Automated Machine Learning Model Solver

Trees are all you need

Overview

forester is an R package designed to be intuive, brief in formula, and easy to use. It covers the whole process of creating Machine Learning Models with different known models, such as XGBoost, ranger, LightGBM, CatBoost, etc. in an unified formula which let users to control how core steps work.

Installation

(From CRAN - not yet available)

install.packages("forester")

Installation of catboost library: https://catboost.ai/docs/concepts/r-installation.html

From GitHub

install.packages("devtools")
devtools::install_github("ModelOriented/forester")

Usage

Core functions in forester package are:

• forester(), for automatically creating the whole ML process and returning optimal model.

data("titanic", package = 'DALEX') 
best_model <- forester(data = titanic, target = "survived", type = "classification",
		       metric = "precision", tune = FALSE)

# Print ranking table for models:
print(best_model)

# The best_model object can be used for giving explanation:
modelStudio::modelStudio(best_model)

• make_...(), for independently creating model with explanations.

data("apartments", package = 'DALEX') 
data("apartments_test", package = 'DALEX')
catboost <- make_catboost(apartments, "m2.price", "regression", tune = TRUE, 
		             metric = "rmse") 

# prediction:
apartments_test <- apartments_test[ ,!(colnames(apartments_test) %in% "m2.price"]
predict(catboost, apartments_test)

# explanation:
model_performance(catboost)
plot(model_performance(catboost))
modelStudio::modelStudio(catboost)

References:

• Other functionalities for explanation:
  • Explainer
  • Connection with DALEX family

This project was implemented by Szymon Szmajdziński and Hoang Thien Ly under the supervision of Anna Kozak and Przemysław Biecek at MI^2 DataLab.