@Override public Result score(long user, long item) { return Results.create(item, model.getIntercept() + model.getUserBias(user) + model.getItemBias(item)); }
@Nonnull @Override public ResultMap scoreWithDetails(long user, @Nonnull Collection<Long> items) { Long2DoubleMap baselines = biasModel.getItemBiases(LongUtils.packedSet(items)); baselines = Vectors.addScalar(baselines, biasModel.getIntercept() + biasModel.getUserBias(user)); RealVector uvec = getUserPreferenceVector(user); if (uvec == null) { return Results.newResultMap(); } List<Result> results = new ArrayList<>(items.size()); LongIterator iter = LongIterators.asLongIterator(items.iterator()); while (iter.hasNext()) { long item = iter.nextLong(); RealVector ivec = model.getItemVector(item); if (ivec != null) { double score = computeScore(baselines.get(item), uvec, ivec); results.add(Results.create(item, score)); } } return Results.newResultMap(results); } }
@Override public Long2DoubleMap unapply(Long2DoubleMap input) { Long2DoubleMap biases = model.getUserBiases(input.keySet()); return Vectors.combine(input, biases, 1.0, itemBias); }
Transform(long uid) { user = uid; userBias = model.getIntercept() + model.getUserBias(user); }
Transform(long iid) { item = iid; itemBias = model.getIntercept() + model.getItemBias(item); }
@Test public void testManyItems() { Generator<Double> globals = doubles(); for (Map<Long,Double> map: someMaps(positiveLongs(), doubles())) { double bias = globals.next(); Long2DoubleMap itemBiases = Long2DoubleSortedArrayMap.create(map); BiasModel model = new UserItemBiasModel(bias, Long2DoubleMaps.EMPTY_MAP, itemBiases); assertThat(model.getIntercept(), equalTo(bias)); assertThat(model.getItemBiases(itemBiases.keySet()), equalTo(itemBiases)); for (Set<Long> users : someSets(positiveLongs())) { Long2DoubleMap biases = model.getItemBiases(LongUtils.packedSet(users)); for (long user: users) { if (itemBiases.containsKey(user)) { assertThat(biases.get(user), equalTo(itemBiases.get(user))); } else { assertThat(biases.get(user), equalTo(0.0)); } } } } }
@Test public void testManyUsers() { Generator<Double> globals = doubles(); for (Map<Long,Double> map: someMaps(positiveLongs(), doubles())) { double bias = globals.next(); Long2DoubleMap userBiases = Long2DoubleSortedArrayMap.create(map); BiasModel model = new UserItemBiasModel(bias, userBiases, Long2DoubleMaps.EMPTY_MAP); assertThat(model.getIntercept(), equalTo(bias)); assertThat(model.getUserBiases(userBiases.keySet()), equalTo(userBiases)); for (Set<Long> users : someSets(positiveLongs())) { Long2DoubleMap biases = model.getUserBiases(LongUtils.packedSet(users)); for (long user: users) { if (userBiases.containsKey(user)) { assertThat(biases.get(user), equalTo(userBiases.get(user))); } else { assertThat(biases.get(user), equalTo(0.0)); } } } } }
@Override public Long2DoubleMap apply(Long2DoubleMap input) { Long2DoubleMap biases = model.getItemBiases(input.keySet()); return Vectors.combine(input, biases, -1.0, -userBias); } }
@Test public void testComputeGlobalMean() { EntityFactory efac = new EntityFactory(); EntityCollectionDAOBuilder daoBuilder = new EntityCollectionDAOBuilder(); daoBuilder.addEntities(efac.rating(100, 200, 3.0), efac.rating(101, 200, 4.0), efac.rating(101, 201, 2.5), efac.rating(102, 203, 4.5)); Provider<GlobalBiasModel> biasProvider = new GlobalAverageRatingBiasModelProvider(daoBuilder.build()); BiasModel model = biasProvider.get(); assertThat(model.getIntercept(), closeTo(3.5, 1.0e-1)); } }
Transform(long uid) { user = uid; userBias = model.getIntercept() + model.getUserBias(user); }
Transform(long iid) { item = iid; itemBias = model.getIntercept() + model.getItemBias(item); }
@Override public Long2DoubleMap unapply(Long2DoubleMap input) { Long2DoubleMap biases = model.getItemBiases(input.keySet()); return Vectors.combine(input, biases, 1.0, userBias); }
@Nonnull @Override public Map<Long, Double> score(long user, @Nonnull Collection<Long> items) { LongSet itemSet = LongUtils.frozenSet(items); double base = model.getIntercept() + model.getUserBias(user); return LongUtils.flyweightMap(itemSet, iid -> base + model.getItemBias(iid)); }
@Nonnull @Override public ResultMap scoreWithDetails(long user, @Nonnull Collection<Long> items) { Long2DoubleMap baselines = biasModel.getItemBiases(LongUtils.packedSet(items)); baselines = Vectors.addScalar(baselines, biasModel.getIntercept() + biasModel.getUserBias(user)); RealVector uvec = getUserPreferenceVector(user); if (uvec == null) { return Results.newResultMap(); } List<Result> results = new ArrayList<>(items.size()); LongIterator iter = LongIterators.asLongIterator(items.iterator()); while (iter.hasNext()) { long item = iter.nextLong(); RealVector ivec = model.getItemVector(item); if (ivec != null) { double score = kernel.apply(baselines.get(item), uvec, ivec); results.add(Results.create(item, score)); } } return Results.newResultMap(results); } }
@Test public void testComputeUserMeans() { EntityFactory efac = new EntityFactory(); EntityCollectionDAOBuilder daoBuilder = new EntityCollectionDAOBuilder(); daoBuilder.addEntities(efac.rating(100, 200, 3.0), efac.rating(101, 200, 4.0), efac.rating(102, 201, 2.5), efac.rating(102, 203, 4.5), efac.rating(101, 203, 3.5)); LenskitConfiguration config = new LenskitConfiguration(); config.addRoot(BiasModel.class); config.bind(BiasModel.class).toProvider(UserAverageRatingBiasModelProvider.class); LenskitRecommender rec = LenskitRecommender.build(config, daoBuilder.build()); BiasModel model = rec.get(BiasModel.class); assertThat(model.getIntercept(), closeTo(3.5, 1.0e-3)); assertThat(model.getUserBias(100), closeTo(-0.5, 1.0e-3)); assertThat(model.getUserBias(101), closeTo(0.25, 1.0e-3)); assertThat(model.getUserBias(102), closeTo(0.0, 1.0e-3)); }
@Test public void testComputeMeans() { EntityFactory efac = new EntityFactory(); EntityCollectionDAOBuilder daoBuilder = new EntityCollectionDAOBuilder(); daoBuilder.addEntities(efac.rating(100, 200, 3.0), efac.rating(101, 200, 4.0), efac.rating(101, 201, 2.5), efac.rating(102, 203, 4.5), efac.rating(103, 203, 3.5)); LenskitConfiguration config = new LenskitConfiguration(); config.addRoot(BiasModel.class); config.bind(BiasModel.class).toProvider(ItemAverageRatingBiasModelProvider.class); LenskitRecommender rec = LenskitRecommender.build(config, daoBuilder.build()); BiasModel model = rec.get(BiasModel.class); assertThat(model.getIntercept(), closeTo(3.5, 1.0e-3)); assertThat(model.getItemBias(200), closeTo(0.0, 1.0e-3)); assertThat(model.getItemBias(201), closeTo(-1.0, 1.0e-3)); assertThat(model.getItemBias(203), closeTo(0.5, 1.0e-3)); } }
@Override public Long2DoubleMap apply(Long2DoubleMap input) { Long2DoubleMap biases = model.getItemBiases(input.keySet()); return Vectors.combine(input, biases, -1.0, -userBias); } }
@Override public Long2DoubleMap apply(Long2DoubleMap input) { Long2DoubleMap biases = model.getUserBiases(input.keySet()); return Vectors.combine(input, biases, -1.0, -itemBias); } }
@Nonnull @Override public ResultMap scoreWithDetails(long user, @Nonnull Collection<Long> items) { List<Result> results = new ArrayList<>(); double base = model.getIntercept() + model.getUserBias(user); LongIterator iter = LongIterators.asLongIterator(items.iterator()); while (iter.hasNext()) { long item = iter.nextLong(); results.add(Results.create(item, base + model.getItemBias(item))); } return Results.newResultMap(results); } }
@Test public void testComputeMeans() { EntityFactory efac = new EntityFactory(); EntityCollectionDAOBuilder daoBuilder = new EntityCollectionDAOBuilder(); daoBuilder.addEntities(efac.rating(100, 200, 3.0), efac.rating(101, 200, 4.0), efac.rating(102, 201, 2.5), efac.rating(102, 203, 4.5), efac.rating(101, 203, 3.5)); LenskitConfiguration config = new LenskitConfiguration(); config.addRoot(BiasModel.class); config.bind(BiasModel.class).to(UserBiasModel.class); LenskitRecommender rec = LenskitRecommender.build(config, daoBuilder.build()); BiasModel model = rec.get(BiasModel.class); assertThat(model.getIntercept(), closeTo(3.5, 1.0e-3)); assertThat(model.getUserBias(100), closeTo(-0.5, 1.0e-3)); assertThat(model.getUserBias(101), closeTo(0.25, 1.0e-3)); assertThat(model.getUserBias(102), closeTo(0.0, 1.0e-3)); } }