fastml: Fast Machine Learning Model Training and Evaluation

fastml is a streamlined R package designed to simplify the training, evaluation, and comparison of multiple machine learning models. It offers comprehensive data preprocessing, supports a wide range of algorithms with hyperparameter tuning, and provides performance metrics alongside visualization tools to facilitate efficient and effective machine learning workflows.

Features

• Comprehensive Data Preprocessing: Handle missing values, encode categorical variables, and apply various scaling methods with minimal code.
• Support for Multiple Algorithms: Train a wide array of machine learning models including XGBoost, Random Forest, SVMs, KNN, Neural Networks, and more.
• Hyperparameter Tuning: Customize and automate hyperparameter tuning for each algorithm to optimize model performance.
• Performance Evaluation: Evaluate models using metrics like Accuracy, Kappa, Sensitivity, Specificity, Precision, F1 Score, and ROC AUC.
• Visualization Tools: Generate comparison plots to visualize and compare the performance of different models effortlessly.
• Easy Integration: Designed to integrate seamlessly into your existing R workflows with intuitive function interfaces.

Installation

From CRAN

You can install the latest stable version of fastml from CRAN using:

install.packages("fastml")

You can install all dependencies (additional models) using:

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

From GitHub

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

# Install devtools if you haven't already
install.packages("devtools")

# Install fastml from GitHub
devtools::install_github("selcukorkmaz/fastml")

Quick Start

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

library(fastml)

# Example dataset
data(iris)
iris <- iris[iris$Species != "setosa", ]  # Binary classification
iris$Species <- factor(iris$Species)

# Train models
model <- fastml(
  data = iris,
  label = "Species"
)

# View model summary
summary(model)

Tuning Strategies

fastml supports both grid search and Bayesian optimization through the tuning_strategy argument. Use "grid" for a regular parameter grid or "bayes" for Bayesian hyperparameter search. The tuning_iterations parameter controls the number of iterations only when tuning_strategy = "bayes" and is ignored otherwise.

Explainability

fastexplain() provides several ways to understand trained models. Set the method argument to choose an approach:

# LIME explanations
explain_lime(model)

# ICE curves
fastexplain(model, method = "ice", features = "Sepal.Length")

# Accumulated Local Effects
fastexplain(model, method = "ale", features = "Sepal.Length")

# Surrogate tree
fastexplain(model, method = "surrogate")

# Interaction strength
fastexplain(model, method = "interaction")

# Counterfactual explanation for a single observation
fastexplain(model, method = "counterfactual", observation = iris[1, ])