Movatterモバイル変換

fastml is a streamlined R package designed tosimplify the training, evaluation, and comparison of multiple machinelearning models. It offers comprehensive data preprocessing, supports awide range of algorithms with hyperparameter tuning, and providesperformance metrics alongside visualization tools to facilitateefficient and effective machine learning workflows.

Features

• Comprehensive Data Preprocessing: Handle missingvalues, encode categorical variables, and apply various scaling methodswith minimal code.
• Support for Multiple Algorithms: Train a wide arrayof machine learning models including XGBoost, Random Forest, SVMs, KNN,Neural Networks, and more.
• Hyperparameter Tuning: Customize and automatehyperparameter tuning for each algorithm to optimize modelperformance.
• Performance Evaluation: Evaluate models usingmetrics like Accuracy, Kappa, Sensitivity, Specificity, Precision, F1Score, and ROC AUC.
• Visualization Tools: Generate comparison plots tovisualize and compare the performance of different modelseffortlessly.
• Easy Integration: Designed to integrate seamlesslyinto your existing R workflows with intuitive function interfaces.

Installation

From CRAN

You can install the latest stable version offastmlfrom CRAN using:

install.packages("fastml")

You can install all dependencies (additional models) using:

# install all dependencies - recommendedinstall.packages("fastml",dependencies =TRUE)

From GitHub

For the development version, install directly from GitHub using thedevtools package:

# Install devtools if you haven't alreadyinstall.packages("devtools")# Install fastml from GitHubdevtools::install_github("selcukorkmaz/fastml")

Quick Start

Here’s a simple workflow to get you started with fastml:

library(fastml)# Example datasetdata(iris)iris<- iris[iris$Species!="setosa", ]# Binary classificationiris$Species<-factor(iris$Species)# Train modelsmodel<-fastml(data = iris,label ="Species")# View model summarysummary(model)

Tuning Strategies

fastml supports both grid search and Bayesian optimization throughthetuning_strategy argument. Use"grid" for aregular parameter grid or"bayes" for Bayesianhyperparameter search. Thetuning_iterations parametercontrols the number of iterationsonly whentuning_strategy = "bayes" and is ignored otherwise.

Explainability

fastexplain() provides several ways to understandtrained models. Set themethod argument to choose anapproach:

# LIME explanationsexplain_lime(model)# ICE curvesfastexplain(model,method ="ice",features ="Sepal.Length")# Accumulated Local Effectsfastexplain(model,method ="ale",features ="Sepal.Length")# Surrogate treefastexplain(model,method ="surrogate")# Interaction strengthfastexplain(model,method ="interaction")# Counterfactual explanation for a single observationfastexplain(model,method ="counterfactual",observation = iris[1, ])