Port XGBoost model trained in python to another system written in C/C++

Question

Suppose I have successfully trained a XGBoost machine learning model in python.

x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.2, random_state=7)
model = XGBClassifier()
model.fit(x_train, y_train)
y_pred = model.predict(x_test)

I want to port this model to another system which will be writte in C/C++. To do this, I need to know the internal logic of the XGboost trained model and translate them into a series of if-then-else statements like decision trees, if I am not wrong.

How can this be done? How to find out the internal logic of the XGBoost trained model to implement it on another system?

I am using python 3.7.

Another way of thinking about this question is: "How can I serialize/deserialize the model in a format that both languages can understand?" Pickling may be the answer. This answer has a few potential options: https://stackoverflow.com/questions/1296162/how-can-i-read-a-python-pickle-database-file-from-c — JacobIRR, Sep 28 '19 at 03:49
This does something similar, but doesn't yet offer C++ output: https://github.com/modusdatascience/sklearn2code. C++ could be added with a little work. — jcrudy, Apr 27 '20 at 16:00
have you been able to find what's the internal logic of XGBoost models? — SBF, Aug 25 '23 at 19:45

score 11 · Accepted Answer · answered Apr 27 '20 at 20:41

m2cgen Is an awesome package that will convert Scikit-Learn compatible models into raw code. If you are using XGBoosts sklearn wrappers (which it looks like you are), then you can simply call something like this:

model = XGBClassifier()
model.fit(x_train, y_train)
 ...
import m2cgen as m2c

with open('./model.c','w') as f:
    code = m2c.export_to_c(model)
    f.write(code)

The really awesome thing about this package, is that it supports many different types of models, such as

Linear
SVM
Tree
Random Forest
Boosting

One more thing. m2cgen also supports multiple languages such as

C
C#
Dart
Go
Haskell
Java
JavaScript
PHP
PowerShell
Python
R
Visual Basic

I hope this helps!

score 7 · Answer 2 · answered Apr 27 '20 at 15:38

7

Someone wrote a script that does exactly this. Check out https://github.com/popcorn/xgb2cpp

answered Apr 27 '20 at 15:38

Aaron

801
6
13

score 0 · Answer 3 · answered May 02 '20 at 20:45

The recommended way of using any ml/dl model is making simple RESTful API with flask/bottle (these are lightweight python frameworks) and use them globally with any language.

You can also containerize RESTful API with docker in case you are developing a big project with a lot of models. Even containerized Restful APIs are used to deploy models on the cloud, ex- aws.

If you are interested in getting the logic behind any ml model, always have a look on its source code (on GitHub).

Port XGBoost model trained in python to another system written in C/C++

3 Answers3

Linked