int main(){
std::cout << setCross (SetFactory(0), SetFactory(1)) << std::endl;
std::cout << setIntersect (SetFactory(0), SetFactory(1)) << std::endl;
std::cout << setMinus (SetFactory(0), SetFactory(1)) << std::endl;
std::cout << setUnion (SetFactory(0), SetFactory(1)) << std::endl;
std::cout << setCardinality (SetFactory(1).add( 2).add( 3)) << std::endl;
return 0;
}
void compJaccSimAccu(Data& data, Params& params) {
std::vector<ModelMF> mfModels;
std::vector<ModelMF> origModels;
for (int i = 1; i < 3; i++) {
std::string prefix = "mf_0_0_10_" + std::to_string(i);
ModelMF fullModel(params, params.seed);
fullModel.loadFacs(prefix.c_str());
mfModels.push_back(fullModel);
std::string uFName = "uFac_" + std::to_string(fullModel.nUsers) + "_10_0.txt";
std::string iFName = "iFac_" + std::to_string(fullModel.nItems+1) + "_10_0.txt";
ModelMF origModel(params, uFName.c_str(), iFName.c_str(), params.seed);
origModels.push_back(origModel);
}
/*
for (int i = 0; i < 2; i++) {
std::string prefix = "mfrand_" + std::to_string(i) + "_0_20";
ModelMF fullModel(params, params.seed);
fullModel.loadFacs(prefix.c_str());
mfModels.push_back(fullModel);
std::string uFName = "uFac_" + std::to_string(fullModel.nUsers) + "_20_"
+ std::to_string(i) + ".txt";
std::string iFName = "iFac_" + std::to_string(fullModel.nItems) + "_20_"
+ std::to_string(i) + ".txt";
ModelMF origModel(params, uFName.c_str(), iFName.c_str(), params.seed);
origModels.push_back(origModel);
}
*/
int nModels = mfModels.size();
std::unordered_set<int> invalidUsers;
std::unordered_set<int> invalidItems;
std::string modelSign = mfModels[0].modelSignature();
std::cout << "\nModel sign: " << modelSign << std::endl;
std::string prefix = std::string(params.prefix) + "_" + modelSign + "_invalUsers.txt";
std::vector<int> invalUsersVec = readVector(prefix.c_str());
prefix = std::string(params.prefix) + "_" + modelSign + "_invalItems.txt";
std::vector<int> invalItemsVec = readVector(prefix.c_str());
for (auto v: invalUsersVec) {
invalidUsers.insert(v);
}
for (auto v: invalItemsVec) {
invalidItems.insert(v);
}
for (int i = 0; i < nModels; i++) {
auto& fullModel = mfModels[i];
auto& origModel = origModels[i];
std::cout << "Model: " << i << std::endl;
std::cout << "Train RMSE: " << fullModel.RMSE(data.trainMat, invalidUsers,
invalidItems, origModel) << std::endl;
std::cout << "Test RMSE: " << fullModel.RMSE(data.testMat, invalidUsers,
invalidItems, origModel) << std::endl;
std::cout << "Val RMSE: " << fullModel.RMSE(data.valMat, invalidUsers,
invalidItems, origModel) << std::endl;
//std::cout << "Full RMSE: " << fullModel.fullLowRankErr(data, invalidUsers,
// invalidItems, origModel) << std::endl;
}
std::vector<double> epsilons = {0.025, 0.05, 0.1, 0.25, 0.5, 1.0};
//std::vector<double> epsilons = {0.5};
int nUsers = data.trainMat->nrows;
int nItems = data.trainMat->ncols;
std::cout << "nModels: " << nModels << std::endl;
for (auto&& epsilon: epsilons) {
std::cout << "epsilon: " << epsilon << std::endl;
std::vector<std::vector<float>> itemsJacSims(nItems);
std::vector<std::vector<float>> itemAccuCount(nItems);
std::vector<std::vector<float>> itemPearsonCorr(nItems);
#pragma omp parallel for
for (int item = 0; item < nItems; item++) {
if (invalidItems.count(item) > 0) {
continue;
}
std::vector<std::unordered_set<int>> modelAccuItems(nModels);
std::vector<std::vector<float>> itemErr(nModels);
std::vector<float> itemErrMean(nModels, 0);
std::vector<bool> ratedUsers(nUsers, false);
int count = 0;
for (int uu = data.trainMat->colptr[item];
uu < data.trainMat->colptr[item+1]; uu++) {
ratedUsers[data.trainMat->colind[uu]] = true;
}
for (int user = 0; user < nUsers; user++) {
if (invalidUsers.count(user) > 0) {
continue;
}
if (ratedUsers[user]) {
continue;
}
for (int i = 0; i < nModels; i++) {
auto& mfModel = mfModels[i];
auto& origModel = origModels[i];
float r_ui_est = mfModel.estRating(user, item);
float r_ui = origModel.estRating(user, item);
float diff = fabs(r_ui - r_ui_est);
itemErr[i].push_back(diff);
itemErrMean[i] += diff;
if (diff <= epsilon) {
modelAccuItems[i].insert(user);
}
}
count++;
}
for (int i = 0; i < nModels; i++) {
itemErrMean[i] /= count;
}
for (int i = 0; i < nModels; i++) {
itemAccuCount[item].push_back(modelAccuItems[i].size());
for (int j = i+1; j < nModels; j++) {
//compute overlap between model i and model j
float intersectCount = (float) setIntersect(modelAccuItems[i], modelAccuItems[j]);
float unionCount = (float) setUnion(modelAccuItems[i], modelAccuItems[j]);
float jacSim = 0;
if (unionCount > 0) {
jacSim = intersectCount/unionCount;
}
itemsJacSims[item].push_back(jacSim);
//compute correlation between model i & j for the item
itemPearsonCorr[item].push_back(pearsonCorr(itemErr[i], itemErr[j],
itemErrMean[i], itemErrMean[j]));
}
}
}
std::string opFName = std::string(params.prefix) + "_" +
std::to_string(epsilon) + "_modelsJacSim.txt";
std::string opFName2 = std::string(params.prefix) + "_" +
std::to_string(epsilon) + "_modelsPearson.txt";
std::cout << "Writing... " << opFName << std::endl;
std::ofstream opFile(opFName.c_str());
std::ofstream opFile2(opFName2.c_str());
for (int item = 0; item < nItems; item++) {
opFile << item << " ";
for (auto&& sim: itemsJacSims[item]) {
opFile << sim << " ";
}
for (auto&& accu: itemAccuCount[item]) {
opFile << accu << " ";
}
opFile << std::endl;
opFile2 << item << " ";
for (auto&& corr: itemPearsonCorr[item]) {
opFile2 << corr << " ";
}
opFile2 << std::endl;
}
opFile.close();
opFile2.close();
}
}