• Status: Solved
  • Priority: Medium
  • Security: Public
  • Views: 563
  • Last Modified:

pseudo-code formula

there is a formula in the following article:Article link
page 503: Levine and Nazif

where simply it does the following:

ListOfClasses = [c1,c2,c3,...,cn], and where ci = [value1,value2,...,valuen]
so it does the intra-region uniformity, the closer to 1 the better the result.

Well what is really disturbing is that i cannot fix it while working with vectors instead of values:

ListOfClasses = [c1,c2,c3,...,cn], and where ci = [v1,v2,...,vn] where v = VECTOR
and where vi = [value1,value2,value3,...valuen]

Can you pls give me the pseudo-code for that algorithm with vectors!!

I am getting opposite results.
  • 2
  • 2
1 Solution
What does the vector represent? What is the current algorithm you are using? Since the uniformity should just be the inverse of the variance, it shouldn't give you the opposite unless you are actually calculating the variance. Then you would just need to flip it (realAnswer = 1/oldAnswer or something).
dadadudeAuthor Commented:
What does the vector represent?
u can think of each vector as f(x,y) the value of each element in a class.
instead of f(x,y) = x (value), f(x,y) = [e1,e2,...,en]
What is the current algorithm you are using?
def M_intraClasses(self,classList):
		somme1 = 0.0
		#for each class in the classlist
		for i in range(0,len(classList)):
			#take a cluster
			cluster = classList[i].getListOfGraphems()
			s1 = []
			#now for each element in the cluster ( s1 is f(x,y) which contains values, so s1 will contain vectors instead of values)
			for sommet in range(0,len(cluster)):
			#take fmax and fmin
			fmax = amax(s1)
			fmin = amin(s1)
			#compute the standard deviation divided by fmax - fmin
			somme1 += std(s1)**2/(((fmax - fmin)**2 + self.epsilon)/2.0)
		#since the intra-region uniformity should be low i maximize it but taking 1 out of it.
		uniforme = 1.0-(somme1/float(len(classList)))
		return uniforme

Open in new window

So why don't you just do a nested for loop and add all the elements of all the vectors into s1? Is that what you tried already? Why didn't it work? It should have.

Or should you be computing multiple values? Should you be getting a different uniformity for each place in the vector or are all the values in the vector related to the same thing?
dadadudeAuthor Commented:
yes it did work now when i have only 1 class the uniformity criteria is high, but when many classes the uniformity is low.

the more i have the classes the lower the uniformity criteria. So i think yes it's working perfectly.

Thank you for your help as usual!

Featured Post

Get expert help—faster!

Need expert help—fast? Use the Help Bell for personalized assistance getting answers to your important questions.

  • 2
  • 2
Tackle projects and never again get stuck behind a technical roadblock.
Join Now