Package

mimir

models

Permalink

package models

Visibility
  1. Public
  2. All

Type Members

  1. class CommentModel extends Model with Serializable with SourcedFeedback

    Permalink

    A model representing a key-repair choice.

    A model representing a key-repair choice.

    The index is ignored. The one argument is a value for the key. The return value is an integer identifying the ordinal position of the selected value, starting with 0.

    Annotations
    @SerialVersionUID()
  2. trait DataIndependentFeedback extends SourcedFeedbackT[Int]

    Permalink
  3. class DefaultMetaModel extends Model with DataIndependentFeedback with NoArgModel with FiniteDiscreteDomain

    Permalink

    A dumb, default Meta-Model to stand in until we get something better.

    A dumb, default Meta-Model to stand in until we get something better.

    This meta model always ignores VG arguments and picks the first model in the list.

    Annotations
    @SerialVersionUID()
  4. class DetectHeaderModel extends Model with Serializable with SourcedFeedback with NeedsDatabase

    Permalink
    Annotations
    @SerialVersionUID()
  5. class EditDistanceMatchModel extends Model with DataIndependentFeedback with NoArgModel with FiniteDiscreteDomain

    Permalink
    Annotations
    @SerialVersionUID()
  6. case class FeedbackSourceIdentifier(id: String = "", name: String = "truth") extends Product with Serializable

    Permalink
  7. trait FiniteDiscreteDomain extends AnyRef

    Permalink
  8. class GeocodingModel extends Model with Serializable with NeedsReconnectToDatabase with ModelCache with SourcedFeedback with FiniteDiscreteDomain

    Permalink

    A model representing a key-repair choice.

    A model representing a key-repair choice.

    The index is ignored. The one argument is a value for the key. The return value is an integer identifying the ordinal position of the selected value, starting with 0.

    Annotations
    @SerialVersionUID()
  9. class MissingKeyModel extends Model with Serializable with FiniteDiscreteDomain with SourcedFeedback

    Permalink

    A model representing a key-repair choice.

    A model representing a key-repair choice.

    The index is ignored. The one argument is a value for the key. The return value is an integer identifying the ordinal position of the selected value, starting with 0.

    Annotations
    @SerialVersionUID()
  10. abstract class Model extends Serializable

    Permalink

    Root class for Model objects.

    Root class for Model objects.

    Models are one of the primitive building blocks in Mimir. While VGTerms (defined in the mimir.ctables package) serve to create placeholders in relational data, Models serve to dictate how these placeholders are filled in.

    Each discrete placeholder value in a relation is associated with a single variable. Variables are identified by an index (idx), and by zero or more argument expressions.

    - Indexes allow a single model object to define multiple categories of variables. Indexes are not data-dependent: For any given query there may only ever be a finite number of index values. However, variables with different indexes may follow different typing rules.

    - Argument expressions allow variables to be dynamically created based on the data, for example one variable per row. However, variables distinguished only by argument expressions must follow the same typing rules.

    In short, the main distinction between indexes and arguments is how the variables interacts with Mimir's typesystem. Arguments can create an arbitrary number of variable instances per query, but must all follow the same typescheme. Meanwhile Indexes can create variables with different types, but there can only be a finite number of indexes in use per query.

    Models are left intentionally abstract. For the moment, at least, we do not try to dictate whether the model should be defined using probability theory, fuzzy logic, belief theory, or any other type of principled mechanism. Rather, the interface simply requires the model to be able to generate a most likely bestGuess value, and be able to draw sample values of possible outputs.

    That said, there are specific specific classes of model that are intended to fulfil specific roles (see ModelRegistry for more details). Models that follow these patterns are expected to conform to specific conventions in terms of their types, how they use arguments, and how they are constructed.

    Annotations
    @SerialVersionUID()
  11. trait ModelCache extends AnyRef

    Permalink
  12. case class ModelException(error: String) extends RAException with Product with Serializable

    Permalink
  13. class ModelManager extends LazyLogging

    Permalink

    The ModelManager handles model persistence.

    The ModelManager handles model persistence.

    The main function of the ModelManager is to provide a persistent (Name -> Model) mapping. Associations are created through persist, removed with drop, and accessed with get. The name of the model is given by Model's name field

    The secondary function is garbage collection. The manager also tracks a second set of 'owner' entities. Owners can be used to cascade deletes on the owner entity to the models, and allows for reference counting with multiple owners.

    See below in this file for some traits used to decode things.

  14. trait NeedsDatabase extends NeedsReconnectToDatabase

    Permalink
  15. trait NeedsReconnectToDatabase extends AnyRef

    Permalink
  16. trait NoArgModel extends AnyRef

    Permalink
  17. case class NoOpModel(name: String, reasonText: String) extends Model with Serializable with Product

    Permalink
  18. class PickerModel extends Model with Serializable with NeedsReconnectToDatabase with FiniteDiscreteDomain with SourcedFeedback with ModelCache

    Permalink

    A model representing a key-repair choice.

    A model representing a key-repair choice.

    The index is ignored. The one argument is a value for the key. The return value is an integer identifying the ordinal position of the selected value, starting with 0.

    Annotations
    @SerialVersionUID()
  19. class RepairKeyModel extends Model with FiniteDiscreteDomain with NeedsReconnectToDatabase with SourcedFeedbackT[List[PrimitiveValue]]

    Permalink

    A model representing a key-repair choice.

    A model representing a key-repair choice.

    The index is ignored. The one argument is a value for the key. The return value is an integer identifying the ordinal position of the selected value, starting with 0.

    Annotations
    @SerialVersionUID()
  20. class SimpleFuncDepModel extends Model with NeedsReconnectToDatabase

    Permalink
    Annotations
    @SerialVersionUID()
  21. class SimpleSeriesModel extends Model with NeedsReconnectToDatabase with SourcedFeedback with ModelCache

    Permalink

    A model performs estimation of missing value column based on the column that follows a series.

    A model performs estimation of missing value column based on the column that follows a series. Best Guess : Performs the best guess based on the weighted-average of upper and lower bound values. Sample : Picks a random value within the range of uppper and lower bound Train : Performs best guess on missing value fields. For each missing value, a map in created [ ROWID -> (Best Guess Value, Lower Bound Value, Upper Bound Value, Recieved Feedback) ]

    Annotations
    @SerialVersionUID()
  22. class SimpleSparkClassifierModel extends Model with NeedsReconnectToDatabase with SourcedFeedback with ModelCache

    Permalink
    Annotations
    @SerialVersionUID()
  23. class SimpleWekaModel extends Model with NeedsReconnectToDatabase with SourcedFeedback

    Permalink
    Annotations
    @SerialVersionUID()
  24. trait SourcedFeedback extends SourcedFeedbackT[String]

    Permalink
  25. trait SourcedFeedbackT[T] extends AnyRef

    Permalink
  26. class TypeInferenceModel extends Model with SourcedFeedback with FiniteDiscreteDomain

    Permalink
    Annotations
    @SerialVersionUID()

Value Members

  1. object DetectHeader

    Permalink
  2. object EditDistanceMatchModel extends Serializable

    Permalink
  3. object FeedbackSource

    Permalink
  4. object FuncDepModel

    Permalink
  5. object ModelRegistry

    Permalink

    The Model Registry is a central point for organizing Models that fulfil specific roles.

    The Model Registry is a central point for organizing Models that fulfil specific roles. As of right now, we consider three basic categories of model:

    • Imputation: Models that 'fill in' missing or erroneous values
    • SchemaMatch: Models that align columns of two relations
    • EntityMatch: Models that align rows of two relations (TBD)
  6. object SeriesMissingValueModel

    Permalink
  7. object SparkClassifierModel

    Permalink
  8. object TypeInferenceModel extends Serializable

    Permalink
  9. object UniformDistribution extends Model with Serializable

    Permalink
  10. object WekaModel

    Permalink

Ungrouped