eval

type LearningDataStats struct {
	NumProcessed       int     `msgpack:"numProcessed"`
	NumFailed          int     `msgpack:"numFailed"`
	DeduplicationRatio float64 `msgpack:"deduplicationRatio"`
}

type MLModel interface {

	// Train trains the model based on input data. In case the model
	// supports only inference (e.g. our XGBoost), this should just prepare
	// data to a format required by actual program performing the learning.
	Train(ctx context.Context, data []feats.QueryEvaluation, slowQueriesTime float64, comment string) error

	Predict(feats.QueryEvaluation) predict.Prediction
	SetClassThreshold(v float64)
	GetClassThreshold() float64
	GetSlowQueriesThresholdTime() float64
	SaveToFile(string) error
	GetInfo() string

	// IsInferenceOnly specifies whether the model also supports
	IsInferenceOnly() bool

	// CreateModelFileName should generate proper model filename based
	// on the feature (i.e. input) file name. This should keep data and
	// model names organized and easy to search through.
	CreateModelFileName(featFile string) string
}

type PrecAndRecall struct {
	Precision float64
	Recall    float64
	FBeta     float64
}

type Predictor struct {
	Evaluations []feats.QueryEvaluation

	LearningDataStats LearningDataStats
	// contains filtered or unexported fields
}

type QueryStatsRecord struct {
	Corpus        string  `json:"corpus"`
	CorpusSize    int64   `json:"corpusSize"`
	SubcorpusSize int64   `json:"subcorpusSize"`
	TimeProc      float64 `json:"timeProc"`
	Query         string  `json:"query"`

	// IsSynthetic specifies whether the record comes from
	// production KonText stats log or if it is generated
	// using a benchmarking module (= MQuery).
	IsSynthetic bool `json:"isSynthetic,omitempty"`

	FlaggedAsSlow bool `json:"flaggedAsSlow,omitempty"`
}

type Reporter struct {
	RFAccuracyScript string

	MisclassQueriesOutPath string
	// contains filtered or unexported fields
}