Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[R-package] minor improvements on weight demo #5212

Merged
merged 1 commit into from
May 17, 2022
Merged

[R-package] minor improvements on weight demo #5212

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
23 changes: 8 additions & 15 deletions R-package/demo/weight_param.R
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,13 +6,13 @@
library(lightgbm)

# We will train a model with the following scenarii:
# - Run 1: sum of weights equal to 0.06513 without adjusted regularization (not learning)
# - Run 2: sum of weights equal to 0.06513 with adjusted regularization (learning)
# - Run 3: sum of weights equal to 6513 (x 1e5) with adjusted regularization (learning)
# - Run 1: sum of weights equal to 6513 (x 1e-5) without adjusted regularization (not learning)
# - Run 2: sum of weights equal to 6513 (x 1e-5) adjusted regularization (learning)
# - Run 3: sum of weights equal to 6513 with adjusted regularization (learning)

# Setup small weights
weights1 <- rep(1.0 / 100000.0, 6513L)
weights2 <- rep(1.0 / 100000.0, 1611L)
weights1 <- rep(1e-5, 6513L)
weights2 <- rep(1e-5, 1611L)

# Load data and create datasets
data(agaricus.train, package = "lightgbm")
Expand All @@ -23,7 +23,7 @@ test <- agaricus.test
dtest <- lgb.Dataset.create.valid(dtrain, test$data, label = test$label, weight = weights2)
valids <- list(test = dtest)

# Run 1: sum of weights equal to 0.06513 without adjusted regularization (not learning)
# Run 1: sum of weights equal to 6513 (x 1e-5) without adjusted regularization (not learning)
# It cannot learn because regularization is too large!
# min_sum_hessian alone is bigger than the sum of weights, thus you will never learn anything
params <- list(
Expand All @@ -47,7 +47,7 @@ model <- lgb.train(
weight_loss <- as.numeric(model$record_evals$test$l2$eval)
plot(weight_loss) # Shows how poor the learning was: a straight line!

# Run 2: sum of weights equal to 0.06513 with adjusted regularization (learning)
# Run 2: sum of weights equal to 6513 (x 1e-5) with adjusted regularization (learning)
# Adjusted regularization just consisting in multiplicating results by 1e4 (x10000)
# Notice how it learns, there is no issue as we adjusted regularization ourselves
params <- list(
Expand All @@ -71,15 +71,8 @@ model <- lgb.train(
small_weight_loss <- as.numeric(model$record_evals$test$l2$eval)
plot(small_weight_loss) # It learns!

# Run 3: sum of weights equal to 6513 (x 1e5) with adjusted regularization (learning)

# And now, we are doing as usual
library(lightgbm)
data(agaricus.train, package = "lightgbm")
train <- agaricus.train
# Run 3: sum of weights equal to 6513 with adjusted regularization (learning)
dtrain <- lgb.Dataset(train$data, label = train$label)
data(agaricus.test, package = "lightgbm")
test <- agaricus.test
dtest <- lgb.Dataset.create.valid(dtrain, test$data, label = test$label)
valids <- list(test = dtest)

Expand Down