private static FastByIDMap<float[]> buildIdentity(long[] idsInOrder) {
int n = idsInOrder.length;
FastByIDMap<float[]> identity = new FastByIDMap<float[]>(n);
for (int i = 0; i < n; i++) {
float[] rowOrCol = new float[n];
rowOrCol[i] = 1.0f;
identity.put(idsInOrder[i], rowOrCol);
}
return identity;
}
private static long[] idsInOrder(FastByIDMap<?> input) {
int n = input.size();
long[] idsInOrder = new long[n];
int count = 0;
LongPrimitiveIterator it = input.keySetIterator();
while (it.hasNext()) {
idsInOrder[count] = it.nextLong();
count++;
}
Preconditions.checkState(n == count);
return idsInOrder;
}
private static FastByIDMap<float[]> buildBinarizedMatrix(
FastByIDMap<FastByIDFloatMap> input, long[] idsInOrder) {
int n = idsInOrder.length;
FastByIDMap<float[]> result = new FastByIDMap<float[]>();
for (FastByIDMap.MapEntry<FastByIDFloatMap> entry : input.entrySet()) {
float[] rowOrCol = new float[n];
FastByIDFloatMap inputValues = entry.getValue();
for (int i = 0; i < n; i++) {
if (inputValues.containsKey(idsInOrder[i])) {
rowOrCol[i] = 1.0f;
}
}
result.put(entry.getKey(), rowOrCol);
}
return result;
}
public static void writeUnreducedModel(File inputDir) throws IOException {
Preconditions.checkNotNull(inputDir);
Preconditions.checkArgument(inputDir.exists());
Preconditions.checkArgument(inputDir.isDirectory());
FastByIDMap<FastIDSet> knownItemIDs = new FastByIDMap<FastIDSet>(10000, 1.25f);
FastByIDMap<FastByIDFloatMap> RbyRow = new FastByIDMap<FastByIDFloatMap>(10000, 1.25f);
FastByIDMap<FastByIDFloatMap> RbyColumn = new FastByIDMap<FastByIDFloatMap>(10000, 1.25f);
FastIDSet itemTagIDs = new FastIDSet(1000, 1.25f);
FastIDSet userTagIDs = new FastIDSet(1000, 1.25f);
InputFilesReader.readInputFiles(
knownItemIDs, RbyRow, RbyColumn, itemTagIDs, userTagIDs, inputDir);
int numUsers = RbyRow.size();
int numItems = RbyColumn.size();
if (numUsers == 0 || numItems == 0) {
log.warn("No input?");
return;
}
FastByIDMap<float[]> X;
FastByIDMap<float[]> Y;
if (numUsers < numItems) {
log.info(
"{} users < {} items; input will be written as the feature-item matrix",
numUsers,
numItems);
long[] idsInOrder = idsInOrder(RbyRow);
X = buildIdentity(idsInOrder);
Y = buildBinarizedMatrix(RbyColumn, idsInOrder);
} else {
log.info(
"{} users >= {} items; input will be written as the user-feature matrix",
numUsers,
numItems);
long[] idsInOrder = idsInOrder(RbyColumn);
X = buildBinarizedMatrix(RbyRow, idsInOrder);
Y = buildIdentity(idsInOrder);
}
Generation generation = new Generation(knownItemIDs, X, Y, itemTagIDs, userTagIDs);
GenerationSerializer.writeGeneration(generation, new File(inputDir, "model.bin.gz"));
}
