private double doItemSimilarity(long itemID1, long itemID2, FastIDSet preferring1)
throws TasteException {
double intersection = 0.0;
double union = 0.0;
for (Preference pref : getDataModel().getPreferencesForItem(itemID2)) {
long userID = pref.getUserID();
double weight = (double) getDataModel().getNumItems() / mUserPrefNum.get(userID);
if (preferring1.contains(userID)) {
intersection += weight;
union -= weight;
}
union += weight;
}
for (LongPrimitiveIterator it_user = preferring1.iterator(); it_user.hasNext(); ) {
long userID = (long) it_user.nextLong();
double weight = (double) getDataModel().getNumItems() / mUserPrefNum.get(userID);
union += weight;
}
if (intersection == 0) {
return Double.NaN;
}
return Math.log(intersection) / Math.log(union);
}
@Override
public double userSimilarity(long userID1, long userID2) throws TasteException {
DataModel dataModel = getDataModel();
FastIDSet xPrefs = dataModel.getItemIDsFromUser(userID1);
FastIDSet yPrefs = dataModel.getItemIDsFromUser(userID2);
int xPrefsSize = xPrefs.size();
int yPrefsSize = yPrefs.size();
if (xPrefsSize == 0 && yPrefsSize == 0) {
return Double.NaN;
}
if (xPrefsSize == 0 || yPrefsSize == 0) {
return 0.0;
}
double intersection = 0.0;
double union = 0.0;
for (LongPrimitiveIterator it_item = xPrefs.iterator(); it_item.hasNext(); ) {
long itemID = (long) it_item.nextLong();
double weight = (double) getDataModel().getNumUsers() / mItemPrefNum.get(itemID);
if (yPrefs.contains(itemID)) {
intersection += weight;
union -= weight;
}
union += weight;
}
for (LongPrimitiveIterator it_item = yPrefs.iterator(); it_item.hasNext(); ) {
long itemID = (long) it_item.nextLong();
double weight = (double) getDataModel().getNumUsers() / mItemPrefNum.get(itemID);
union += weight;
}
return Math.log(intersection) / Math.log(union);
}
@Override
public PreferenceArray getPreferencesForItem(long itemID) throws NoSuchItemException {
FastIDSet userIDs = preferenceForItems.get(itemID);
if (userIDs == null) {
throw new NoSuchItemException(itemID);
}
PreferenceArray prefArray = new BooleanItemPreferenceArray(userIDs.size());
int i = 0;
LongPrimitiveIterator it = userIDs.iterator();
while (it.hasNext()) {
prefArray.setUserID(i, it.nextLong());
prefArray.setItemID(i, itemID);
i++;
}
return prefArray;
}
@Override
public List<RecommendedItem> recommend(
long userID, int howMany, IDRescorer rescorer, boolean includeKnownItems)
throws TasteException {
Preconditions.checkArgument(howMany >= 1, "howMany must be at least 1");
log.debug("Recommending items for user ID '{}'", userID);
PreferenceArray preferencesFromUser = getDataModel().getPreferencesFromUser(userID);
FastIDSet possibleItemIDs = getAllOtherItems(userID, preferencesFromUser, includeKnownItems);
TopItems.Estimator<Long> estimator = new Estimator(userID);
List<RecommendedItem> topItems =
TopItems.getTopItems(howMany, possibleItemIDs.iterator(), rescorer, estimator);
log.debug("Recommendations are: {}", topItems);
return topItems;
}
@Override
protected void reduce(Vertex vertex, Iterable<Vertex> vertices, Context ctx)
throws IOException, InterruptedException {
int howmany = 0;
FastIDSet bufferedVertexIDs = new FastIDSet();
for (Vertex firstVertexOfMissingEdge : vertices) {
LongPrimitiveIterator bufferedVertexIdsIterator = bufferedVertexIDs.iterator();
while (bufferedVertexIdsIterator.hasNext()) {
Vertex secondVertexOfMissingEdge = new Vertex(bufferedVertexIdsIterator.nextLong());
UndirectedEdge missingEdge =
new UndirectedEdge(firstVertexOfMissingEdge, secondVertexOfMissingEdge);
JoinableUndirectedEdge key = new JoinableUndirectedEdge(missingEdge, false);
VertexOrMarker value = new VertexOrMarker(vertex);
log.trace("{} -> {}", key, value);
howmany++;
ctx.write(key, value);
}
bufferedVertexIDs.add(firstVertexOfMissingEdge.getId());
}
log.debug("{} triads on vertex {}", howmany, vertex);
}
/**
* Creates a new {@link GenericDataModel} from the given users (and their preferences). This
* {@link DataModel} retains all this information in memory and is effectively immutable.
*
* @param userData users to include
* @param timestamps optionally, provided timestamps of preferences as milliseconds since the
* epoch. User IDs are mapped to maps of item IDs to Long timestamps.
*/
public GenericBooleanPrefDataModel(
FastByIDMap<FastIDSet> userData, FastByIDMap<FastByIDMap<Long>> timestamps) {
Preconditions.checkArgument(userData != null, "userData is null");
this.preferenceFromUsers = userData;
this.preferenceForItems = new FastByIDMap<>();
FastIDSet itemIDSet = new FastIDSet();
for (Map.Entry<Long, FastIDSet> entry : preferenceFromUsers.entrySet()) {
long userID = entry.getKey();
FastIDSet itemIDs = entry.getValue();
itemIDSet.addAll(itemIDs);
LongPrimitiveIterator it = itemIDs.iterator();
while (it.hasNext()) {
long itemID = it.nextLong();
FastIDSet userIDs = preferenceForItems.get(itemID);
if (userIDs == null) {
userIDs = new FastIDSet(2);
preferenceForItems.put(itemID, userIDs);
}
userIDs.add(userID);
}
}
this.itemIDs = itemIDSet.toArray();
itemIDSet = null; // Might help GC -- this is big
Arrays.sort(itemIDs);
this.userIDs = new long[userData.size()];
int i = 0;
LongPrimitiveIterator it = userData.keySetIterator();
while (it.hasNext()) {
userIDs[i++] = it.next();
}
Arrays.sort(userIDs);
this.timestamps = timestamps;
}
