Method and device for determining statistical model parameter based on expectation-maximization algorithm

Method and device for determining statistical model parameter based on expectation-maximization algorithm

  • CN 104,809,098 A
  • Filed: 01/27/2014
  • Published: 07/29/2015
  • Est. Priority Date: 01/27/2014
  • Status: Active Application
First Claim
Patent Images

1. determine a method for statistical model parameter, for the parameter based on N number of data point determination statistical model, wherein N be more than or equal to 2 integer, it is characterized in that, comprising:

  • Receive comprise D attribute of N number of data point and N number of data point data set to be organized into input matrix, wherein D be more than or equal to 1 integer;

    According to described input matrix, set K cluster centre, the initial value of described parameter and posterior probability matrix μ

    n ×

    K


    n,k) initial value wherein, posterior probability μ

    n,krepresent the posterior probability of the n-th data point on a kth cluster centre, wherein K be more than or equal to 2 integer, 1≤

    n≤

    N, 1≤

    k≤

    K, and according to and the calculation of initial value of described parameter Based on calculate the described parameter of the t time circulation, and calculate the residual error of described N number of data point at a described K cluster centre wherein t>

    =1;

    From described N number of data point, select M data point in the residual error of a described K cluster centre based on described N number of data point, and select L cluster centre from a described K cluster centre, wherein 1≤

    M≤

    N, 1≤

    L≤

    K;

    The posterior probability of a described M data point on a described L cluster centre is calculated according to the described parameter that the t time cycle calculations obtains According to calculated described posterior probability upgrade the posterior probability matrix of described N number of data point on a described K cluster centre and based on the described posterior probability calculated upgrade the value of the described parameter of the t+1 time circulation;

    Judge whether the described parameter circulated for the t+1 time restrains, when described parameter is for convergence, stop circulating and exporting described parameter.

View all claims
    ×
    ×

    Thank you for your feedback

    ×
    ×