Classes
class	XBGoostTextNode

Functions
	dump_tree (tree_structure)
	dump2ROOT (model, output_filename, output_treename='xgboost')
	convertXGBoostToRootTree (model, output_filename, tree_name='xgboost')
	test (model_file, tree_file, objective, tree_name='xgboost', ntests=10000, test_file=None)
	test_regression (booster, mva_utils, objective, ntests=10000, test_file=None)
	test_binary (booster, mva_utils, objective, ntests=10000, test_file=None)
	test_multiclass (booster, mva_utils, objective, ntests=10000, test_file=None)
	check_file (fn)

Variables
str	__author__ = "Yuan-Tang Chou"
	level
	parser = argparse.ArgumentParser(description=__doc__)
	help
	type
	str
	default
	action
	int
	args = parser.parse_args()
list	supported_objective = ['binary:logistic', 'reg:linear', 'reg:squarederror','multi:softprob']
	output_treename = convertXGBoostToRootTree(args.input, args.output, args.tree_name)
	result = test(args.input, args.output, args.objective, args.tree_name, args.ntests, args.test_file)
	objective = args.objective
	data = np.load(args.test_file)

Detailed Description

Convert XGBoost model to TTree to be used with MVAUtils.

Function Documentation

◆ check_file()

util.convertXGBoostToRootTree.check_file ( fn )

Definition at line 279 of file convertXGBoostToRootTree.py.

def check_file(fn):
    f = ROOT.TFile.Open(fn)
    keys = f.GetListOfKeys()
    keys = list(keys)
    if len(keys) != 1:
        logging.info("file %s is empty", fn)
        return False
    tree = f.Get(keys[0].GetName())
    if type(tree) is not ROOT.TTree:
        logging.info("cannot find TTree in file %s", fn)
        return False
    if not tree.GetEntries():
        logging.info("tree is empty")
        return False
    return True
 
 

◆ convertXGBoostToRootTree()

util.convertXGBoostToRootTree.convertXGBoostToRootTree	(	model,
		output_filename,
		tree_name = 'xgboost' )

Model: - a string, in this case, it is the name of the input file containing the xgboost model
         you can get this model with xgboost with `bst.save_model('my_model.model')
       - directly a xgboost booster object

Definition at line 133 of file convertXGBoostToRootTree.py.

def convertXGBoostToRootTree(model, output_filename, tree_name='xgboost'):
    """
    Model: - a string, in this case, it is the name of the input file containing the xgboost model
             you can get this model with xgboost with `bst.save_model('my_model.model')
           - directly a xgboost booster object
    """
    if type(model) is str:
        bst = xgb.Booster()
        bst.load_model(model)
        return dump2ROOT(bst, output_filename, tree_name)
    else:
        return dump2ROOT(model, output_filename, tree_name)
 
 

◆ dump2ROOT()

util.convertXGBoostToRootTree.dump2ROOT	(	model,
		output_filename,
		output_treename = 'xgboost' )

Definition at line 93 of file convertXGBoostToRootTree.py.

def dump2ROOT(model, output_filename, output_treename='xgboost'):
    model.dump_model('dump_model.json', dump_format='json')
    with open('dump_model.json', 'r') as dump_json:
        model_dump = dump_json.read()
    trees = json.loads(model_dump)
    fout = ROOT.TFile.Open(output_filename, 'recreate')
 
    features_array = ROOT.std.vector('int')()
    values_array = ROOT.std.vector('float')()
    default_lefts_array = ROOT.std.vector('bool')()
 
    title = 'creator=xgboost'
    root_tree = ROOT.TTree(output_treename, title)
    root_tree.Branch('vars', 'vector<int>', ROOT.AddressOf(features_array))
    root_tree.Branch('values', 'vector<float>', ROOT.AddressOf(values_array))
    root_tree.Branch('default_left', 'vector<bool>', ROOT.AddressOf(default_lefts_array))
 
    logging.info("tree support nan: using XGBoost implementation")
 
    for tree in trees:
        tree_structure = tree
        features, values, default_lefts = dump_tree(tree_structure)
 
        features_array.clear()
        values_array.clear()
        default_lefts_array.clear()
 
        for value in values:
            values_array.push_back(value)
        for feature in features:
            features_array.push_back(feature)
        for default_left in default_lefts:
            default_lefts_array.push_back(default_left)
 
        root_tree.Fill()
 
    root_tree.Write()
    fout.Close()
    return output_treename
 

◆ dump_tree()

util.convertXGBoostToRootTree.dump_tree ( tree_structure )

dump a single decision tree to arrays to be written into the TTree

Definition at line 67 of file convertXGBoostToRootTree.py.

def dump_tree(tree_structure):
    """
    dump a single decision tree to arrays to be written into the TTree
    """
 
    split_values = []
    split_features = []
    default_left = []
    top = XBGoostTextNode(tree_structure)
 
    def preorder(node):
        # visit root
        split_features.append(node.get_split_feature())
        split_values.append(node.get_value())
        default_left.append(node.get_default_left())
 
        # visit (yes)left
        if node.get_left() is not None:
            preorder(node.get_left())
        # visit (no)right
        if node.get_right() is not None:
            preorder(node.get_right())
 
    preorder(top)
    return split_features, split_values, default_left
 

◆ test()

util.convertXGBoostToRootTree.test	(	model_file,
		tree_file,
		objective,
		tree_name = 'xgboost',
		ntests = 10000,
		test_file = None )

Definition at line 147 of file convertXGBoostToRootTree.py.

def test(model_file, tree_file, objective, tree_name='xgboost', ntests=10000, test_file=None):
    bst = xgb.Booster()
    bst.load_model(model_file)
    f = ROOT.TFile.Open(tree_file)
    tree = f.Get(tree_name)
    try:
        _ = ROOT.MVAUtils.BDT
    except Exception:
        print("cannot import MVAUtils")
        return None
 
    mva_utils = ROOT.MVAUtils.BDT(tree)
 
    if 'binary' in objective:
        logging.info("testing binary")
        return test_binary(bst, mva_utils, objective, ntests, test_file)
    elif 'multi' in objective:
        logging.info("testing multi-class")
        return test_multiclass(bst,mva_utils, objective, ntests, test_file)
    else:
        logging.info("testing regression")
        return test_regression(bst, mva_utils, objective, ntests, test_file)
 

◆ test_binary()

util.convertXGBoostToRootTree.test_binary	(	booster,
		mva_utils,
		objective,
		ntests = 10000,
		test_file = None )

Definition at line 206 of file convertXGBoostToRootTree.py.

def test_binary(booster, mva_utils, objective, ntests=10000, test_file=None):
    import numpy as np
    logging.info("Testing input features with binary classification")
    if test_file is not None:
        data_input = np.load(test_file)
        logging.info("using as input %s inputs from file %s", len(data_input), test_file)
    else:
        logging.error("Please provide an input test file for testing")
 
    start = time.time()
    dTest = xgb.DMatrix(data_input)
    results_xgboost = booster.predict(dTest)
    logging.info("xgboost (vectorized) timing = %s ms/input", (time.time() - start) * 1000 / len(data_input))
 
    input_values_vector = ROOT.std.vector("float")()
    results_MVAUtils = []
    start = time.time()
    for input_values in data_input:
        input_values_vector.clear()
        for v in input_values:
            input_values_vector.push_back(v)
        output_MVAUtils = mva_utils.GetClassification(input_values_vector)
        results_MVAUtils.append(output_MVAUtils)
    logging.info("mvautils (not vectorized+overhead) timing = %s ms/input", (time.time() - start) * 1000 / len(data_input))
 
    for input_values, output_xgb, output_MVAUtils in zip(data_input, results_xgboost, results_MVAUtils):
        if not np.allclose(output_xgb, output_MVAUtils):
            logging.info("output are different:"
                         "mvautils: %s\n"
                         "xgboost: %s\n"
                         "inputs: %s", output_MVAUtils, output_xgb, input_values)
            return False
    return True
 

◆ test_multiclass()

util.convertXGBoostToRootTree.test_multiclass	(	booster,
		mva_utils,
		objective,
		ntests = 10000,
		test_file = None )

Definition at line 240 of file convertXGBoostToRootTree.py.

def test_multiclass(booster, mva_utils, objective, ntests=10000, test_file=None):
    import numpy as np
    logging.info("using multiclass model")
 
    if test_file is not None:
        data_input = np.load(test_file)
        logging.info("using as input %s inputs from file %s", len(data_input), test_file)
    else:
        logging.error("Please provide an input test file for testing")
 
    start = time.time()
    dTest = xgb.DMatrix(data_input)
    results_xgboost = booster.predict(dTest)
    
    nclasses = results_xgboost.shape[1]
    logging.info("xgboost (vectorized) timing = %s ms/input", (time.time() - start) * 1000 / len(data_input))
 
    input_values_vector = ROOT.std.vector("float")()
    results_MVAUtils = []
    start = time.time()
    for input_values in data_input:
        input_values_vector.clear()
        for v in input_values:
            input_values_vector.push_back(v)
        output_MVAUtils = mva_utils.GetMultiResponse(input_values_vector, nclasses)
        results_MVAUtils.append(output_MVAUtils)
 
    logging.info("mvautils (not vectorized+overhead) timing = %s ms/input", (time.time() - start) * 1000 / len(data_input))
 
    for input_values, output_xgb, output_MVAUtils in zip(data_input, results_xgboost, results_MVAUtils):
        if not np.allclose(output_xgb, output_MVAUtils):
            logging.info("output are different:"
                         "mvautils: %s\n"
                         "xgboost: %s\n"
                         "inputs: %s", output_MVAUtils, output_xgb, input_values)
            return False
    return True
 
 

◆ test_regression()

util.convertXGBoostToRootTree.test_regression	(	booster,
		mva_utils,
		objective,
		ntests = 10000,
		test_file = None )

Definition at line 170 of file convertXGBoostToRootTree.py.

def test_regression(booster, mva_utils, objective, ntests=10000, test_file=None):
    import numpy as np
    logging.info("Tesing input features with regression")
 
    if test_file is not None:
        data_input = np.load(test_file)
        logging.info("using as input %s inputs from file %s", len(data_input), test_file)
    else:
        logging.error("Please provide an input test file for testing")
 
    start = time.time()
    dTest = xgb.DMatrix(data_input)
    results_xgboost = booster.predict(dTest)
    logging.info("xgboost (vectorized) timing = %s ms/input", (time.time() - start) * 1000 / len(data_input))
 
    input_values_vector = ROOT.std.vector("float")()
    results_MVAUtils = []
    start = time.time()
    for input_values in data_input:
        input_values_vector.clear()
        for v in input_values:
            input_values_vector.push_back(v)
        output_MVAUtils = mva_utils.GetResponse(input_values_vector)
        results_MVAUtils.append(output_MVAUtils)
    logging.info("mvautils (not vectorized+overhead) timing = %s ms/input", (time.time() - start) * 1000 / len(data_input))
 
    for input_values, output_xgb, output_MVAUtils in zip(data_input, results_xgboost, results_MVAUtils):
        if not np.allclose(output_xgb, output_MVAUtils, rtol=1E-4):
            logging.info("output are different:"
                         "mvautils: %s\n"
                         "xgboost: %s\n"
                         "inputs: %s", output_MVAUtils, output_xgb, input_values)
            return False
    return True
 
 