I have problems with long-time consumption in grid search. Can I have some help for hyperparameter tunning by grid search in catboost? R language
`Code 1: 7 seconds
grid <- expand.grid(depth = 2,
learning_rate = 0.01,
iterations = 100,
l2_leaf_reg = 10,
rsm = 0.8,
border_count=32)
report <- train( HAPO_7yrs_imputed1[, -1], as.factor(make.names(HAPO_7yrs_imputed1[, 1])),
class_weights=c(1,1.5,15),
method = catboost.caret,
logging_level = 'Verbose', preProc = NULL,
tuneGrid = grid, trControl = fit_control)
Code 2: 7 mins
grid <- expand.grid(depth = 10,
learning_rate = 0.01,
iterations = 1000,
l2_leaf_reg = 0.1,
rsm = 1,
border_count=256)
report <- train( HAPO_7yrs_imputed1[, -1], as.factor(make.names(HAPO_7yrs_imputed1[, 1])),
class_weights=c(1,1.5,15),
method = catboost.caret,
logging_level = 'Verbose', preProc = NULL,
tuneGrid = grid, trControl = fit_control)
Code 3: still running after 10 hours
grid <- expand.grid(depth = seq(2,10,length.out=2),
learning_rate = seq(0.01,0.3,length.out=2),
iterations = c(100,1000),
l2_leaf_reg = 10^seq(-1,2,length.out=2),
rsm = seq(0.8,1,length.out=2),
border_count=32*2^seq(0,3,length.out=2)
)
report <- train( HAPO_7yrs_imputed1[, -1], as.factor(make.names(HAPO_7yrs_imputed1[, 1])),
class_weights=c(1,1.5,15),
method = catboost.caret,
logging_level = 'Verbose', preProc = NULL,
tuneGrid = grid, trControl = fit_control) `
I tried to set only one pair of hyperparameters (plz see code 1 and code 2) and record their time consumption.
I would like to know how to reduce time in grid search for hyperparameter tuning.
Thank you!
New contributor