train_data, ['-s 2 -t 2 -n' num2str (nu) '-g' num2str (gamma) '-h 0']); [Predict_label_Tr, accuracy_tr, dec_values_Tr] = svmpredict (train_label, train_data, model); Accuracy by using Gamma = 0.01 TAGR = <0.01> I get the accuracy of trained data using 97.50 / gamma = 100. I get the accuracy of training data as 42.50 only when the only part of the outliers in training dataset, Is selected?   
 
 
 Actually I discovered the same problem. The performance of SVM generally depends on the contact of γ and nu. If you are fixing a parameter while trying to tune each other, then there is not even one in the learning curve.  
 I have to draw training accuracy, accuracy test (5 times in depth of heart), and three images on their difference. From γ to  10 ^ (- 4)  to  10 ^ (1) , and nu from  10 ^ (- 3) to  to < Code> 10 ^ (- 1) :  
 enter image details here  
 To observe the small parameters more clearly, I applied logarithm to γ and New axis, see the following figure:  
  Enter image details here   
 In fact, looks more clearly than with the given 120 data.  
   
  Enter the image details here   
 Not quite clear or not perfect at all! Generalization on parameters seems a bit smoother as the dependence of the error, possibly due to the optimization algorithm used in Libsevim instead of 'true' solution ...

Comments

Post a Comment

Popular posts from this blog

c - Mpirun hangs when mpi send and recieve is put in a loop -

I was trying to use the program to use mifrewan on 4 node clusters. Distributing node 0 data in nodes 1, 2 and 3. In the program, from '90 'to variable values ​​of' dior 'to be calculated. Therefore the node is distributing data and collecting results in an omitted fashion (for different values ​​of var 'dir') when do {*******} Whereas (dir Loop is given, Mippun is hanging, and no output is found. But when I comment on two {*******} (dir The loop output is obtained for the initial value of variable dir dir = -90 ), and that output is correct. The problem occurs when the loop is given. Can anyone help me in solving this problem. #include "mpi.h" int main (int argc, char * argv []) Float dir = -90; Int rank, nanoprakash; MPI_Status status; MPI_Init (& amp; argc, & argv); MPI_Comm_rank (MPI_COMM_WORLD, & amp; Rank); MPI_Comm_size (MPI_COMM_WORLD, & amp; numprocs); If (rank == 0) {/ / data to begin / / (dest = 1; dest ...
Read more

python - Apply coupon to a customer's subscription based on non-stripe related actions on the site -

I can not find anything in the source about using coupons after subscribing to a user's plan. Is it possible to use a coupon to credit a user's membership after initial subscription? If not, what is the best option? Specifically, I want to give a user 100% coupon for that month, when they take 5 [site specific actions]. This is possible, but may not be the best idea. Coupons are currently applied to the stripe, then apply again at the customer level and each client can have a single coupon. So if you are already offering discounts to the coupon, then the customer who is currently in the present will also give up. It also means that if you are using a recently-added multiple subscription capacity, then all of the user's subscription will be ready for the month. Coupons can be set to "Once", only once (as the only name) can be used by any customer, so if this credit can earn more than once, You will have to maintain a 100% discount coupon continuousl...
Read more

java - Unable to get JDBC connection in Spring application to MySQL -

मैंने स्प्रिंग में डीबी कॉन्फ़िगरेशन का पालन किया है: & lt; bean id = "dataSource" Class = "org.apache.commons.dbcp.BasicDataSource" नष्ट-विधि = "बंद" & gt; & Lt; प्रॉपर्टी नाम = "url" मान = "jdbc: dburl" / & gt; & Lt; संपत्ति नाम = "driverClassName" मान = "com.mysql.jdbc.Driver" / & gt; & Lt; संपत्ति नाम = "उपयोगकर्ता नाम" मान = "उपयोगकर्ता नाम" / & gt; & Lt; संपत्ति नाम = "पासवर्ड" मान = "पासवॉर्फ" / & gt; & Lt; प्रॉपर्टी नाम = "हटाए गए हटाए गए" मान = "सही" / & gt; & Lt; गुण नाम = "प्रारंभिक आकार" मान = "3" / & gt; & Lt; संपत्ति नाम = "अधिकतमअक्टिव" मान = "5" / & gt; & Lt; / सेम ​​& gt; डीबी से गुण मूल्यों को लोड करने के लिए अतिरिक्त कॉन्फ़िगरेशन। & lt; bean id = "config1" class = "org.apache.commons.configuration....
Read more