A detailed interpretation of the Cvtrimweights function

Cvtrimweights source in the OpenCV cvboost.cpp file, the specific content and some comments are as follows

/* *cvtrimweights * Role: The weight of less than a certain threshold culling, because the smaller weight of the sample on the training results have little impact, after culling, so that training samples can shorten the training time */cv_boost_implcvmat* cvtrimweights ( cvmat* weights,//Training Sample weight matrix  cvmat* idx,//Training sample index sequence matrix float factor)//culling    The fraction of the sample remaining after a small weight is calculated based on this parameter to calculate the threshold value {cvmat* ptr = 0) that needs to be excluded from the smaller sample.    Cv_funcname ("Cvtrimweights");    __begin__;    int i, index, NUM;    float sum_weights;    uchar* Wdata;    size_t Wstep;    int wnum;    float threshold;    int count;    float* sorted_weights;    Cv_assert (Cv_mat_type (weights->type) = = CV_32FC1);    ptr = idx;    Sorted_weights = NULL;        if (Factor > 0.0F && Factor < 1.0F) {size_t data_size;        Cv_mat2vec (*weights, Wdata, Wstep, wnum); num = (idx = = NULL)?        Wnum:max (Idx->rows, idx->cols);        data_size = num * sizeof (*sorted_weights);        Sorted_weights = (float*) cvalloc (data_size);        memset (sorted_weights, 0, data_size);        Sum_weights = 0.0F; for (i = 0; i < num; i++) {index = Icvgetidxat (idx, I);//Remove the sample sequence from the first position in the matrix idx sorted_weights[i] = * ((float*) (  Wdata + index * wstep));//Remove the weight of the IDX I position sum_weights + = sorted_weights[i];//sample Weights and} icvsort_32f (      sorted_weights, num, 0);//Sort Samples sum_weights *= (1.0f-factor);        Based on the total weight and factor of the training samples, the weights and i =-1 of the small weight training samples which need to be rejected are calculated. do {sum_weights-= Sorted_weights[++i];}        The sorted sample is subtracted from the sample with the smallest small weight until the condition ends while the while (Sum_weights > 0.0F && i < (num-1));             threshold = Sorted_weights[i]; /* After jumping out of the Do-while loop, assign the weight of the sample with the smallest weight to threshold without being stripped of the weight. External performance: If the weight of the sample is less than this threshold, the direct culling * * while (i > 0 && sorted_weights[i-1] = = threshold) i--;//The small weight of the sample, and then To the original culling of the portion of the threshold equal to the sample picked back, very humane ha if (i > 0 | |        (idx = NULL && cv_mat_type (idx->type)! = CV_32FC1))            {Cv_call (ptr = Cvcreatemat (1, Num-i, CV_32FC1));            Count = 0; for (i = 0; i < num;                i++) {index = Icvgetidxat (idx, i); if (* (* (float*) (Wdata + index * wstep)) >= threshold)//weight operation only for threshold >threshold {cv_m                    At_elem (*ptr, float, 0, count) = (float) index;                count++;        }} assert (count = = Ptr->cols);    } cvfree (&sorted_weights);    } __end__;     return ptr; Only samples with weights greater than threshold are returned}

Copyright NOTICE: This article for Bo Master original article, without Bo Master permission not reproduced.

A detailed interpretation of the Cvtrimweights function