How to modify lightgbm for my custom loss functions? - lightgbm

Question

Which files should be change while adding my own custom loss function? I know I can add my objective and gradient/hessian computation in ObjectiveFunction, just wondering if there is anything else I need to do or if there are other alternatives for custom loss functions.

Answer 1

According to demo files in lightGBM early stopping example,
Setting objective function as:
# User define objective function, given prediction, return gradient and second order gradient
# This is loglikelihood loss
logregobj <- function(preds, dtrain) {
labels <- getinfo(dtrain, "label")
preds <- 1 / (1 + exp(-preds))
grad <- preds - labels
hess <- preds * (1 - preds)
return(list(grad = grad, hess = hess))
}
Setting error function as:
# User defined evaluation function, return a pair metric_name, result, higher_better
# NOTE: when you do customized loss function, the default prediction value is margin
# This may make buildin evalution metric not function properly
# For example, we are doing logistic loss, the prediction is score before logistic transformation
# The buildin evaluation error assumes input is after logistic transformation
# Take this in mind when you use the customization, and maybe you need write customized evaluation function
evalerror <- function(preds, dtrain) {
labels <- getinfo(dtrain, "label")
err <- as.numeric(sum(labels != (preds > 0.5))) / length(labels)
return(list(name = "error", value = err, higher_better = FALSE))
}
And then you can run lightgbm as:
bst <- lgb.train(param,
dtrain,
num_round,
valids,
objective = logregobj,
eval = evalerror,
early_stopping_round = 3)