void gk_find_frequent_itemsets(int ntrans, int *tranptr, int *tranind,
int minfreq, int maxfreq, int minlen, int maxlen,
void (*process_itemset)(void *stateptr, int nitems, int *itemids,
int ntrans, int *transids),
void *stateptr)
{
ssize_t i;
gk_csr_t *mat, *pmat;
isparams_t params;
int *pattern;
/* Create the matrix */
mat = gk_csr_Create();
mat->nrows = ntrans;
mat->ncols = tranind[gk_iargmax(tranptr[ntrans], tranind)]+1;
mat->rowptr = gk_zmalloc(ntrans+1, "gk_find_frequent_itemsets: mat.rowptr");
for (i=0; i<ntrans+1; i++)
mat->rowptr[i] = tranptr[i];
mat->rowind = gk_icopy(tranptr[ntrans], tranind, gk_imalloc(tranptr[ntrans], "gk_find_frequent_itemsets: mat.rowind"));
mat->colids = gk_iincset(mat->ncols, 0, gk_imalloc(mat->ncols, "gk_find_frequent_itemsets: mat.colids"));
/* Setup the parameters */
params.minfreq = minfreq;
params.maxfreq = (maxfreq == -1 ? mat->nrows : maxfreq);
params.minlen = minlen;
params.maxlen = (maxlen == -1 ? mat->ncols : maxlen);
params.tnitems = mat->ncols;
params.callback = process_itemset;
params.stateptr = stateptr;
params.rmarker = gk_ismalloc(mat->nrows, 0, "gk_find_frequent_itemsets: rmarker");
params.cand = gk_ikvmalloc(mat->ncols, "gk_find_frequent_itemsets: cand");
/* Perform the initial projection */
gk_csr_CreateIndex(mat, GK_CSR_COL);
pmat = itemsets_project_matrix(¶ms, mat, -1);
gk_csr_Free(&mat);
pattern = gk_imalloc(pmat->ncols, "gk_find_frequent_itemsets: pattern");
itemsets_find_frequent_itemsets(¶ms, pmat, 0, pattern);
gk_csr_Free(&pmat);
gk_free((void **)&pattern, ¶ms.rmarker, ¶ms.cand, LTERM);
}
void itemsets_find_frequent_itemsets(isparams_t *params, gk_csr_t *mat,
int preflen, int *prefix)
{
ssize_t i;
gk_csr_t *cmat;
/* Project each frequent column */
for (i=0; i<mat->ncols; i++) {
prefix[preflen] = mat->colids[i];
if (preflen+1 >= params->minlen)
(*params->callback)(params->stateptr, preflen+1, prefix,
mat->colptr[i+1]-mat->colptr[i], mat->colind+mat->colptr[i]);
if (preflen+1 < params->maxlen) {
cmat = itemsets_project_matrix(params, mat, i);
itemsets_find_frequent_itemsets(params, cmat, preflen+1, prefix);
gk_csr_Free(&cmat);
}
}
}
void ComputeNeighbors(params_t *params)
{
int i, j, nhits;
gk_csr_t *mat;
int32_t *marker;
gk_fkv_t *hits, *cand;
FILE *fpout;
printf("Reading data for %s...\n", params->infstem);
mat = gk_csr_Read(params->infstem, GK_CSR_FMT_CSR, 1, 0);
printf("#docs: %d, #nnz: %d.\n", mat->nrows, mat->rowptr[mat->nrows]);
/* compact the column-space of the matrices */
gk_csr_CompactColumns(mat);
/* perform auxiliary normalizations/pre-computations based on similarity */
gk_csr_Normalize(mat, GK_CSR_ROW, 2);
/* create the inverted index */
gk_csr_CreateIndex(mat, GK_CSR_COL);
/* create the output file */
fpout = (params->outfile ? gk_fopen(params->outfile, "w", "ComputeNeighbors: fpout") : NULL);
/* allocate memory for the necessary working arrays */
hits = gk_fkvmalloc(mat->nrows, "ComputeNeighbors: hits");
marker = gk_i32smalloc(mat->nrows, -1, "ComputeNeighbors: marker");
cand = gk_fkvmalloc(mat->nrows, "ComputeNeighbors: cand");
/* find the best neighbors for each query document */
gk_startwctimer(params->timer_1);
for (i=0; i<mat->nrows; i++) {
if (params->verbosity > 0)
printf("Working on query %7d\n", i);
/* find the neighbors of the ith document */
nhits = gk_csr_GetSimilarRows(mat,
mat->rowptr[i+1]-mat->rowptr[i],
mat->rowind+mat->rowptr[i],
mat->rowval+mat->rowptr[i],
GK_CSR_COS, params->nnbrs, params->minsim, hits,
marker, cand);
/* write the results in the file */
if (fpout) {
for (j=0; j<nhits; j++)
fprintf(fpout, "%8d %8d %.3f\n", i, hits[j].val, hits[j].key);
}
}
gk_stopwctimer(params->timer_1);
/* cleanup and exit */
if (fpout) gk_fclose(fpout);
gk_free((void **)&hits, &marker, &cand, LTERM);
gk_csr_Free(&mat);
return;
}
void FreeVault(vault_t *vault)
{
gk_csr_Free(&vault->mat);
gk_free((void **)&vault, LTERM);
}
int main(int argc, char *argv[])
{
ssize_t i, j, niter;
params_t *params;
gk_csr_t *mat;
FILE *fpout;
/* get command-line options */
params = parse_cmdline(argc, argv);
/* read the data */
mat = gk_csr_Read(params->infile, GK_CSR_FMT_METIS, 1, 1);
/* display some basic stats */
print_init_info(params, mat);
if (params->ntvs != -1) {
/* compute the pr for different randomly generated restart-distribution vectors */
float **prs;
prs = gk_fAllocMatrix(params->ntvs, mat->nrows, 0.0, "main: prs");
/* generate the random restart vectors */
for (j=0; j<params->ntvs; j++) {
for (i=0; i<mat->nrows; i++)
prs[j][i] = RandomInRange(931);
gk_fscale(mat->nrows, 1.0/gk_fsum(mat->nrows, prs[j], 1), prs[j], 1);
niter = gk_rw_PageRank(mat, params->lamda, params->eps, params->niter, prs[j]);
printf("tvs#: %zd; niters: %zd\n", j, niter);
}
/* output the computed pr scores */
fpout = gk_fopen(params->outfile, "w", "main: outfile");
for (i=0; i<mat->nrows; i++) {
for (j=0; j<params->ntvs; j++)
fprintf(fpout, "%.4e ", prs[j][i]);
fprintf(fpout, "\n");
}
gk_fclose(fpout);
gk_fFreeMatrix(&prs, params->ntvs, mat->nrows);
}
else if (params->ppr != -1) {
/* compute the personalized pr from the specified vertex */
float *pr;
pr = gk_fsmalloc(mat->nrows, 0.0, "main: pr");
pr[params->ppr-1] = 1.0;
niter = gk_rw_PageRank(mat, params->lamda, params->eps, params->niter, pr);
printf("ppr: %d; niters: %zd\n", params->ppr, niter);
/* output the computed pr scores */
fpout = gk_fopen(params->outfile, "w", "main: outfile");
for (i=0; i<mat->nrows; i++)
fprintf(fpout, "%.4e\n", pr[i]);
gk_fclose(fpout);
gk_free((void **)&pr, LTERM);
}
else {
/* compute the standard pr */
int jmax;
float diff, maxdiff;
float *pr;
pr = gk_fsmalloc(mat->nrows, 1.0/mat->nrows, "main: pr");
niter = gk_rw_PageRank(mat, params->lamda, params->eps, params->niter, pr);
printf("pr; niters: %zd\n", niter);
/* output the computed pr scores */
fpout = gk_fopen(params->outfile, "w", "main: outfile");
for (i=0; i<mat->nrows; i++) {
for (jmax=i, maxdiff=0.0, j=mat->rowptr[i]; j<mat->rowptr[i+1]; j++) {
if ((diff = fabs(pr[i]-pr[mat->rowind[j]])) > maxdiff) {
maxdiff = diff;
jmax = mat->rowind[j];
}
}
fprintf(fpout, "%.4e %10zd %.4e %10d\n", pr[i],
mat->rowptr[i+1]-mat->rowptr[i], maxdiff, jmax+1);
}
gk_fclose(fpout);
gk_free((void **)&pr, LTERM);
}
gk_csr_Free(&mat);
/* display some final stats */
print_final_info(params);
}