Task and Model Types

Task Types

Tangram determines the task, either regression, binary classification, or multiclass classification, based on the type of the target column. If the target column is a Number column, the task will be regression. If the target column is an Enum column with two variants, the task will be binary classification. If the target column is an Enum column with more than two variants, the task will be multiclass classification. Tangram will automatically infer the column types or you can provide the column types explicitly in a JSON configuration file passed tot tangram train with the --config flag. See the guide Train with Custom Configuration.

Model Types

Tangram trains a grid of Linear and Gradient Boosted Decision Tree (GBDT) models for the selected task. The definition of the default grid for each task is defined here: https://github.com/tangramdotdev/tangram/blob/main/crates/core/grid.rs. Alternatively, you can specify your own grid in a JSON configuration file passed to tangram train with the --config flag. See the guide Train with Custom Configuration.

Linear Models

Linear models are models where the relationship between the target column and the features is modeled by a linear function. Linear models are highly interpretable, offer very fast prediction times, and the model size grows linearly with the number of features. However, linear models are limited in learning only linear relationships between the features and the target column.

Gradient Boosted Decision Trees

Gradient Boosted Decision Trees (GBDT) consist of many decision tree models where each subsequent decision tree is trained to learn the error of the previous trees. GBDT’s can learn non-linear relationships between the features and the target column and are among the best perfoming models for tabular data. Tangram’s GBDT implementation is written entirely in Rust and has the lowest memory footprint and fastest training times as compared with the most well known GBDT implementations. See benchmarks.