[R package question] Is that possible to retrieve the score for each sample from lgb.cv?

[ajing]

I just wondering whether this is possible. This function can help to analyze the errors during training and do better feature engineering.

[blewy]

• Hi ajing, i cant get the lgb.cv to work for modeling a continous variable with regression. Where you able to make it work?
  lgb.train worked just fine with an validation dataset, but gb.cv still puzzles me.

• R gets stuck and the folowing warning shows:
  "Warning messages:
  1: In foldVector[y == dimnames(numInClass)$y[i]] <- sample(seqVector) :
  number of items to replace is not a multiple of replacement length "

It seams its searching for a class label. I've used the folowing the code:

"params <- list(objective = "regression", metric = "l2")
model <- lgb.cv(params,
dtrain,
10,
nfold = 5,
min_data = 1,
learning_rate = 1,
early_stopping_rounds = 10)"

[ajing]

What is in your dtrain? What does your y look like?

[blewy]

Its a continous variable

I created dtrain like this:

• dtrain <- lgb.Dataset(as.matrix(train.dataset), label = train.target)

and a summary from train.target:

• summary(train.target)
  Min. 1st Qu. Median Mean 3rd Qu. Max.
  0.67 1204.49 2116.89 3032.27 3859.20 106863.00

[ajing]

Is train.dataset a data.frame? Does it have any missing value?

[blewy]

No Missings and ist a data.frame but I used as.matrix to creata suitable dataset for lgbm. This thing is that training with "lgb.train" works fine, but cv.lgb does not. Have you used cv.lgb for regression before ? did it worked ?

…

On 11 August 2017 at 17:41, ajing ***@***.***> wrote: Is train.dataset a data.frame? Does it have any missing value? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#808 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AGQfk-muqMQHtUT_PfHriV1qYFUEHbrSks5sXIRHgaJpZM4Oxf5V> .

[ajing]

Yes, it worked.

[blewy]

Hum.. ok so the problem is on my data set. Thanks.

…

On 11 Aug 2017 7:59 pm, "ajing" ***@***.***> wrote: Yes, it worked. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#808 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AGQfk1GIKv6d2_bfZTWLzhSdfYpoXRE7ks5sXKSlgaJpZM4Oxf5V> .

[blewy]

I just tried thos code, its still show warnings :( " data<-read.table("./data/lowbwt.dat", header=TRUE, skip=0) data_train <- data[, c(2:11)] target <- data_train$BWT data_train$BWT <- NULL library(lightgbm) dtrain <- lgb.Dataset(as.matrix(data_train), label = target) params <- list(objective = "regression", metric = "l2") model <- lgb.cv(params, dtrain, 10, nfold = 5, min_data = 1, learning_rate = 1, early_stopping_rounds = 10) " The data set its the simple birds dataset, lowbwt.dat.

…

On 11 August 2017 at 20:04, Filipe Santos ***@***.***> wrote: Hum.. ok so the problem is on my data set. Thanks. On 11 Aug 2017 7:59 pm, "ajing" ***@***.***> wrote: > Yes, it worked. > > — > You are receiving this because you commented. > Reply to this email directly, view it on GitHub > <#808 (comment)>, > or mute the thread > <https://github.com/notifications/unsubscribe-auth/AGQfk1GIKv6d2_bfZTWLzhSdfYpoXRE7ks5sXKSlgaJpZM4Oxf5V> > . >

[ajing]

Have you tried this example:

library(lightgbm)
data(agaricus.train, package='lightgbm')
train <- agaricus.train
dtrain <- lgb.Dataset(train$data, label=train$label)
params <- list(objective="regression", metric="l2")
model <- lgb.cv(params, dtrain, 10, nfold=5, min_data=1, learning_rate=1, early_stopping_rounds=10)

This is a classification problem, but you can use regression to fit the model.

[blewy]

Yes, I tried that one, it works fine. But the label (or target variable) its a 0/1 value. I was just checking the possibility of an issue for true continuous variables.

…

On 12 August 2017 at 02:20, ajing ***@***.***> wrote: Have you try this example: library(lightgbm) data(agaricus.train, package='lightgbm') train <- agaricus.train dtrain <- lgb.Dataset(train$data, label=train$label) params <- list(objective="regression", metric="l2") model <- lgb.cv(params, dtrain, 10, nfold=5, min_data=1, learning_rate=1, early_stopping_rounds=10) This is a classification problem, but you can use regression to fit the model. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#808 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AGQfk8z6Ltw9b63x9chLXV4Dbrnwlgbhks5sXP3lgaJpZM4Oxf5V> .

[guolinke]

duplicated, refer to #283