in the following code in euclidean hash some of the result of the rand.nextGaussian() implementation is negative when set dimension 100 and w is 2. which affect in the hash method.
public EuclideanHash(int dimensions,int w){
Random rand = new Random();
this.w = w;
this.offset = rand.nextInt(w);
randomProjection = new Vector(dimensions);
for(int d=0; d
Hello.
If I am correct, the dot-Method of Vector computes a simple scalar-product.
The problem occured is in my opinion an inaccurate understanding of the function "rand.nextGaussian". This function returns a negative value with a probability of 50%. As it states in the api of Random, the method returns the next pseudorandom, Gaussian ("normally") distributed double value with a mean of 0.0 and standard deviation of 1.0 from this random number generator's sequence. Thus 50% of all doubles returned are negative.
Therefore the randomProjection-Vector contains in average about 50 negative numbers. When you compute the scalar product you may get a negative value which is less than -2. This depends on the values of "vector" and of "randomProjection".
Example with dimension = 6, w = 2:
- vector = (2, 8, 3, 4, 9, 2),
- randomProjection = (0.713, 0.797, 0.382, -0.639, -1.131, 0.132)
- dot-product = -3.515
- hashCode = (-3.515 + offset) / 2 = -1.78 + offset/2
If offset is less than 3.515 in this example, the result will be negative.
I hope you understood the general problem of your code.
Now to my recommendations as what to do: As a negative hashCode may occur whatever hashCode-Function you call, you should be aware of this fact and add an appropriate handling, e.g. by Math.abs(hashCode) in the calling method/class.
I hope you can solve your problem.
Mareike Przysucha, if i add math.abs to hashCode method will change the method itself. how can i change next.gaussin method that result negative value without change the method it self.
Hello Amal. In order to answer this question, I need a little bit more information about the "vector". Are there any limitations about the entries, like: Each entry of vector lies in the interval [x, y]?
Mareike Przysucha. the vector that i want to hash it is 0s and 1s. the result of that hash vector i want to use as index of dimensions of other data structure Bloom Filter.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Algorithms