SGSparseVector<ST> CHashedSparseFeatures<ST>::hash_vector(SGSparseVector<ST> vec, int32_t dim,
bool use_quadratic, bool keep_linear_terms)
{
SGVector<ST> h_vec(dim);
SGVector<ST>::fill_vector(h_vec, dim, 0);
int32_t hash_cache_size = use_quadratic ? vec.num_feat_entries : 0;
SGVector<uint32_t> hash_cache(hash_cache_size);
for (index_t i=0; i<vec.num_feat_entries; i++)
{
uint32_t hash = CHash::MurmurHash3((uint8_t* ) &vec.features[i].feat_index, sizeof (index_t),
vec.features[i].feat_index);
if (use_quadratic)
hash_cache[i] = hash;
if ( (!use_quadratic) || keep_linear_terms )
h_vec[hash % dim] += vec.features[i].entry;
}
if (use_quadratic)
{
for (index_t i=0; i<vec.num_feat_entries; i++)
{
index_t n_idx = vec.features[i].feat_index + vec.features[i].feat_index;
index_t idx = CHash::MurmurHash3((uint8_t* ) &n_idx, sizeof(index_t),
vec.features[i].feat_index) % dim;
h_vec[idx] += vec.features[i].entry * vec.features[i].entry;
for (index_t j=i+1; j<vec.num_feat_entries; j++)
{
idx = (hash_cache[i] ^ hash_cache[j]) % dim;
h_vec[idx] += vec.features[i].entry * vec.features[j].entry;
}
}
}
int32_t num_nnz_features = 0;
for (index_t i=0; i<dim; i++)
{
if (h_vec[i]!=0)
num_nnz_features++;
}
SGSparseVector<ST> sv(num_nnz_features);
int32_t sparse_index = 0;
for (index_t i=0; i<dim; i++)
{
if (h_vec[i]!=0)
{
sv.features[sparse_index].entry = h_vec[i];
sv.features[sparse_index++].feat_index = i;
}
}
return sv;
}
int main(void)
{
// generate 20 random numbers on the host
thrust::host_vector<int> h_vec(20);
thrust::generate(h_vec.begin(), h_vec.end(), rand);
// interface to CUDA code
sort_on_device(h_vec);
// print sorted array
thrust::copy(h_vec.begin(), h_vec.end(), std::ostream_iterator<int>(std::cout, "\n"));
return 0;
}
