Basic Walkthrough

1. Introduction
2. The dataset
3. Training the model
- 3.1 Using the lightgbm() function
- 3.2 Using the lgb.train() function
4. References

1. Introduction

Welcome to the world of LightGBM, a highly efficient gradient boosting implementation (Ke et al. 2017).

library(lightgbm)

This vignette will guide you through its basic usage. It will show how to build a simple binary classification model based on a subset of the bank dataset (Moro, Cortez, and Rita 2014). You will use the two input features “age” and “balance” to predict whether a client has subscribed a term deposit.

2. The dataset

The dataset looks as follows.

data(bank, package = "lightgbm")

bank[1L:5L, c("y", "age", "balance")]
#>         y   age balance
#>    <char> <int>   <int>
#> 1:     no    30    1787
#> 2:     no    33    4789
#> 3:     no    35    1350
#> 4:     no    30    1476
#> 5:     no    59       0

# Distribution of the response
table(bank$y)
#> 
#>   no  yes 
#> 4000  521

3. Training the model

The R-package of LightGBM offers two functions to train a model:

lgb.train(): This is the main training logic. It offers full flexibility but requires a Dataset object created by the lgb.Dataset() function.
lightgbm(): Simpler, but less flexible. Data can be passed without having to bother with lgb.Dataset().

3.1 Using the `lightgbm()` function

In a first step, you need to convert data to numeric. Afterwards, you are ready to fit the model by the lightgbm() function.

# Numeric response and feature matrix
y <- as.numeric(bank$y == "yes")
X <- data.matrix(bank[, c("age", "balance")])

# Train
fit <- lightgbm(
  data = X
  , label = y
  , params = list(
    num_leaves = 4L
    , learning_rate = 1.0
    , objective = "binary"
  )
  , nrounds = 10L
  , verbose = -1L
)

# Result
summary(predict(fit, X))
#>    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
#> 0.01192 0.07370 0.09871 0.11593 0.14135 0.65796

It seems to have worked! And the predictions are indeed probabilities between 0 and 1.

3.2 Using the `lgb.train()` function

Alternatively, you can go for the more flexible interface lgb.train(). Here, as an additional step, you need to prepare y and X by the data API lgb.Dataset() of LightGBM. Parameters are passed to lgb.train() as a named list.

# Data interface
dtrain <- lgb.Dataset(X, label = y)

# Parameters
params <- list(
  objective = "binary"
  , num_leaves = 4L
  , learning_rate = 1.0
)

# Train
fit <- lgb.train(
  params
  , data = dtrain
  , nrounds = 10L
  , verbose = -1L
)

Try it out! If stuck, visit LightGBM’s documentation for more details.

4. References

Ke, Guolin, Qi Meng, Thomas Finley, Taifeng Wang, Wei Chen, Weidong Ma, Qiwei Ye, and Tie-Yan Liu. 2017. “LightGBM: A Highly Efficient Gradient Boosting Decision Tree.” In Advances in Neural Information Processing Systems 30 (NIPS 2017).

Moro, Sérgio, Paulo Cortez, and Paulo Rita. 2014. “A Data-Driven Approach to Predict the Success of Bank Telemarketing.” Decision Support Systems 62: 22–31.