/** @return the transpose of current matrix */
public SparseMatrix transpose() {
if (isCCSUsed) {
SparseMatrix tr = new SparseMatrix(numColumns, numRows);
tr.copyCRS(this.rowData, this.rowPtr, this.colInd);
tr.copyCCS(this.colData, this.colPtr, this.rowInd);
return tr;
} else {
Table<Integer, Integer, Double> dataTable = HashBasedTable.create();
for (MatrixEntry me : this) dataTable.put(me.column(), me.row(), me.get());
return new SparseMatrix(numColumns, numRows, dataTable);
}
}
@Override
protected void initModel() throws Exception {
testIndexMatrix = new SparseMatrix(testMatrix);
for (MatrixEntry me : testIndexMatrix) {
int u = me.row();
int i = me.column();
testIndexMatrix.set(u, i, 0.0);
}
// global svd P Q to calculate the kernel value between users (or items)
P = new DenseMatrix(numUsers, numFactors);
Q = new DenseMatrix(numItems, numFactors);
// initialize model
if (initByNorm) {
P.init(initMean, initStd);
Q.init(initMean, initStd);
} else {
P.init(); // P.init(smallValue);
Q.init(); // Q.init(smallValue);
}
this.buildGlobalModel();
}
@Override
protected void buildModel() throws Exception {
for (int iter = 1; iter <= numIters; iter++) {
// update W by fixing H
for (int u = 0; u < W.numRows(); u++) {
SparseVector uv = V.row(u);
if (uv.getCount() > 0) {
SparseVector euv = new SparseVector(V.numColumns());
for (int j : uv.getIndex()) euv.set(j, predict(u, j));
for (int f = 0; f < W.numColumns(); f++) {
DenseVector fv = H.row(f, false);
double real = fv.inner(uv);
double estm = fv.inner(euv) + 1e-9;
W.set(u, f, W.get(u, f) * (real / estm));
}
}
}
// update H by fixing W
DenseMatrix trW = W.transpose();
for (int j = 0; j < H.numColumns(); j++) {
SparseVector jv = V.column(j);
if (jv.getCount() > 0) {
SparseVector ejv = new SparseVector(V.numRows());
for (int u : jv.getIndex()) ejv.set(u, predict(u, j));
for (int f = 0; f < H.numRows(); f++) {
DenseVector fv = trW.row(f, false);
double real = fv.inner(jv);
double estm = fv.inner(ejv) + 1e-9;
H.set(f, j, H.get(f, j) * (real / estm));
}
}
}
// compute errors
loss = 0;
errs = 0;
for (MatrixEntry me : V) {
int u = me.row();
int j = me.column();
double ruj = me.get();
if (ruj > 0) {
double euj = predict(u, j) - ruj;
errs += euj * euj;
loss += euj * euj;
}
}
errs *= 0.5;
loss *= 0.5;
if (isConverged(iter)) break;
}
}
@Override
protected double predict(int u, int i) throws Exception {
return predictMatrix.get(u, i);
}
@Override
protected void buildModel() throws Exception {
// Pre-calculating similarity:
int completeModelCount = 0;
LLORMAUpdater[] learners = new LLORMAUpdater[multiThreadCount];
int[] anchorUser = new int[modelMax];
int[] anchorItem = new int[modelMax];
int modelCount = 0;
int[] runningThreadList = new int[multiThreadCount];
int runningThreadCount = 0;
int waitingThreadPointer = 0;
int nextRunningSlot = 0;
cumPrediction = new SparseMatrix(testIndexMatrix);
cumWeight = new SparseMatrix(testIndexMatrix);
// Parallel training:
while (completeModelCount < modelMax) {
int u_t = (int) Math.floor(Math.random() * numUsers);
List<Integer> itemList = trainMatrix.getColumns(u_t);
if (itemList != null) {
if (runningThreadCount < multiThreadCount && modelCount < modelMax) {
// Selecting a new anchor point:
int idx = (int) Math.floor(Math.random() * itemList.size());
int i_t = itemList.get(idx);
anchorUser[modelCount] = u_t;
anchorItem[modelCount] = i_t;
// Preparing weight vectors:
DenseVector w =
kernelSmoothing(numUsers, u_t, KernelSmoothing.EPANECHNIKOV_KERNEL, 0.8, false);
DenseVector v =
kernelSmoothing(numItems, i_t, KernelSmoothing.EPANECHNIKOV_KERNEL, 0.8, true);
// Starting a new local model learning:
learners[nextRunningSlot] =
new LLORMAUpdater(
modelCount,
localNumFactors,
numUsers,
numItems,
u_t,
i_t,
localLRate,
localRegU,
localRegI,
localNumIters,
w,
v,
trainMatrix);
learners[nextRunningSlot].start();
runningThreadList[runningThreadCount] = modelCount;
runningThreadCount++;
modelCount++;
nextRunningSlot++;
} else if (runningThreadCount > 0) {
// Joining a local model which was done with learning:
try {
learners[waitingThreadPointer].join();
} catch (InterruptedException ie) {
System.out.println("Join failed: " + ie);
}
int mp = waitingThreadPointer;
int mc = completeModelCount;
completeModelCount++;
// Predicting with the new local model and all previous
// models:
predictMatrix = new SparseMatrix(testIndexMatrix);
for (MatrixEntry me : testMatrix) {
int u = me.row();
int i = me.column();
double weight =
KernelSmoothing.kernelize(
getUserSimilarity(anchorUser[mc], u),
0.8,
KernelSmoothing.EPANECHNIKOV_KERNEL)
* KernelSmoothing.kernelize(
getItemSimilarity(anchorItem[mc], i),
0.8,
KernelSmoothing.EPANECHNIKOV_KERNEL);
double newPrediction =
(learners[mp].getUserFeatures().row(u).inner(learners[mp].getItemFeatures().row(i)))
* weight;
cumWeight.set(u, i, cumWeight.get(u, i) + weight);
cumPrediction.set(u, i, cumPrediction.get(u, i) + newPrediction);
double prediction = cumPrediction.get(u, i) / cumWeight.get(u, i);
if (Double.isNaN(prediction) || prediction == 0.0) {
prediction = globalMean;
}
if (prediction < minRate) {
prediction = minRate;
} else if (prediction > maxRate) {
prediction = maxRate;
}
predictMatrix.set(u, i, prediction);
}
if (completeModelCount % 5 == 0) {
Logs.debug(
"{}{} iter {}:[MAE,RMSE,NMAE,rMAE,rRMSE,MPE] {}",
algoName,
foldInfo,
completeModelCount,
"[" + Recommender.getEvalInfo(evalRatings()) + "]");
}
nextRunningSlot = waitingThreadPointer;
waitingThreadPointer = (waitingThreadPointer + 1) % multiThreadCount;
runningThreadCount--;
}
}
}
}
