(New page: The LGB algorithm is another clustering technique. It has a slightly different approach compared to [K-means]. In this algorithm the user inputs the number of clusters he wants to split hi...) |
|||
| Line 1: | Line 1: | ||
| − | The LGB algorithm is another clustering technique. It has a slightly different approach compared to [K-means]. In this algorithm the user inputs the number of clusters he wants to split his data set into. It has to be a power of 2. | + | The LGB algorithm is another clustering technique. It has a slightly different approach compared to [K-means]. In this algorithm the user inputs the number of clusters he wants to split his data set into. It has to be a power of 2. |
| − | The algo works as follows: | + | The algo works as follows: |
1) Find the sample mean of the data. Let that me some mu. | 1) Find the sample mean of the data. Let that me some mu. | ||
| − | |||
| − | |||
| − | |||
| − | This algorithm will ensure that we have the desired number of clusters. | + | 2) To this mean add and subtract a small value epsilon. Thus we will have 2 new means mu +epsilon and mu-epsilon.For every data point in the set find which mean it belongs to. Now we have thus clustered the data into 2 parts. |
| + | |||
| + | 3) For each of this part evaluate the new means. Now again split each of the means into 2 more. | ||
| + | |||
| + | 4) Continue till we get the desired number of clusters. | ||
| + | |||
| + | This algorithm will ensure that we have the desired number of clusters. | ||
The algorithm is named after Linde,Buzo and Gray. | The algorithm is named after Linde,Buzo and Gray. | ||
Revision as of 14:17, 16 March 2008
The LGB algorithm is another clustering technique. It has a slightly different approach compared to [K-means]. In this algorithm the user inputs the number of clusters he wants to split his data set into. It has to be a power of 2.
The algo works as follows:
1) Find the sample mean of the data. Let that me some mu.
2) To this mean add and subtract a small value epsilon. Thus we will have 2 new means mu +epsilon and mu-epsilon.For every data point in the set find which mean it belongs to. Now we have thus clustered the data into 2 parts.
3) For each of this part evaluate the new means. Now again split each of the means into 2 more.
4) Continue till we get the desired number of clusters.
This algorithm will ensure that we have the desired number of clusters.
The algorithm is named after Linde,Buzo and Gray.
