ctr

package
v0.5.10 Latest Latest
Warning

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

 Go to latest Published: Jun 19, 2026 License: Apache-2.0 Imports: 30 Imported by: 0

Documentation

Index

Constants

This section is empty.

Variables

This section is empty.

Functions

func AUC 

func AUC(posPrediction, negPrediction []float32) float32

func Accuracy 

func Accuracy(posPrediction, negPrediction []float32) float32

func LoadLibFMFile 

func LoadLibFMFile(path string) (features [][]lo.Tuple2[int32, float32], targets []float32, maxLabel int32, err error)

LoadLibFMFile loads libFM format file.

func MarshalModel 

func MarshalModel(w io.Writer, m FactorizationMachines) error

func Precision 

func Precision(posPrediction, negPrediction []float32) float32

func Recall 

func Recall(posPrediction, _ []float32) float32

Types

type AFM added in v0.5.2 

type AFM struct {
	BaseFactorizationMachines

	// parameters
	B *nn.Tensor
	W nn.Layer
	V nn.Layer
	A []nn.Layer
	E []nn.Layer

	// numerical feature scalers: feature_index -> AutoScaler
	Scalers map[int32]*AutoScaler
	// contains filtered or unexported fields
}

func NewAFM added in v0.5.2 

func NewAFM(params model.Params) *AFM

func (*AFM) BatchInternalPredict added in v0.5.2 

func (fm *AFM) BatchInternalPredict(x []lo.Tuple2[[]int32, []float32], e [][][]uint16, jobs int) []float32

func (*AFM) BatchPredict added in v0.5.2 

func (fm *AFM) BatchPredict(inputs []lo.Tuple4[string, string, []Label, []Label], embeddings [][]Embedding, jobs int) []float32

func (*AFM) Clear added in v0.5.2 

func (fm *AFM) Clear()

func (*AFM) Fit added in v0.5.2 

func (fm *AFM) Fit(ctx context.Context, trainSet, testSet dataset.CTRSplit, config *FitConfig) Score

func (*AFM) Forward added in v0.5.2 

func (fm *AFM) Forward(indices, values *nn.Tensor, embeddings []*nn.Tensor, jobs int) *nn.Tensor

func (*AFM) Init added in v0.5.2 

func (fm *AFM) Init(trainSet dataset.CTRSplit)

func (*AFM) InternalPredict added in v0.5.2 

func (fm *AFM) InternalPredict(_ []int32, _ []float32) float32

func (*AFM) Invalid added in v0.5.2 

func (fm *AFM) Invalid() bool

func (*AFM) Marshal added in v0.5.2 

func (fm *AFM) Marshal(w io.Writer) error

func (*AFM) Parameters added in v0.5.2 

func (fm *AFM) Parameters() []*nn.Tensor

func (*AFM) Predict added in v0.5.2 

func (fm *AFM) Predict(_, _ string, _, _ []Label) float32

func (*AFM) SetParams added in v0.5.2 

func (fm *AFM) SetParams(params model.Params)

func (*AFM) SuggestParams added in v0.5.2 

func (fm *AFM) SuggestParams(trial goptuna.Trial) model.Params

func (*AFM) Unmarshal added in v0.5.2 

func (fm *AFM) Unmarshal(r io.Reader) error

type AutoScaler added in v0.5.7 

type AutoScaler struct {
	UseLog bool         // true if log1p preprocessing is applied, false if RobustScaler is used
	MinMax MinMaxScaler // for non-negative values (after log1p if UseLog)
	Robust RobustScaler // for data with negative values
}

AutoScaler automatically selects the appropriate scaling method based on data distribution. - If data contains negative values: uses RobustScaler - If all values are non-negative: applies log1p then MinMaxScaler

func NewAutoScaler added in v0.5.7 

func NewAutoScaler() *AutoScaler

NewAutoScaler creates an AutoScaler.

func (*AutoScaler) Fit added in v0.5.7 

func (s *AutoScaler) Fit(values []float32)

Fit analyzes the data and selects the appropriate scaling method.

func (*AutoScaler) Marshal added in v0.5.7 

func (s *AutoScaler) Marshal(w io.Writer) error

Marshal writes the scaler to a writer.

func (*AutoScaler) Transform added in v0.5.7 

func (s *AutoScaler) Transform(value float32) float32

Transform scales a value using the selected method.

func (*AutoScaler) Unmarshal added in v0.5.7 

func (s *AutoScaler) Unmarshal(r io.Reader) error

Unmarshal reads the scaler from a reader.

type BaseFactorizationMachines 

type BaseFactorizationMachines struct {
	model.BaseModel
	Index dataset.UnifiedIndex
}

func (*BaseFactorizationMachines) Init 

func (b *BaseFactorizationMachines) Init(trainSet dataset.CTRSplit)

type BatchInference 

type BatchInference interface {
	BatchPredict(inputs []lo.Tuple4[string, string, []Label, []Label], e [][]Embedding, jobs int) []float32
	BatchInternalPredict(x []lo.Tuple2[[]int32, []float32], e [][][]uint16, jobs int) []float32
}

type Dataset 

type Dataset struct {
	Index                  dataset.UnifiedIndex
	UserLabels             [][]lo.Tuple2[int32, float32]
	ItemLabels             [][]lo.Tuple2[int32, float32]
	ContextLabels          [][]lo.Tuple2[int32, float32]
	Users                  []int32
	Items                  []int32
	Target                 []float32
	Timestamps             []time.Time
	ItemEmbeddings         [][][]uint16 // Index by row id, embedding id, embedding dimension; stored as BF16 bits
	ItemEmbeddingDimension []int
	ItemEmbeddingIndex     *dataset.Index
	PositiveCount          int
	NegativeCount          int
}

Dataset for click-through-rate models.

func LoadDataFromBuiltIn 

func LoadDataFromBuiltIn(name string) (train, test *Dataset, err error)

LoadDataFromBuiltIn loads built-in dataset.

func (*Dataset) Count 

func (dataset *Dataset) Count() int

Count returns the number of samples.

func (*Dataset) CountContextLabels 

func (dataset *Dataset) CountContextLabels() int

func (*Dataset) CountItemLabels 

func (dataset *Dataset) CountItemLabels() int

func (*Dataset) CountItems 

func (dataset *Dataset) CountItems() int

CountItems returns the number of items.

func (*Dataset) CountNegative 

func (dataset *Dataset) CountNegative() int

func (*Dataset) CountPositive 

func (dataset *Dataset) CountPositive() int

func (*Dataset) CountUserLabels 

func (dataset *Dataset) CountUserLabels() int

func (*Dataset) CountUsers 

func (dataset *Dataset) CountUsers() int

CountUsers returns the number of users.

func (*Dataset) Get 

func (dataset *Dataset) Get(i int) ([]int32, []float32, [][]uint16, float32)

Get returns the i-th sample.

func (*Dataset) GetIndex 

func (dataset *Dataset) GetIndex() dataset.UnifiedIndex

func (*Dataset) GetItemEmbeddingDim added in v0.5.2 

func (dataset *Dataset) GetItemEmbeddingDim() []int

func (*Dataset) GetItemEmbeddingIndex added in v0.5.2 

func (dataset *Dataset) GetItemEmbeddingIndex() *dataset.Index

func (*Dataset) GetTarget 

func (dataset *Dataset) GetTarget(i int) float32

func (*Dataset) Split 

func (dataset *Dataset) Split(ratio float32, seed int64) (*Dataset, *Dataset)

Split a dataset to training set and test set.

func (*Dataset) SplitByUserTime added in v0.5.8 

func (dataset *Dataset) SplitByUserTime(ratio float32) (*Dataset, *Dataset)

SplitByUserTime splits the dataset within each user by timestamp. Earlier samples are used for training and the latest samples are used for testing. At least one sample is kept in train for every user with two or more samples.

type Embedding 

type Embedding struct {
	Name  string
	Value []uint16
}

func ConvertEmbeddings 

func ConvertEmbeddings(o any) []Embedding

type FactorizationMachines 

type FactorizationMachines interface {
	model.Model
	Predict(userId, itemId string, userFeatures, itemFeatures []Label) float32
	InternalPredict(x []int32, values []float32) float32
	Fit(ctx context.Context, trainSet, testSet dataset.CTRSplit, config *FitConfig) Score
	Marshal(w io.Writer) error
}

func UnmarshalModel 

func UnmarshalModel(r io.Reader) (FactorizationMachines, error)

type FitConfig 

type FitConfig struct {
	Jobs     int
	Verbose  int
	Patience int
}

func NewFitConfig 

func NewFitConfig() *FitConfig

func (*FitConfig) LoadDefaultIfNil 

func (config *FitConfig) LoadDefaultIfNil() *FitConfig

func (*FitConfig) SetJobs 

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

func (*FitConfig) SetPatience 

func (config *FitConfig) SetPatience(patience int) *FitConfig

func (*FitConfig) SetVerbose 

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

type Label 

type Label struct {
	Name  string
	Value float32
}

func ConvertLabels 

func ConvertLabels(o any) []Label

type MinMaxScaler added in v0.5.7 

type MinMaxScaler struct {
	Min float32
	Max float32
}

MinMaxScaler transforms a single feature by scaling to [0, 1] range. The transformation is given by: 

X_scaled = (X - X.min) / (X.max - X.min)

func NewMinMaxScaler added in v0.5.7 

func NewMinMaxScaler() *MinMaxScaler

NewMinMaxScaler creates a MinMaxScaler.

func (*MinMaxScaler) Fit added in v0.5.7 

func (s *MinMaxScaler) Fit(values []float32)

Fit computes the minimum and maximum values from the given values.

func (*MinMaxScaler) Marshal added in v0.5.7 

func (s *MinMaxScaler) Marshal(w io.Writer) error

Marshal writes the scaler to a writer.

func (*MinMaxScaler) Transform added in v0.5.7 

func (s *MinMaxScaler) Transform(value float32) float32

Transform scales a value to [0, 1] range.

func (*MinMaxScaler) Unmarshal added in v0.5.7 

func (s *MinMaxScaler) Unmarshal(r io.Reader) error

Unmarshal reads the scaler from a reader.

type ModelCreator 

type ModelCreator func() FactorizationMachines

type ModelSearch 

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

func NewModelSearch 

func NewModelSearch(models map[string]ModelCreator, trainSet, testSet dataset.CTRSplit, config *FitConfig) *ModelSearch

func (*ModelSearch) Objective 

func (ms *ModelSearch) Objective(trial goptuna.Trial) (float64, error)

func (*ModelSearch) Result 

func (ms *ModelSearch) Result() meta.Model[Score]

func (*ModelSearch) WithContext 

func (ms *ModelSearch) WithContext(ctx context.Context) *ModelSearch

func (*ModelSearch) WithSpan 

func (ms *ModelSearch) WithSpan(span *monitor.Span) *ModelSearch

type RobustScaler added in v0.5.7 

type RobustScaler struct {
	Median float32
	Q1     float32 // 25th percentile
	Q3     float32 // 75th percentile
	IQR    float32 // Interquartile range (Q3 - Q1)
}

RobustScaler transforms features using statistics that are robust to outliers. It uses the median and interquartile range (IQR) for scaling: 

X_scaled = (X - median) / IQR

where IQR = Q3 - Q1 (75th percentile - 25th percentile)

func NewRobustScaler added in v0.5.7 

func NewRobustScaler() *RobustScaler

NewRobustScaler creates a RobustScaler.

func (*RobustScaler) Fit added in v0.5.7 

func (s *RobustScaler) Fit(values []float32)

Fit computes the median and IQR from the given values.

func (*RobustScaler) Marshal added in v0.5.7 

func (s *RobustScaler) Marshal(w io.Writer) error

Marshal writes the scaler to a writer.

func (*RobustScaler) Transform added in v0.5.7 

func (s *RobustScaler) Transform(value float32) float32

Transform scales a value using median and IQR.

func (*RobustScaler) Unmarshal added in v0.5.7 

func (s *RobustScaler) Unmarshal(r io.Reader) error

Unmarshal reads the scaler from a reader.

type Score 

type Score struct {
	RMSE      float32
	Precision float32
	Recall    float32
	Accuracy  float32
	AUC       float32
}

func EvaluateClassification 

func EvaluateClassification(estimator FactorizationMachines, testSet dataset.CTRSplit, jobs int) Score

EvaluateClassification evaluates factorization machines in classification task.

func EvaluateRegression 

func EvaluateRegression(estimator FactorizationMachines, testSet *Dataset) Score

EvaluateRegression evaluates factorization machines in regression task.

func (Score) BetterThan 

func (score Score) BetterThan(s Score) bool

func (Score) GetValue 

func (score Score) GetValue() float32

func (Score) ZapFields 

func (score Score) ZapFields() []zap.Field

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.