MyMediaLite
3.10
|
Matrix factorization model for item prediction (ranking) optimized for BPR More...
Public Member Functions | |
override void | AddFeedback (ICollection< Tuple< int, int >> feedback) |
Add positive feedback events and perform incremental training | |
BPRMF () | |
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 current optimization objective (usually loss plus regularization term) of the model | |
override void | Iterate () |
Perform one iteration of stochastic gradient ascent over the training data | |
override void | LoadModel (string file) |
Get the model parameters from a file | |
override float | Predict (int user_id, int item_id) |
Predict the weight for a given user-item combination | |
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 | RemoveFeedback (ICollection< Tuple< int, int >> feedback) |
Remove all feedback events by the given user-item combinations | |
override void | RemoveItem (int item_id) |
Remove all feedback by one item | |
override void | RemoveUser (int user_id) |
Remove all feedback by one user | |
override void | SaveModel (string file) |
Save the model parameters to a file | |
IList< Tuple< int, float > > | ScoreItems (IList< int > accessed_items, IList< int > candidate_items) |
Score a list of items given a list of items 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 |
Protected Member Functions | |
override void | AddItem (int item_id) |
override void | AddUser (int user_id) |
override void | InitModel () |
virtual void | IterateWithoutReplacementUniformPair () |
Iterate over the training data, uniformly sample from user-item pairs without replacement. | |
virtual void | IterateWithoutReplacementUniformPair (IList< int > indices) |
Iterate over the training data, uniformly sample from user-item pairs without replacement. | |
virtual void | IterateWithoutReplacementUniformUser () |
Iterate over the training data, uniformly sample from users without replacement. | |
virtual void | IterateWithReplacementUniformPair () |
Iterate over the training data, uniformly sample from user-item pairs with replacement. | |
virtual void | IterateWithReplacementUniformUser () |
Iterate over the training data, uniformly sample from users with replacement. | |
override void | RetrainItem (int item_id) |
Retrain the latent factors of a given item | |
override void | RetrainUser (int user_id) |
Retrain the latent factors of a given user | |
virtual void | SampleItemPair (ICollection< int > user_items, out int item_id, out int other_item_id) |
Sample a pair of items, given a user | |
virtual bool | SampleOtherItem (int user_id, int item_id, out int other_item_id) |
Sample another item, given the first one and the user | |
virtual void | SampleTriple (out int user_id, out int item_id, out int other_item_id) |
Sample a triple for BPR learning | |
virtual int | SampleUser () |
Uniformly sample a user that has viewed at least one and not all items | |
virtual void | UpdateFactors (int user_id, int item_id, int other_item_id, bool update_u, bool update_i, bool update_j) |
Update latent factors according to the stochastic gradient descent update rule |
Protected Attributes | |
float[] | item_bias |
Item bias terms | |
Matrix< float > | item_factors |
Latent item factor matrix | |
float | learn_rate = 0.05f |
Learning rate alpha | |
int | num_factors = 10 |
Number of latent factors per user/item | |
float | reg_i = 0.0025f |
Regularization parameter for positive item factors | |
float | reg_j = 0.00025f |
Regularization parameter for negative item factors | |
float | reg_u = 0.0025f |
Regularization parameter for user factors | |
bool | update_j = true |
If set (default), update factors for negative sampled items during learning | |
Matrix< float > | user_factors |
Latent user factor matrix |
Static Protected Attributes | |
static System.Random | random |
Reference to (per-thread) singleton random number generator |
Properties | |
float | BiasReg [get, set] |
Regularization parameter for the bias term | |
virtual IPosOnlyFeedback | Feedback [get, set] |
the feedback data to be used for training | |
double | InitMean [get, set] |
Mean of the normal distribution used to initialize the latent factors | |
double | InitStdDev [get, set] |
Standard deviation of the normal distribution used to initialize the latent factors | |
float | LearnRate [get, set] |
Learning rate alpha | |
int | MaxItemID [get, set] |
Maximum item ID | |
int | MaxUserID [get, set] |
Maximum user ID | |
uint | NumFactors [get, set] |
Number of latent factors per user/item | |
uint | NumIter [get, set] |
Number of iterations over the training data | |
float | RegI [get, set] |
Regularization parameter for positive item factors | |
float | RegJ [get, set] |
Regularization parameter for negative item factors | |
float | RegU [get, set] |
Regularization parameter for user factors | |
bool | UniformUserSampling [get, set] |
Sample uniformly from users | |
bool | UpdateItems [get, set] |
bool | UpdateJ [get, set] |
If set (default), update factors for negative sampled items during learning | |
bool | UpdateUsers [get, set] |
bool | WithReplacement [get, set] |
Sample positive observations with (true) or without (false) replacement |
Matrix factorization model for item prediction (ranking) optimized for BPR
BPR reduces ranking to pairwise classification. The different variants (settings) of this recommender roughly optimize the area under the ROC curve (AUC).
where and
.
represents the parameters of the model and
is a regularization constant.
is the logistic function.
In this implementation, we distinguish different regularization updates for users and positive and negative items, which means we do not have only one regularization constant. The optimization problem specified above thus is only an approximation.
Literature:
Different sampling strategies are configurable by setting the UniformUserSampling and WithReplacement accordingly. To get the strategy from the original paper, set UniformUserSampling=false and WithReplacement=false. WithReplacement=true (default) gives you usually a slightly faster convergence, and UniformUserSampling=true (default) (approximately) optimizes the average AUC over all users.
This recommender supports incremental updates.
|
inline |
Default constructor
|
inlinevirtualinherited |
Add positive feedback events and perform incremental training
feedback | collection of user id - item id tuples |
Reimplemented from IncrementalItemRecommender.
|
inlinevirtualinherited |
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.
user_id | the user ID |
item_id | the item ID |
Implements IRecommender.
Reimplemented in ExternalItemRecommender, ExternalRatingPredictor, BiPolarSlopeOne, SlopeOne, Constant, GlobalAverage, UserAverage, ItemAverage, and Random.
|
inlineinherited |
create a shallow copy of the object
|
inlinevirtual |
Compute the current optimization objective (usually loss plus regularization term) of the model
Implements MF.
Reimplemented in SoftMarginRankingMF.
|
inlinevirtual |
Perform one iteration of stochastic gradient ascent over the training data
One iteration is samples number of positive entries in the training matrix times
Implements MF.
|
inlineprotectedvirtual |
Iterate over the training data, uniformly sample from user-item pairs without replacement.
Reimplemented in MultiCoreBPRMF.
|
inlineprotectedvirtual |
Iterate over the training data, uniformly sample from user-item pairs without replacement.
|
inlineprotectedvirtual |
Iterate over the training data, uniformly sample from users without replacement.
|
inlineprotectedvirtual |
Iterate over the training data, uniformly sample from user-item pairs with replacement.
|
inlineprotectedvirtual |
Iterate over the training data, uniformly sample from users with replacement.
|
inlinevirtual |
Get the model parameters from a file
filename | the name of the file to read from |
Reimplemented from MF.
|
inlinevirtual |
Predict the weight for a given user-item combination
If the user or the item are not known to the recommender, zero is returned. To avoid this behavior for unknown entities, use CanPredict() to check before.
user_id | the user ID |
item_id | the item ID |
Reimplemented from MF.
|
inherited |
Recommend items for a given user
user_id | the user ID |
n | the number of items to recommend, -1 for as many as possible |
ignore_items | collection if items that should not be returned; if null, use empty collection |
candidate_items | the candidate items to choose from; if null, use all items |
Implemented in WeightedEnsemble, and Ensemble.
|
inlinevirtualinherited |
Remove all feedback events by the given user-item combinations
feedback | collection of user id - item id tuples |
Reimplemented from IncrementalItemRecommender.
|
inlinevirtual |
|
inlinevirtualinherited |
Remove all feedback by one user
user_id | the user ID |
Reimplemented from IncrementalItemRecommender.
|
inlineprotectedvirtual |
|
inlineprotectedvirtual |
|
inlineprotectedvirtual |
Sample a pair of items, given a user
user_items | the items accessed by the given user |
item_id | the ID of the first item |
other_item_id | the ID of the second item |
|
inlineprotectedvirtual |
Sample another item, given the first one and the user
user_id | the user ID |
item_id | the ID of the given item |
other_item_id | the ID of the other item |
|
inlineprotectedvirtual |
Sample a triple for BPR learning
user_id | the user ID |
item_id | the ID of the first item |
other_item_id | the ID of the second item |
Reimplemented in WeightedBPRMF.
|
inlineprotectedvirtual |
Uniformly sample a user that has viewed at least one and not all items
|
inlinevirtual |
Save the model parameters to a file
filename | the name of the file to write to |
Reimplemented from MF.
|
inline |
Score a list of items given a list of items that represent a new user
accessed_items | the ratings (item IDs and rating values) representing the new user |
candidate_items | the items to be rated |
Implements IFoldInItemRecommender.
|
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.
Implements IRecommender.
Reimplemented in SoftMarginRankingMF, WeightedBPRMF, and MultiCoreBPRMF.
|
inlinevirtual |
Learn the model parameters of the recommender from the training data
Reimplemented from MF.
Reimplemented in MultiCoreBPRMF, and WeightedBPRMF.
|
inlineprotectedvirtual |
Update latent factors according to the stochastic gradient descent update rule
user_id | the user ID |
item_id | the ID of the first item |
other_item_id | the ID of the second item |
update_u | if true, update the user latent factors |
update_i | if true, update the latent factors of the first item |
update_j | if true, update the latent factors of the second item |
Reimplemented in SoftMarginRankingMF.
|
protected |
Item bias terms
|
protectedinherited |
Latent item factor matrix
|
protected |
Learning rate alpha
|
protectedinherited |
Number of latent factors per user/item
|
staticprotected |
Reference to (per-thread) singleton random number generator
|
protected |
Regularization parameter for positive item factors
|
protected |
Regularization parameter for negative item factors
|
protected |
Regularization parameter for user factors
|
protected |
If set (default), update factors for negative sampled items during learning
|
protectedinherited |
Latent user factor matrix
|
getset |
Regularization parameter for the bias term
|
getsetinherited |
the feedback data to be used for training
|
getsetinherited |
Mean of the normal distribution used to initialize the latent factors
|
getsetinherited |
Standard deviation of the normal distribution used to initialize the latent factors
|
getset |
Learning rate alpha
|
getsetinherited |
Maximum item ID
|
getsetinherited |
Maximum user ID
|
getsetinherited |
Number of latent factors per user/item
|
getsetinherited |
Number of iterations over the training data
|
getset |
Regularization parameter for positive item factors
|
getset |
Regularization parameter for negative item factors
|
getset |
Regularization parameter for user factors
|
getset |
Sample uniformly from users
|
getset |
If set (default), update factors for negative sampled items during learning
|
getset |
Sample positive observations with (true) or without (false) replacement