ranking

package
v0.2.6 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Sep 26, 2021 License: Apache-2.0 Imports: 19 Imported by: 0

Documentation

Index

Constants

View Source 
const (
	CollaborativeBPR = "bpr"
	CollaborativeALS = "als"
	CollaborativeCCD = "ccd"
)

Variables

This section is empty.

Functions

func EncodeModel 

func EncodeModel(m Model) ([]byte, error)

func Evaluate 

func Evaluate(estimator MatrixFactorization, testSet, trainSet *DataSet, topK, numCandidates, nJobs int, scorers ...Metric) []float32

Evaluate evaluates a model in top-n tasks.

func GetModelName added in v0.2.5 

func GetModelName(m Model) string

func HR 

func HR(targetSet *i32set.Set, rankList []int32) float32

HR means Hit Ratio.

func LoadDataFromBuiltIn 

func LoadDataFromBuiltIn(dataSetName string) (*DataSet, *DataSet, error)

LoadDataFromBuiltIn loads a built-in Data set. Now support:

func MAP 

func MAP(targetSet *i32set.Set, rankList []int32) float32

MAP means Mean Average Precision. mAP: http://sdsawtelle.github.io/blog/output/mean-average-precision-MAP-for-recommender-systems.html

func MRR 

func MRR(targetSet *i32set.Set, rankList []int32) float32

MRR means Mean Reciprocal Rank.

The mean reciprocal rank is a statistic measure for evaluating any process that produces a list of possible responses to a sample of queries, ordered by probability of correctness. The reciprocal rank of a query response is the multiplicative inverse of the rank of the first correct answer: 1 for first place, ​1⁄2 for second place, ​1⁄3 for third place and so on. The mean reciprocal rank is the average of the reciprocal ranks of results for a sample of queries Q: 

MRR = \frac{1}{Q} \sum^{|Q|}_{i=1} \frac{1}{rank_i}

func NDCG 

func NDCG(targetSet *i32set.Set, rankList []int32) float32

NDCG means Normalized Discounted Cumulative Gain.

func Precision 

func Precision(targetSet *i32set.Set, rankList []int32) float32

Precision is the fraction of relevant ItemFeedback among the recommended ItemFeedback. 

\frac{|relevant documents| \cap |retrieved documents|} {|{retrieved documents}|}

func Rank 

func Rank(model MatrixFactorization, userId int32, candidates []int32, topN int) ([]int32, []float32)

func Recall 

func Recall(targetSet *i32set.Set, rankList []int32) float32

Recall is the fraction of relevant ItemFeedback that have been recommended over the total amount of relevant ItemFeedback. 

\frac{|relevant documents| \cap |retrieved documents|} {|{relevant documents}|}

Types

type ALS 

type ALS struct {
	BaseMatrixFactorization
	// Model parameters
	UserFactor *mat.Dense // p_u
	ItemFactor *mat.Dense // q_i
	// contains filtered or unexported fields
}

ALS [7] is the Weighted Regularized Matrix Factorization, which exploits unique properties of implicit feedback datasets. It treats the data as indication of positive and negative preference associated with vastly varying confidence levels. This leads to a factor model which is especially tailored for implicit feedback recommenders. Authors also proposed a scalable optimization procedure, which scales linearly with the data size. Hyper-parameters: 

NFactors   - The number of latent factors. Default is 10.
NEpochs    - The number of training epochs. Default is 50.
InitMean   - The mean of initial latent factors. Default is 0.
InitStdDev - The standard deviation of initial latent factors. Default is 0.1.
Reg        - The strength of regularization.

func NewALS 

func NewALS(params model.Params) *ALS

NewALS creates a ALS model.

func (*ALS) Clear 

func (als *ALS) Clear()

func (*ALS) Fit 

func (als *ALS) Fit(trainSet, valSet *DataSet, config *FitConfig) Score

Fit the ALS model.

func (*ALS) GetParamsGrid 

func (als *ALS) GetParamsGrid() model.ParamsGrid

func (*ALS) Init 

func (als *ALS) Init(trainSet *DataSet)

func (*ALS) InternalPredict 

func (als *ALS) InternalPredict(userIndex, itemIndex int32) float32

func (*ALS) Invalid added in v0.2.2 

func (als *ALS) Invalid() bool

func (*ALS) Predict 

func (als *ALS) Predict(userId, itemId string) float32

Predict by the ALS model.

func (*ALS) SetParams 

func (als *ALS) SetParams(params model.Params)

SetParams sets hyper-parameters for the ALS model.

type BPR 

type BPR struct {
	BaseMatrixFactorization
	// Model parameters
	UserFactor [][]float32 // p_u
	ItemFactor [][]float32 // q_i
	// contains filtered or unexported fields
}

BPR means Bayesian Personal Ranking, is a pairwise learning algorithm for matrix factorization model with implicit feedback. The pairwise ranking between item i and j for user u is estimated by: 

p(i >_u j) = \sigma( p_u^T (q_i - q_j) )

Hyper-parameters: 

 Reg 		- The regularization parameter of the cost function that is
			  optimized. Default is 0.01.
 Lr 		- The learning rate of SGD. Default is 0.05.
 nFactors	- The number of latent factors. Default is 10.
 NEpochs	- The number of iteration of the SGD procedure. Default is 100.
 InitMean	- The mean of initial random latent factors. Default is 0.
 InitStdDev	- The standard deviation of initial random latent factors. Default is 0.001.

func NewBPR 

func NewBPR(params model.Params) *BPR

NewBPR creates a BPR model.

func (*BPR) Clear 

func (bpr *BPR) Clear()

func (*BPR) Fit 

func (bpr *BPR) Fit(trainSet, valSet *DataSet, config *FitConfig) Score

Fit the BPR model.

func (*BPR) GetParamsGrid 

func (bpr *BPR) GetParamsGrid() model.ParamsGrid

func (*BPR) Init 

func (bpr *BPR) Init(trainSet *DataSet)

func (*BPR) InternalPredict 

func (bpr *BPR) InternalPredict(userIndex, itemIndex int32) float32

func (*BPR) Invalid added in v0.2.2 

func (bpr *BPR) Invalid() bool

func (*BPR) Predict 

func (bpr *BPR) Predict(userId, itemId string) float32

Predict by the BPR model.

func (*BPR) SetParams 

func (bpr *BPR) SetParams(params model.Params)

SetParams sets hyper-parameters of the BPR model.

type BaseMatrixFactorization 

type BaseMatrixFactorization struct {
	model.BaseModel
	UserIndex base.Index
	ItemIndex base.Index
}

func (*BaseMatrixFactorization) GetItemIndex 

func (model *BaseMatrixFactorization) GetItemIndex() base.Index

func (*BaseMatrixFactorization) GetUserIndex 

func (model *BaseMatrixFactorization) GetUserIndex() base.Index

func (*BaseMatrixFactorization) Init 

func (model *BaseMatrixFactorization) Init(trainSet *DataSet)

type CCD 

type CCD struct {
	BaseMatrixFactorization
	// Model parameters
	UserFactor [][]float32
	ItemFactor [][]float32
	// contains filtered or unexported fields
}

func NewCCD 

func NewCCD(params model.Params) *CCD

NewCCD creates a eALS model.

func (*CCD) Clear 

func (ccd *CCD) Clear()

func (*CCD) Fit 

func (ccd *CCD) Fit(trainSet, valSet *DataSet, config *FitConfig) Score

func (*CCD) GetParamsGrid 

func (ccd *CCD) GetParamsGrid() model.ParamsGrid

func (*CCD) Init 

func (ccd *CCD) Init(trainSet *DataSet)

func (*CCD) InternalPredict 

func (ccd *CCD) InternalPredict(userIndex, itemIndex int32) float32

func (*CCD) Invalid added in v0.2.2 

func (ccd *CCD) Invalid() bool

func (*CCD) Predict 

func (ccd *CCD) Predict(userId, itemId string) float32

Predict by the ALS model.

func (*CCD) SetParams 

func (ccd *CCD) SetParams(params model.Params)

SetParams sets hyper-parameters for the ALS model.

type DataSet 

type DataSet struct {
	UserIndex     base.Index
	ItemIndex     base.Index
	FeedbackUsers base.Integers
	FeedbackItems base.Integers
	UserFeedback  [][]int32
	ItemFeedback  [][]int32
	Negatives     [][]int32
	ItemLabels    [][]int32
	UserLabels    [][]int32
	// statistics
	NumItemLabels int32
	NumUserLabels int32
}

DataSet contains preprocessed data structures for recommendation models.

func LoadDataFromCSV 

func LoadDataFromCSV(fileName, sep string, hasHeader bool) *DataSet

LoadDataFromCSV loads Data from a CSV file. The CSV file should be: 

[optional header]
<userId 1> <sep> <itemId 1> <sep> <rating 1> <sep> <extras>
<userId 2> <sep> <itemId 2> <sep> <rating 2> <sep> <extras>
<userId 3> <sep> <itemId 3> <sep> <rating 3> <sep> <extras>
...

For example, the `u.Data` from MovieLens 100K is: 

196\t242\t3\t881250949
186\t302\t3\t891717742
22\t377\t1\t878887116

func NewDirectIndexDataset 

func NewDirectIndexDataset() *DataSet

func NewMapIndexDataset 

func NewMapIndexDataset() *DataSet

NewMapIndexDataset creates a data set.

func (*DataSet) AddFeedback 

func (dataset *DataSet) AddFeedback(userId, itemId string, insertUserItem bool)

func (*DataSet) AddItem 

func (dataset *DataSet) AddItem(itemId string)

func (*DataSet) AddUser 

func (dataset *DataSet) AddUser(userId string)

func (*DataSet) Count 

func (dataset *DataSet) Count() int

func (*DataSet) GetIndex 

func (dataset *DataSet) GetIndex(i int) (int32, int32)

GetIndex gets the i-th record by <user index, item index, rating>.

func (*DataSet) ItemCount 

func (dataset *DataSet) ItemCount() int

ItemCount returns the number of ItemFeedback.

func (*DataSet) NegativeSample 

func (dataset *DataSet) NegativeSample(excludeSet *DataSet, numCandidates int) [][]int32

func (*DataSet) SetNegatives 

func (dataset *DataSet) SetNegatives(userId string, negatives []string)

func (*DataSet) Split 

func (dataset *DataSet) Split(numTestUsers int, seed int64) (*DataSet, *DataSet)

Split dataset by user-leave-one-out method. The argument `numTestUsers` determines the number of users in the test set. If numTestUsers is equal or greater than the number of total users or numTestUsers <= 0, all users are presented in the test set.

func (*DataSet) UserCount 

func (dataset *DataSet) UserCount() int

UserCount returns the number of UserFeedback.

type FitConfig 

type FitConfig struct {
	Jobs       int
	Verbose    int
	Candidates int
	TopK       int
	Tracker    model.Tracker
}

func NewFitConfig added in v0.2.2 

func NewFitConfig() *FitConfig

func (*FitConfig) LoadDefaultIfNil 

func (config *FitConfig) LoadDefaultIfNil() *FitConfig

func (*FitConfig) SetJobs added in v0.2.2 

func (config *FitConfig) SetJobs(nJobs int) *FitConfig

func (*FitConfig) SetTracker added in v0.2.4 

func (config *FitConfig) SetTracker(tracker model.Tracker) *FitConfig

func (*FitConfig) SetVerbose added in v0.2.5 

func (config *FitConfig) SetVerbose(verbose int) *FitConfig

type MatrixFactorization 

type MatrixFactorization interface {
	Model
	// Predict the rating given by a user (userId) to a item (itemId).
	Predict(userId, itemId string) float32
	// InternalPredict predicts rating given by a user index and a item index
	InternalPredict(userIndex, itemIndex int32) float32
	// GetUserIndex returns user index.
	GetUserIndex() base.Index
}

func DecodeModel 

func DecodeModel(buf []byte) (MatrixFactorization, error)

type Metric 

type Metric func(targetSet *i32set.Set, rankList []int32) float32

Metric is used by evaluators in personalized ranking tasks.

type Model 

type Model interface {
	model.Model
	// Fit a model with a train set and parameters.
	Fit(trainSet *DataSet, validateSet *DataSet, config *FitConfig) Score
	// GetItemIndex returns item index.
	GetItemIndex() base.Index
}

func Clone 

func Clone(m Model) Model

Clone a model with deep copy.

func NewModel 

func NewModel(name string, params model.Params) (Model, error)

type ModelSearcher 

type ModelSearcher struct {
	// contains filtered or unexported fields
}

ModelSearcher is a thread-safe personal ranking model searcher.

func NewModelSearcher 

func NewModelSearcher(nEpoch, nTrials, nJobs int) *ModelSearcher

NewModelSearcher creates a thread-safe personal ranking model searcher.

func (*ModelSearcher) Fit 

func (searcher *ModelSearcher) Fit(trainSet, valSet *DataSet, tracker model.Tracker, runner model.Runner) error

func (*ModelSearcher) GetBestModel 

func (searcher *ModelSearcher) GetBestModel() (string, Model, Score)

GetBestModel returns the optimal personal ranking model.

type ParamsSearchResult 

type ParamsSearchResult struct {
	BestModel  Model
	BestScore  Score
	BestParams model.Params
	BestIndex  int
	Scores     []Score
	Params     []model.Params
}

ParamsSearchResult contains the return of grid search.

func GridSearchCV 

func GridSearchCV(estimator MatrixFactorization, trainSet *DataSet, testSet *DataSet, paramGrid model.ParamsGrid,
	_ int64, fitConfig *FitConfig, runner model.Runner) ParamsSearchResult

GridSearchCV finds the best parameters for a model.

func RandomSearchCV 

func RandomSearchCV(estimator MatrixFactorization, trainSet *DataSet, testSet *DataSet, paramGrid model.ParamsGrid,
	numTrials int, seed int64, fitConfig *FitConfig, runner model.Runner) ParamsSearchResult

RandomSearchCV searches hyper-parameters by random.

func (*ParamsSearchResult) AddScore 

func (r *ParamsSearchResult) AddScore(params model.Params, score Score)

type Score 

type Score struct {
	NDCG      float32
	Precision float32
	Recall    float32
}

type SnapshotManger 

type SnapshotManger struct {
	BestWeights []interface{}
	BestScore   Score
}

SnapshotManger manages the best snapshot.

func (*SnapshotManger) AddSnapshot 

func (sm *SnapshotManger) AddSnapshot(score Score, weights ...interface{})

AddSnapshot adds a copied snapshot.

func (*SnapshotManger) AddSnapshotNoCopy 

func (sm *SnapshotManger) AddSnapshotNoCopy(score Score, weights ...interface{})

AddSnapshotNoCopy adds a snapshot without copy.

Source Files

View all Source files

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.