Skip to content

Latest commit

 

History

History
32 lines (21 loc) · 1.88 KB

SPECS.md

File metadata and controls

32 lines (21 loc) · 1.88 KB

fn-cv-ml

Provided a GeoJSON of points with numbers examined and numbers positive, as well as a GeoJSON of prediction points, this function fits a highly adaptive lasso using 10-fold cross validation

Parameters

A nested JSON object containing:

  • points - {GeoJSON FeatureCollection} Required. Features with following properties:

    • n_trials - {integer} Required. Number of individuals examined/tested at each location (‘null’ for points without observations)
    • n_positive - {integer} Required. Number of individuals positive at each location (‘null’ for points without observations)
    • id - {string} Optional id for each point. Must be unique. If not provided, 1:n (where n is the number of Features in the FeatureCollection) will be used.
    • Additional covariate values as obtained using this function.

  • layer_names - {array of strings} Optional. Default is to run with only latitude and longitude. Names relating to the covariate to use to model and predict. Corresponding layer must be present in points. See here for options.

  • importance - {boolean}. Should random forest importance (gini) be returned? Defaults to TRUE.

Constraints

  • maximum number of points/features
  • maximum number of layers is XX
  • can only include points within a single country

Response

A JSON containing

  • points {GeoJSON FeatureCollection} with the following additional fields:

    • fitted_prediction - predicted probability (equivalent of fitted values at observation points and predictions at points without observations)
    • cv_predictions - cross-validated predicted probability (only available at observation points)

  • importance {array}. Gini impurity values for each covariate. Only returned if importance == TRUE