sparse_joint_feature_vector.cpp

/*
 * Copyright(C) 2014 Abinash Panda
 * Written(W) 2014 Abinash Panda
 */

#include <shogun/base/init.h>
#include <shogun/lib/Hash.h>
#include <shogun/lib/SGSparseVector.h>
#include <shogun/features/SparseFeatures.h>
#include <shogun/mathematics/Random.h>
#include <shogun/io/SGIO.h>
#include <shogun/lib/Time.h>

using namespace shogun;

SGSparseVector<float64_t> get_sparse_joint_feature_vector(
        SGSparseVector<float64_t> vec,
        SGVector<int32_t> labels, int32_t hash_dim)
{
	SGSparseVector<float64_t> h_vec(vec.num_feat_entries * labels.vlen);
	index_t k = 0;

	for (int32_t i = 0; i < labels.vlen; i++)
	{
		uint32_t seed = (uint32_t)labels[i];

		for (int32_t j = 0; j < vec.num_feat_entries; j++)
		{
			uint32_t hash = CHash::MurmurHash3(
			                        (uint8_t *)&vec.features[j].feat_index,
			                        sizeof(index_t), seed);
			h_vec.features[k].feat_index = (hash >> 1) % hash_dim;
			h_vec.features[k++].entry =
			        (hash % 2 == 1 ? -1.0 : 1.0) * vec.features[j].entry;
		}

	}

	h_vec.sort_features(true);
	return h_vec;
}


int main(int argc, char ** argv)
{
	init_shogun_with_defaults();

	int32_t num_features = 1 << 12;
	int32_t dim_features = 1 << 28;

	uint32_t hash_seed = 23;
	int32_t hash_dim = 1 << 18;

	CRandom * prng = new CRandom(12345);
	SGSparseVector<float64_t> sparse = SGSparseVector<float64_t>(num_features);

	for (int32_t i = 0; i < num_features; i++)
	{
		sparse.features[i].feat_index = prng->random(0, dim_features);
		sparse.features[i].entry = 1.0;
	}

	sparse.sort_features(true);

	SGVector<int32_t> labels(1000);
	SGVector<int32_t>::range_fill_vector(labels.vector, labels.vlen, 123);

	CTime * timer = new CTime();
	timer->start();

	SGSparseVector<float64_t> psi = get_sparse_joint_feature_vector(
	                                        sparse, labels, hash_dim);

	timer->stop();
	float64_t iter_time = timer->cur_time_diff();
	SG_SPRINT("[%.1f sec] sparse joint feature vector implementation \n",
	          iter_time);
	SG_SPRINT("Number of features in sparse vector = %d\n",
	          sparse.num_feat_entries);
	SG_SPRINT("Number of featurs in joint vector   = %d\n",
	          psi.num_feat_entries)

	SG_UNREF(timer);

	exit_shogun();

	return 0;
}

[0.9 sec] sparse joint feature vector implementation
Number of features in sparse vector = 4096
Number of featurs in joint vector   = 235383