package yggdrasil_decision_forests.utils.model_analysis.proto

Get desktop application:
View/edit binary Protocol Buffers messages

Results of a model analysis.

Used in: StandaloneAnalysisResult

• optional proto.PartialDependencePlotSet pdp_set = 1

• optional proto.PartialDependencePlotSet cep_set = 2

• map<string, model.proto.VariableImportanceSet> variable_importances = 3

Feature variation in a prediction analysis.

Used in: PredictionAnalysisResult

• repeated FeatureVariationItem items = 1

Used in: FeatureVariation

• repeated FeatureVariationItem.Bin bins = 1

  Bins, i.e., model outputs.

• repeated FeatureVariationItem.Attribute attributes = 2

  Attributes being varied.

Used in: FeatureVariationItem

• optional int32 column_idx = 2

• oneof type

  • Attribute.Numerical numerical = 3

  • Attribute.Categorical categorical = 4

  • Attribute.Boolean boolean = 5

Used in: Attribute

(message has no fields)

Used in: Attribute

• optional int32 num_values = 1

Used in: Attribute

• repeated float values = 1

Used in: FeatureVariationItem

• optional model.proto.Prediction prediction = 1

Configuration for the model analysis

• optional int32 num_threads = 1

  Number of threads used for the computation.

• optional Options.PlotConfig pdp = 2

  Enable the computation of Partial Dependence Plots.

• optional Options.PlotConfig cep = 3

  Enable the computation of Conditional Expectation Plot.

• optional Options.PermutedVariableImportance permuted_variable_importance = 7

  Measure the importance of each input feature using permutation variable importance: https://christophm.github.io/interpretable-ml-book/feature-importance.html. Permuted variable importances are not yet supported for anomaly detection models.

• optional Options.ShapVariableImportance shap_variable_importance = 22

• optional bool include_model_structural_variable_importances = 17

• optional int32 figure_width = 8

  Size, in pixel, of a figure (possibly composed of multiple plots).

• optional int32 plot_width = 9

  Size in pixel of an individual plot.

• optional int32 plot_height = 10

• optional Options.ReportHeader report_header = 11

• optional Options.TableOfContent table_of_content = 12

• optional Options.ReportSetup report_setup = 13

• optional Options.Dataspec dataspec = 14

• optional Options.ModelDescription model_description = 15

• optional Options.Plot plot = 16

• optional string html_id_prefix = 18

  Prefix used to generate unique html element ids. If not set, use a random prefix.

• optional float maximum_duration_seconds = 19

  Maximum duration of the analysis in seconds.

• optional int32 random_seed = 20

  Random seed for randomized tasks.

• repeated string features = 21

  Features to analyse for the PDP and CEP plots. If not set (i.e., empty), all the model features are analyzed. If set, `features` defines the order of the features in the analysis. Does not impact the variable importance features.

Used in: Options

• optional bool enabled = 1

Used in: Options

• optional bool enabled = 1

Used in: Options

• optional bool enabled = 1

  If the model does not have labels (e.g., anomaly detection without labels), permutation variable importances are not computed, even if enabled=True.

• optional int32 num_rounds = 2

  Number of repetitions of the estimation. More repetitions increase the quality of the variable importance estimates.

Used in: Options

• optional bool show_interactive_menu = 1

Configuration of the computation of the PDP and CEP.

Used in: Options

• optional bool enabled = 1

  Enable the computation.

• optional float example_sampling = 2

• optional int32 num_numerical_bins = 3

Used in: Options

• optional bool enabled = 1

Used in: Options

• optional bool enabled = 1

Used in: Options

• optional bool enabled = 1

  Enable the computation of Shap value variable importances.

• optional float example_sampling = 2

  Ratio of examples used to compute the shap value variable importances. Reduce this value to speed-up the computation.

Used in: Options

• optional bool enabled = 1

• optional int32 numerical_num_bins = 1

• optional int32 plot_width = 2

  Size in pixel of an individual plot.

• optional int32 plot_height = 3

• optional string html_id_prefix = 4

  Prefix used to generate unique html element ids. If not set, use a random prefix.

• repeated string features = 5

  Features to analyse. If not set (i.e., empty), all the model features are analyzed. If set, `features` defines the order of the features in the analysis.

Result of a prediction analysis.

• optional dataset.proto.DataSpecification data_spec = 1

• optional int32 label_col_idx = 2

• optional model.proto.Task task = 3

• optional FeatureVariation feature_variation = 4

• optional dataset.proto.Example example = 5

• optional model.proto.Prediction prediction = 6

An analysis result that does not need a model or a dataset to be displayed.

• optional AnalysisResult core_analysis = 1

• optional string dataset_path = 2

• optional string model_path = 3

• optional dataset.proto.DataSpecification data_spec = 4

• optional int32 label_col_idx = 5

• optional model.proto.Task task = 6

• optional string model_description = 7