Detailed Interface

Data in XGBoost

All data should be transformed to DMatrix DMatrix support to be built directly from

• libSVM txt format file
• XGBoost buffer file
• Julia Array{Real, 2}
• Julia SparseMatrixCSC{K, V} where K<:Real and V<:Int

Model in XGBoost

Booster is the model struct. To get Booster object, you can load from saved model file or use method

Fit model

To fit a model, use the function bst = xgboost(dtrain, nrounds, param=param, watchlist=watchlist, obj=obj, feval=feval) where

• dtrain training data in DMatrix, required
• nrounds training epoch in Integer, required
• param parameter pairs for XGBoost in Dict or Array{(ASCIIString, Any), 1}, optional
• watchlist data need to be evaulated in Array{(DMatrix, ASCIIString), 1}, optional
• obj customized objective function to be boosted, optional
• feval customized evaluation function, optional

Check demo for all usage

Prediction using fitted model

To do prediction, use the function preds = predict(bst, dmat, output_margin=output_margin, ntree_limit=ntree_limit) where

• bst fitted model Booster object, required
• dmat data to be predicted in DMatrix, required
• output_margin whether to change the output to margin probability, Bool, optional
• ntree_limit number of trees used in prediction, Integer, optional

Other useful function

To save Booster object into binary file, use the function save(bst, fname), where

• bst fitted model Booster object, required
• fname output file name in ASCIIString, required

To save DMatrix object into binary file, use the function save(dmat, fname), where

• dmat data in DMatrix object, required
• fname output file name in ASCIIString, required

To dump Booster model into readable txt file, use the function dump_model(bst, fname, fmap=fmap) where

• bst fitted model Booster object, required
• fname output file name in ASCIIString, required
• fmap file path to feature map file, optional

To slice DMatrix, use the function dnew = slice(dmat, idxset) where

• dmat data in DMatrix object, required
• idxset id of index to be selected in Array{T, 1}, required

To get label, weight, base_margin and group from DMatrix, use function get_info(dmat, field) where

• dmat data in DMatrix object, required
• field ASCIIString of label, weight, base_margin and group

To set label, weight, base_margin and group to DMatrix, use function set_info(dmat, field, array) where

• dmat data in DMatrix object, required
• field ASCIIString of label, weight, base_margin and group
• arraymeta info in Array{T, 1} to be set into dmat

To do cross validation, use the function nfold_cv(param, dtrain, num_boost_round, nfold, metrics=metrics, obj=obj, feval=feval, fpreproc=fpreproc, show_stdv=show_stdv, seed=seed) where

• param parameter pairs for XGBoost in Dict or Array{(ASCIIString, Any), 1}, required
• dtrain training data in DMatrix, required
• num_boost_round training epoch in Integer, required
• nfold fold used for cv in Integer, required
• metrics metrics used in cv in Array{ASCIIString, 1}, optional
• obj customized objective function to be boosted, optional
• feval customized evaluation function, optional
• fpreproc customized preprocessing function, optional
• show_stdv whether show std value for result, optional
• seed random seed in Integer