package yggdrasil_decision_forests.serving.embed.proto

Get desktop application:
View/edit binary Protocol Buffers messages

(message has no fields)

Used in: Options

• UNDEFINED = 0

• IF_ELSE = 1

  Generate a large imbricated if-else structure. This solution is slow, consumes a lot of memory, and is mostly implemented for experimentation.

• ROUTING = 2

  Compute the prediction with the routing algorithm. This solution has the best overall complexity, size and speed (though some other methods might sometime be faster).

Used in: Options

(message has no fields)

Output of a classification model.

(message has no fields)

Used in: Options

• CLASS = 0

  The model returns an integer representing the predicted class. This option is fast and simple to use.

• SCORE = 1

  For a binary classification model, the prediction output is an accumulator of type "accumulator_precision" where the sign (positive or negative) indicates the predicted label value. For a multi-class classification model, the predicted output is an array of accumulators where the largest value defines the predicted label. This option is fast but hard to use.

• PROBABILITY = 2

  For a binary classification model, the prediction output is the probability of the positive class. For a multi-class classification model, the predicted output is an array of probabilities for all the classes. Note: This probability might be not calibrated if the model is not calibrated. This option is easy to use but slow.

(message has no fields)

• UNDEFINED = 0

• INT8 = 1

• INT16 = 2

• INT32 = 3

• UINT8 = 4

• UINT16 = 5

• UINT32 = 6

• FLOAT32 = 7

• BOOL = 8

Used in: Options

• optional string package_name = 1

High level options to control the model embedding.

• optional string name = 1

  Name filename and namespace of the generated library.

• optional ClassificationOutput.Enum classification_output = 2

  What is the Predict function is returning in case of a classification model.

• optional bool integerize_output = 3

  If true, integerize the output computation. The model will be smaller and faster, but the predictions will be less accurate.

• optional int32 accumulator_precision_bytes = 4

  Precision of the accumulator when integerize_output=true;

• optional Algorithm.Enum algorithm = 5

  Algorithm used to compute the tree output.

• optional bool monitor_usage = 6

  Add a monitor of the library used in the generated code.

• optional bool categorical_from_string = 7

  If true, generates functions to create categorical feature values from strings. For example, for a categorical feature "X" with an associated "FeatureX" enum class, the method "FeatureXFromString(absl::string_view name) -> FeatureX" is created. Enabling this option might increase the binary size.

• oneof language

  Which programming language should be produced.

  • CC cc = 8

  • Java java = 9