internal MostSimilarEstimator(User toUser,
UserCorrelation correlation,
Rescorer <Pair <User, User> > rescorer)
{
this.toUser = toUser;
this.correlation = correlation;
this.rescorer = rescorer;
}
public SpearmanCorrelation(DataModel dataModel)
{
if (dataModel == null)
{
throw new ArgumentNullException("dataModel is null");
}
this.rankingUserCorrelation = new PearsonCorrelation(dataModel);
this.refreshLock = new ReentrantLock();
}
public SpearmanCorrelation(UserCorrelation rankingUserCorrelation)
{
if (rankingUserCorrelation == null)
{
throw new ArgumentNullException("rankingUserCorrelation is null");
}
this.rankingUserCorrelation = rankingUserCorrelation;
this.refreshLock = new ReentrantLock();
}
public SpearmanCorrelation(DataModel dataModel)
{
if (dataModel == null)
{
throw new ArgumentNullException("dataModel is null");
}
this.rankingUserCorrelation = new PearsonCorrelation(dataModel);
this.refreshLock = new ReentrantLock();
}
public SpearmanCorrelation(UserCorrelation rankingUserCorrelation)
{
if (rankingUserCorrelation == null)
{
throw new ArgumentNullException("rankingUserCorrelation is null");
}
this.rankingUserCorrelation = rankingUserCorrelation;
this.refreshLock = new ReentrantLock();
}
/**
* <p>Constructs a {@link FarthestNeighborClusterSimilarity} based on the given {@link UserCorrelation}.
* By setting <code>samplingPercentage</code> to a value less than 1.0, this implementation will only examine
* that fraction of all user-user correlations between two clusters, increasing performance at the expense
* of accuracy.</p>
*
* @param Correlation
* @param samplingPercentage
* @since 1.4.5
*/
public FarthestNeighborClusterSimilarity(UserCorrelation correlation, double samplingPercentage)
{
if (correlation == null) {
throw new ArgumentNullException("Correlation is null");
}
if (double.IsNaN(samplingPercentage) || samplingPercentage <= 0.0 || samplingPercentage > 1.0)
{
throw new ArgumentException("samplingPercentage is invalid: " + samplingPercentage);
}
this.correlation = correlation;
this.samplingPercentage = samplingPercentage;
}
/// <summary>
/// construct a NearestNUserNeighborhood
/// </summary>
/// <param name="n">n Neighborhood size</param>
/// <param name="userCorrelation">nearness metric</param>
/// <param name="dataModel">data Model</param>
/// <param name="samplingRate">percentage of users to consider when building Neighborhood -- decrease to
/// trade quality for performance</param>
public NearestNUserNeighborhood(int n,
UserCorrelation userCorrelation,
DataModel dataModel,
double samplingRate)
: base(userCorrelation, dataModel, samplingRate)
{
if (n < 1)
{
throw new ArgumentException("n must be at least 1");
}
this.cache = new SoftCache <Object, ICollection <User> >(new Retriever(this, n), dataModel.GetNumUsers());
}
/**
* @param threshold similarity threshold
* @param userCorrelation similarity metric
* @param dataModel data Model
* @param samplingRate percentage of users to consider when building Neighborhood -- decrease to
* trade quality for performance
* @throws IllegalArgumentException if threshold or samplingRate is {@link Double#NaN},
* or if samplingRate is not positive and less than or equal to 1.0, or if userCorrelation
* or dataModel are <code>null</code>
* @since 1.3
*/
public ThresholdUserNeighborhood(double threshold,
UserCorrelation userCorrelation,
DataModel dataModel,
double samplingRate)
: base(userCorrelation, dataModel, samplingRate)
{
if (Double.IsNaN(threshold))
{
throw new ArgumentException("threshold must not be NaN");
}
this.cache = new SoftCache <Object, ICollection <User> >(new Retriever(this, threshold), dataModel.GetNumUsers());
}
public GenericUserBasedRecommender(DataModel dataModel,
UserNeighborhood neighborhood,
UserCorrelation correlation)
: base(dataModel)
{
if (neighborhood == null)
{
throw new ArgumentNullException("Neighborhood is null");
}
this.neighborhood = neighborhood;
this.correlation = correlation;
this.refreshLock = new ReentrantLock();
}
/// <summary>
/// <p>Constructs a <see cref="taste.Recommender.NearestNeighborClusterSimilarity">NearestNeighborClusterSimilarity</see> based on the given
/// <see cref="taste.Correlation.UserCorrelation">UserCorrelation</see>.
/// By setting <code>samplingPercentage</code> to a value less than 1.0, this implementation will only examine
/// that fraction of all user-user correlations between two clusters, increasing performance at the expense
/// of accuracy.</p>
/// </summary>
/// <param name="Correlation"></param>
/// <param name="samplingPercentage"></param>
public NearestNeighborClusterSimilarity(UserCorrelation correlation, double samplingPercentage)
{
if (correlation == null)
{
throw new ArgumentNullException("Correlation is null");
}
if (double.IsNaN(samplingPercentage) || samplingPercentage <= 0.0 || samplingPercentage > 1.0)
{
throw new ArgumentException("samplingPercentage is invalid: " + samplingPercentage);
}
this.correlation = correlation;
this.samplingPercentage = samplingPercentage;
}
public AbstractUserNeighborhood(UserCorrelation userCorrelation,
DataModel dataModel,
double samplingRate)
{
if (userCorrelation == null || dataModel == null)
{
throw new ArgumentNullException("userCorrelation or dataModel is null");
}
if (Double.IsNaN(samplingRate) || samplingRate <= 0.0 || samplingRate > 1.0)
{
throw new ArgumentException("samplingRate must be in (0,1]");
}
this.userCorrelation = userCorrelation;
this.dataModel = dataModel;
this.samplingRate = samplingRate;
}
public AbstractUserNeighborhood(UserCorrelation userCorrelation,
DataModel dataModel,
double samplingRate)
{
if (userCorrelation == null || dataModel == null)
{
throw new ArgumentNullException("userCorrelation or dataModel is null");
}
if (Double.IsNaN(samplingRate) || samplingRate <= 0.0 || samplingRate > 1.0)
{
throw new ArgumentException("samplingRate must be in (0,1]");
}
this.userCorrelation = userCorrelation;
this.dataModel = dataModel;
this.samplingRate = samplingRate;
}
public ICollection <User> GetValue(Object key)
{
if (log.IsInfoEnabled)
{
log.Info("Computing Neighborhood around user ID '" + key + '\'');
}
DataModel dataModel = _owner.DataModel;
User theUser = dataModel.GetUser(key);
IList <User> neighborhood = new List <User>();
IEnumerable <User> users = dataModel.GetUsers();
UserCorrelation userCorrelationImpl = _owner.UserCorrelation;
IEnumerator <User> en = users.GetEnumerator();
while (en.MoveNext())
{
User user = en.Current;
if (_owner.SampleForUser && !key.Equals(user.ID))
{
double theCorrelation = userCorrelationImpl.GetUserCorrelation(theUser, user);
if (!Double.IsNaN(theCorrelation) && theCorrelation >= threshold)
{
neighborhood.Add(user);
}
}
}
if (log.IsInfoEnabled)
{
log.Info("UserNeighborhood around user ID '" + key + "' is: " + neighborhood);
}
//return Collections.unmodifiableList(Neighborhood);
return(neighborhood);
}
/**
* @param threshold similarity threshold
* @param userCorrelation similarity metric
* @param dataModel data Model
* @throws IllegalArgumentException if threshold is {@link Double#NaN},
* or if samplingRate is not positive and less than or equal to 1.0, or if userCorrelation
* or dataModel are <code>null</code>
*/
public ThresholdUserNeighborhood(double threshold,
UserCorrelation userCorrelation,
DataModel dataModel)
: this(threshold, userCorrelation, dataModel, 1.0)
{
}
public ICollection <User> GetValue(Object key)
{
if (log.IsInfoEnabled)
{
log.Info("Computing Neighborhood around user ID '" + key + '\'');
}
DataModel dataModel = owner.DataModel;
User theUser = dataModel.GetUser(key);
UserCorrelation userCorrelationImpl = owner.UserCorrelation;
LinkedList <UserCorrelationPair> queue = new LinkedList <UserCorrelationPair>();
bool full = false;
foreach (User user in dataModel.GetUsers())
{
if (owner.SampleForUser && !key.Equals(user.ID))
{
double theCorrelation = userCorrelationImpl.GetUserCorrelation(theUser, user);
if (!Double.IsNaN(theCorrelation) && (!full || theCorrelation > queue.Last.Value.Correlation))
{
LinkedListNode <UserCorrelationPair> iterator = queue.Last;
while (iterator != null && iterator.Previous != null)
{
iterator = iterator.Previous;
if (theCorrelation <= iterator.Value.Correlation)
{
iterator = iterator.Next;
break;
}
if (iterator == queue.First)
{
break;
}
}
UserCorrelationPair pair = new UserCorrelationPair(user, theCorrelation);
if (iterator == null)
{
queue.AddFirst(pair);
}
else
{
queue.AddAfter(iterator, pair);
}
if (full)
{
queue.RemoveLast();
}
else if (queue.Count > n)
{
full = true;
queue.RemoveLast();
}
}
}
}
IList <User> neighborhood = new List <User>(queue.Count);
foreach (UserCorrelationPair pair in queue)
{
//assert pair != null;
neighborhood.Add(pair.User);
}
if (log.IsInfoEnabled)
{
log.Info("UserNeighborhood around user ID '" + key + "' is: " + neighborhood);
}
//return Collections.unmodifiableList(Neighborhood);
return(neighborhood);
}
/**
* <p>Constructs a {@link FarthestNeighborClusterSimilarity} based on the given {@link UserCorrelation}.
* All user-user correlations are examined.</p>
*
* @param Correlation
*/
public FarthestNeighborClusterSimilarity(UserCorrelation correlation)
:this(correlation, 1.0)
{
}
/// <summary>
/// construct a NearestNUserNeighborhood
/// </summary>
/// <param name="n">Neighborhood size</param>
/// <param name="userCorrelation">nearness metric</param>
/// <param name="dataModel">data Model</param>
public NearestNUserNeighborhood(int n,
UserCorrelation userCorrelation,
DataModel dataModel)
: this(n, userCorrelation, dataModel, 1.0)
{
}
/**
* <p>Constructs a {@link NearestNeighborClusterSimilarity} based on the given {@link UserCorrelation}.
* All user-user correlations are examined.</p>
*
* @param Correlation
*/
public NearestNeighborClusterSimilarity(UserCorrelation correlation)
: this(correlation, 1.0)
{
}
