MyMediaLite  3.09
Public Member Functions | Protected Member Functions | Protected Attributes | Properties
BiasedMatrixFactorization Class Reference

Matrix factorization with explicit user and item bias, learning is performed by stochastic gradient descent. More...

Inheritance diagram for BiasedMatrixFactorization:
MatrixFactorization IncrementalRatingPredictor IIterativeModel IFoldInRatingPredictor RatingPredictor IIncrementalRatingPredictor IRatingPredictor Recommender IRatingPredictor IRatingPredictor IIncrementalRecommender IRecommender IRecommender IRecommender IRecommender SigmoidCombinedAsymmetricFactorModel SigmoidItemAsymmetricFactorModel SigmoidUserAsymmetricFactorModel SocialMF

List of all members.

Public Member Functions

override void AddRatings (IRatings ratings)
 Add new ratings and perform incremental training.
 BiasedMatrixFactorization ()
 Default constructor.
virtual bool CanPredict (int user_id, int item_id)
 Check whether a useful prediction (i.e. not using a fallback/default answer) can be made for a given user-item combination.
Object Clone ()
 create a shallow copy of the object
override float ComputeObjective ()
 Compute the regularized loss.
override void Iterate ()
 Run one iteration (= pass over the training data)
override void LoadModel (string filename)
 Get the model parameters from a file.
override float Predict (int user_id, int item_id)
 Predict the rating of a given user for a given item.
IList< Tuple< int, float > > Recommend (int user_id, int n=-1, ICollection< int > ignore_items=null, ICollection< int > candidate_items=null)
 Recommend items for a given user.
virtual
System.Collections.Generic.IList
< Tuple< int, float > > 
Recommend (int user_id, int n=-1, System.Collections.Generic.ICollection< int > ignore_items=null, System.Collections.Generic.ICollection< int > candidate_items=null)
override void RemoveItem (int item_id)
 Remove all feedback by one item.
override void RemoveRatings (IDataSet ratings)
 Remove existing ratings and perform "incremental" training.
override void RemoveUser (int user_id)
 Remove all feedback by one user.
override void RetrainItem (int item_id)
 Updates the latent factors of an item.
override void RetrainUser (int user_id)
 Updates the latent factors on a user.
override void SaveModel (string filename)
 Save the model parameters to a file.
IList< Tuple< int, float > > ScoreItems (IList< Tuple< int, float >> rated_items, IList< int > candidate_items)
 Rate a list of items given a list of ratings that represent a new user.
override string ToString ()
 Return a string representation of the recommender.
override void Train ()
 Learn the model parameters of the recommender from the training data.
override void UpdateRatings (IRatings ratings)
 Update existing ratings and perform incremental training.

Protected Member Functions

override void AddItem (int item_id)
override void AddUser (int user_id)
double ComputeLoss ()
 Computes the value of the loss function that is currently being optimized.
override float[] FoldIn (IList< Tuple< int, float >> rated_items)
 Compute parameters (latent factors) for a user represented by ratings.
internal override void InitModel ()
 Initialize the model data structure.
override void Iterate (IList< int > rating_indices, bool update_user, bool update_item)
 Iterate once over rating data and adjust corresponding factors (stochastic gradient descent)
float Predict (int user_id, int item_id, bool bound)
override float Predict (float[] user_vector, int item_id)
 Predict rating for a fold-in user and an item.
void SetupLoss ()
 Set up the common part of the error gradient of the loss function to optimize.
override void UpdateLearnRate ()
 Updates current_learnrate after each epoch.

Protected Attributes

Func< double, double, float > compute_gradient_common
 delegate to compute the common term of the error gradient
internal float current_learnrate
 The learn rate used for the current epoch.
const int FOLD_IN_BIAS_INDEX = 0
 Index of the bias term in the user vector representation for fold-in.
const int FOLD_IN_FACTORS_START = 1
 Start index of the user factors in the user vector representation for fold-in.
float global_bias
 The bias (global average)
internal float[] item_bias
 the item biases
internal Matrix< float > item_factors
 Matrix containing the latent item factors.
double last_loss = double.NegativeInfinity
 Loss for the last iteration, used by bold driver heuristics.
float max_rating
 Maximum rating value.
float min_rating
 Minimum rating value.
float rating_range_size
 size of the interval of valid ratings
IRatings ratings
 rating data
internal float[] user_bias
 the user biases
internal Matrix< float > user_factors
 Matrix containing the latent user factors.

Properties

float BiasLearnRate [get, set]
 Learn rate factor for the bias terms.
float BiasReg [get, set]
 regularization factor for the bias terms
bool BoldDriver [get, set]
 Use bold driver heuristics for learning rate adaption.
float Decay [get, set]
 Multiplicative learn rate decay.
bool FrequencyRegularization [get, set]
 Regularization based on rating frequency.
double InitMean [get, set]
 Mean of the normal distribution used to initialize the factors.
double InitStdDev [get, set]
 Standard deviation of the normal distribution used to initialize the factors.
float LearnRate [get, set]
 Learn rate (update step size)
OptimizationTarget Loss [get, set]
 The optimization target.
int MaxItemID [get, set]
 Maximum item ID.
virtual float MaxRating [get, set]
 Maximum rating value.
int MaxThreads [get, set]
 the maximum number of threads to use
int MaxUserID [get, set]
 Maximum user ID.
virtual float MinRating [get, set]
 Minimum rating value.
bool NaiveParallelization [get, set]
 Use 'naive' parallelization strategy instead of conflict-free 'distributed' SGD.
uint NumFactors [get, set]
 Number of latent factors.
uint NumIter [get, set]
 Number of iterations over the training data.
virtual IRatings Ratings [get, set]
 The rating data.
float RegI [get, set]
 regularization constant for the item factors
float RegU [get, set]
 regularization constant for the user factors
override float Regularization [set]
 Regularization parameter.
bool UpdateItems [get, set]
 true if items shall be updated when doing incremental updates
bool UpdateUsers [get, set]
 true if users shall be updated when doing incremental updates

Detailed Description

Matrix factorization with explicit user and item bias, learning is performed by stochastic gradient descent.

Per default optimizes for RMSE. Alternatively, you can set the Loss property to MAE or LogisticLoss. If set to log likelihood and with binary ratings, the recommender implements a simple version Menon and Elkan's LFL model, which predicts binary labels, has no advanced regularization, and uses no side information.

This recommender makes use of multi-core machines if requested. Just set MaxThreads to a large enough number (usually multiples of the number of available cores). The parallelization is based on ideas presented in the paper by Gemulla et al.

Literature:

This recommender supports incremental updates. See the paper by Rendle and Schmidt-Thieme.


Constructor & Destructor Documentation

Default constructor.


Member Function Documentation

override void AddRatings ( IRatings  ratings) [inline, virtual, inherited]

Add new ratings and perform incremental training.

Parameters:
ratingsthe ratings

Reimplemented from IncrementalRatingPredictor.

virtual bool CanPredict ( int  user_id,
int  item_id 
) [inline, virtual, inherited]

Check whether a useful prediction (i.e. not using a fallback/default answer) can be made for a given user-item combination.

It is up to the recommender implementor to decide when a prediction is useful, and to document it accordingly.

Parameters:
user_idthe user ID
item_idthe item ID
Returns:
true if a useful prediction can be made, false otherwise

Implements IRecommender.

Reimplemented in ExternalItemRecommender, ExternalRatingPredictor, BiPolarSlopeOne, SlopeOne, Constant, GlobalAverage, UserAverage, ItemAverage, and Random.

Object Clone ( ) [inline, inherited]

create a shallow copy of the object

double ComputeLoss ( ) [inline, protected]

Computes the value of the loss function that is currently being optimized.

Returns:
the loss
override float ComputeObjective ( ) [inline, virtual]

Compute the regularized loss.

Returns:
the regularized loss

Reimplemented from MatrixFactorization.

Reimplemented in SigmoidCombinedAsymmetricFactorModel, SigmoidItemAsymmetricFactorModel, SigmoidUserAsymmetricFactorModel, and SocialMF.

override float [] FoldIn ( IList< Tuple< int, float >>  rated_items) [inline, protected, virtual]

Compute parameters (latent factors) for a user represented by ratings.

Returns:
a vector of latent factors
Parameters:
rated_itemsa list of (item ID, rating value) pairs

Reimplemented from MatrixFactorization.

Reimplemented in SigmoidCombinedAsymmetricFactorModel, SigmoidUserAsymmetricFactorModel, and SigmoidItemAsymmetricFactorModel.

internal override void InitModel ( ) [inline, protected, virtual]

Initialize the model data structure.

Reimplemented from MatrixFactorization.

Reimplemented in SigmoidCombinedAsymmetricFactorModel, SigmoidItemAsymmetricFactorModel, SigmoidUserAsymmetricFactorModel, and SocialMF.

override void Iterate ( ) [inline, virtual]

Run one iteration (= pass over the training data)

Reimplemented from MatrixFactorization.

override void Iterate ( IList< int >  rating_indices,
bool  update_user,
bool  update_item 
) [inline, protected, virtual]

Iterate once over rating data and adjust corresponding factors (stochastic gradient descent)

Parameters:
rating_indicesa list of indices pointing to the ratings to iterate over
update_usertrue if user factors to be updated
update_itemtrue if item factors to be updated

Reimplemented from MatrixFactorization.

Reimplemented in SigmoidCombinedAsymmetricFactorModel, SigmoidItemAsymmetricFactorModel, SigmoidUserAsymmetricFactorModel, and SocialMF.

override void LoadModel ( string  filename) [inline]

Get the model parameters from a file.

Parameters:
filenamethe name of the file to read from

Reimplemented from MatrixFactorization.

Reimplemented in SigmoidCombinedAsymmetricFactorModel, SigmoidItemAsymmetricFactorModel, and SigmoidUserAsymmetricFactorModel.

override float Predict ( int  user_id,
int  item_id 
) [inline]

Predict the rating of a given user for a given item.

If the user or the item are not known to the recommender, the global average is returned. To avoid this behavior for unknown entities, use CanPredict() to check before.

Parameters:
user_idthe user ID
item_idthe item ID
Returns:
the predicted rating

Reimplemented from MatrixFactorization.

Reimplemented in SigmoidCombinedAsymmetricFactorModel, SigmoidItemAsymmetricFactorModel, and SigmoidUserAsymmetricFactorModel.

override float Predict ( float[]  user_vector,
int  item_id 
) [inline, protected, virtual]

Predict rating for a fold-in user and an item.

Parameters:
user_vectora float vector representing the user
item_idthe item ID
Returns:
the predicted rating

Reimplemented from MatrixFactorization.

Reimplemented in SigmoidCombinedAsymmetricFactorModel.

IList<Tuple<int, float> > Recommend ( int  user_id,
int  n = -1,
ICollection< int >  ignore_items = null,
ICollection< int >  candidate_items = null 
) [inherited]

Recommend items for a given user.

Parameters:
user_idthe user ID
nthe number of items to recommend, -1 for as many as possible
ignore_itemscollection if items that should not be returned; if null, use empty collection
candidate_itemsthe candidate items to choose from; if null, use all items
Returns:
a sorted list of (item_id, score) tuples

Implemented in WeightedEnsemble, and Ensemble.

override void RemoveItem ( int  item_id) [inline, virtual]

Remove all feedback by one item.

Parameters:
item_idthe item ID

Reimplemented from MatrixFactorization.

override void RemoveRatings ( IDataSet  ratings) [inline, virtual, inherited]

Remove existing ratings and perform "incremental" training.

Parameters:
ratingsthe user and item IDs of the ratings to be removed

Reimplemented from IncrementalRatingPredictor.

override void RemoveUser ( int  user_id) [inline, virtual]

Remove all feedback by one user.

Parameters:
user_idthe user ID

Reimplemented from MatrixFactorization.

override void RetrainItem ( int  item_id) [inline, virtual]

Updates the latent factors of an item.

Parameters:
item_idthe item ID

Reimplemented from MatrixFactorization.

override void RetrainUser ( int  user_id) [inline, virtual]

Updates the latent factors on a user.

Parameters:
user_idthe user ID

Reimplemented from MatrixFactorization.

override void SaveModel ( string  filename) [inline]

Save the model parameters to a file.

Parameters:
filenamethe name of the file to write to

Reimplemented from MatrixFactorization.

Reimplemented in SigmoidCombinedAsymmetricFactorModel, SigmoidItemAsymmetricFactorModel, and SigmoidUserAsymmetricFactorModel.

IList<Tuple<int, float> > ScoreItems ( IList< Tuple< int, float >>  rated_items,
IList< int >  candidate_items 
) [inline, inherited]

Rate a list of items given a list of ratings that represent a new user.

Returns:
a list of int and float pairs, representing item IDs and predicted ratings
Parameters:
rated_itemsthe ratings (item IDs and rating values) representing the new user
candidate_itemsthe items to be rated

Implements IFoldInRatingPredictor.

void SetupLoss ( ) [inline, protected]

Set up the common part of the error gradient of the loss function to optimize.

override string ToString ( ) [inline]

Return a string representation of the recommender.

The ToString() method of recommenders should list the class name and all hyperparameters, separated by space characters.

Reimplemented from MatrixFactorization.

Reimplemented in SigmoidCombinedAsymmetricFactorModel, SigmoidItemAsymmetricFactorModel, SigmoidUserAsymmetricFactorModel, and SocialMF.

override void UpdateLearnRate ( ) [inline, protected, virtual]

Updates current_learnrate after each epoch.

Reimplemented from MatrixFactorization.

override void UpdateRatings ( IRatings  ratings) [inline, virtual, inherited]

Update existing ratings and perform incremental training.

Parameters:
ratingsthe ratings

Reimplemented from IncrementalRatingPredictor.


Member Data Documentation

Func<double, double, float> compute_gradient_common [protected]

delegate to compute the common term of the error gradient

internal float current_learnrate [protected, inherited]

The learn rate used for the current epoch.

const int FOLD_IN_BIAS_INDEX = 0 [protected]

Index of the bias term in the user vector representation for fold-in.

const int FOLD_IN_FACTORS_START = 1 [protected]

Start index of the user factors in the user vector representation for fold-in.

float global_bias [protected, inherited]

The bias (global average)

internal float [] item_bias [protected]

the item biases

internal Matrix<float> item_factors [protected, inherited]

Matrix containing the latent item factors.

double last_loss = double.NegativeInfinity [protected]

Loss for the last iteration, used by bold driver heuristics.

float max_rating [protected, inherited]

Maximum rating value.

float min_rating [protected, inherited]

Minimum rating value.

float rating_range_size [protected]

size of the interval of valid ratings

IRatings ratings [protected, inherited]

rating data

internal float [] user_bias [protected]

the user biases

internal Matrix<float> user_factors [protected, inherited]

Matrix containing the latent user factors.


Property Documentation

float BiasLearnRate [get, set]

Learn rate factor for the bias terms.

float BiasReg [get, set]

regularization factor for the bias terms

bool BoldDriver [get, set]

Use bold driver heuristics for learning rate adaption.

Literature:

float Decay [get, set, inherited]

Multiplicative learn rate decay.

Applied after each epoch (= pass over the whole dataset)

bool FrequencyRegularization [get, set]

Regularization based on rating frequency.

Regularization proportional to the inverse of the square root of the number of ratings associated with the user or item. As described in the paper by Menon and Elkan.

double InitMean [get, set, inherited]

Mean of the normal distribution used to initialize the factors.

double InitStdDev [get, set, inherited]

Standard deviation of the normal distribution used to initialize the factors.

float LearnRate [get, set, inherited]

Learn rate (update step size)

The optimization target.

int MaxItemID [get, set, inherited]

Maximum item ID.

virtual float MaxRating [get, set, inherited]

Maximum rating value.

Implements IRatingPredictor.

int MaxThreads [get, set]

the maximum number of threads to use

For parallel learning, set this number to a multiple of the number of available cores/CPUs

int MaxUserID [get, set, inherited]

Maximum user ID.

virtual float MinRating [get, set, inherited]

Minimum rating value.

Implements IRatingPredictor.

bool NaiveParallelization [get, set]

Use 'naive' parallelization strategy instead of conflict-free 'distributed' SGD.

The exact sequence of updates depends on the thread scheduling. If you want reproducible results, e.g. when setting --random-seed=N, do NOT set this property.

uint NumFactors [get, set, inherited]

Number of latent factors.

uint NumIter [get, set, inherited]

Number of iterations over the training data.

Implements IIterativeModel.

virtual IRatings Ratings [get, set, inherited]

The rating data.

Implements IRatingPredictor.

Reimplemented in KNN, FactorWiseMatrixFactorization, TimeAwareRatingPredictor, ItemKNN, and UserKNN.

float RegI [get, set]

regularization constant for the item factors

float RegU [get, set]

regularization constant for the user factors

override float Regularization [set]

Regularization parameter.

Reimplemented from MatrixFactorization.

bool UpdateItems [get, set, inherited]

true if items shall be updated when doing incremental updates

Set to false if you do not want any updates to the item model parameters when doing incremental updates.

Implements IIncrementalRecommender.

bool UpdateUsers [get, set, inherited]

true if users shall be updated when doing incremental updates

Default should be true. Set to false if you do not want any updates to the user model parameters when doing incremental updates.

Implements IIncrementalRecommender.


The documentation for this class was generated from the following file: