Question about random forest model

johnnix · **Joined:** Sun Jul 03, 2022 7:03 am **Posts:** 2

Hello,

I am evaluating ALGLIB to be used inside one of my commercial software programs in order to offer my users an easy way to use random forest models for their data. It looks ok so far but I was not able to find a way to "save" the trained model so that it can be loaded and used again. Is this possible?

Thank you all for your time!

Regards

John

Sergey.Bochkanov · **Joined:** Fri May 07, 2010 7:06 am **Posts:** 927

Hi!

It can be performed with dfserialize()/dfunserialize() pair of functions. Random forest can be serialized to string or stream. The data are written using compressed text format, i.e. you can write it to text or XML file, change EOL between Windows and Unix style, and it will work. It is also possible to move models between C++ and C# versions of the library (e.g. train in C++ and use in C#).

Sergey

johnnix · **Joined:** Sun Jul 03, 2022 7:03 am **Posts:** 2

That is great! I will proceed with purchasing the commercial version :)

shaos · **Joined:** Tue Aug 15, 2023 2:28 pm **Posts:** 3

hi，can you provide a code demo to use the dfserialize /dfunserialize function?

does it correct if i ues it as:

Code:

        Qstring outModelDir_path = "D:/out"
        decisionforest model;
        string outModelFilePath = outModelDir_path.toStdString() + "/trainedModel.txt";
   string outModel;
   dfserialize(model, outModel);
   ofstream outModelFile(outModelFilePath.c_str(), ios::out);
   outModelFile << outModel;
 

i am not sure whether is the right way to use it.
it seems the serialized data look like:

looking for your help!

Sergey.Bochkanov · **Joined:** Fri May 07, 2010 7:06 am **Posts:** 927

Hi!

Yes, it is a right way to do. The model is serialized in a portable text format which can tolerate all types of issues arising when moving text between machines and operating systems. In particular, it does not contain non-ASCII symbols, can easily tolerate conversion between Windows and Unix newlines and so on.

Sergei

shaos · **Joined:** Tue Aug 15, 2023 2:28 pm **Posts:** 3

Thats cool! Thanks for your thoughtful work to make it easy to use!
Best regards!
shaos

shaos · **Joined:** Tue Aug 15, 2023 2:28 pm **Posts:** 3

hi!
By following your suggestion, i successfully serialized the model to txt.
I didnot find any demo code in guidebook, is it correct to use dfunserialize function following the code below?
i use C++ to finish this work

Code:

string loadModel(string serializedModelFileStringFormatPath) {
   ifstream serializedModelFile(serializedModelFileStringFormatPath);
   if (!serializedModelFile.is_open()) {
      cout << cannot open serialized data" << endl;
      return "cannot open serialized data";
   }
   ostringstream  tmp;
   tmp << serializedModelFile.rdbuf();
   string  data = tmp.str();
   serializedModelFile.close();
   return data;
}

int main(
  string inModelFilePath = "c:/test/inmodel.txt";
  string modelContent = loadModel(inModelFilePath);
  decisionforest model;
  dfunserialize(modelContent, model);
  double esti = dfprocess0(model, rowItem);
{

I would appreciate a lot!!
shaos

Sergey.Bochkanov · **Joined:** Fri May 07, 2010 7:06 am **Posts:** 927

Yes, you are right, it must work

forum.alglib.net

Forum rules

Question about random forest model

Who is online