argument list for class template "shark::RFClassifier" is missing

[cydcydcyd]

Hi, I download shark4.0 from https://github.com/Shark-ML/Shark/tree/4.0
I compile the library and try to use RFTrainer, and I get some errors like:
Severity Code Description Project File Line Suppression State
Error (active) argument list for class template "shark::RFClassifier" is missing forestCounting c:\Users\chenyuduo\Desktop\forestCounting\forestCounting\forestCounting\main.cpp 64

Severity Code Description Project File Line Suppression State
Error (active) argument list for class template "shark::RFTrainer" is missing forestCounting c:\Users\chenyuduo\Desktop\forestCounting\forestCounting\forestCounting\main.cpp 65
Can you help me solve it?thanks

[Ulfgard]

Have you followed the tutorials?

http://www.shark-ml.org/sphinx_pages/build/html/rest_sources/tutorials/algorithms/rf.html

There is also a random forest example:
https://github.com/Shark-ML/Shark/blob/master/examples/Supervised/RFTutorial.tpp

[Ulfgard]

I am not seeing your code, so i can only guess. Please note that examples, especially the ones i linkred, required you to define the template arguments: RFClassifier<unsigned int> model; RFTrainer<unsigned int> trainer;

[cydcydcyd]

@Ulfgard
Thank you. I read examples to solve the problem.
Another problem is that RFTrainer is still slow in debug mode.
Did you speed up?

[Ulfgard]

yeah, try it in release mode and allow your compiler to optimize the code.

…

________________________________ From: cydcydcyd [[email protected]] Sent: Monday, February 25, 2019 1:55 AM To: Shark-ML/Shark Cc: Oswin Krause; Mention Subject: Re: [Shark-ML/Shark] argument list for class template "shark::RFClassifier" is missing (#264) @Ulfgard<https://github.com/Ulfgard> Thank you. I read examples to solve the problem. Another problem is that RFTrainer is still slow in debug mode. Did you speed up? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<#264 (comment)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AOWTBkB1G9G4nG9yG_y6nFKQeqKAMYjLks5vQzSBgaJpZM4bIzmc>.

[cydcydcyd]

hello，I try to use RFTrainer as regression, but the result is different from sklearn.
here is some of my codes:
data.shuffle(); data.makeIndependent(); RFTrainer<RealVector> trainer(false,false); trainer.setMaxDepth(10); trainer.setMinSplit(20); RFClassifier<RealVector> model; trainer.train(model, data); Data<RealVector> testInputs = createTestInputs(fMat); Data<RealVector> pred = model(testInputs);

[Ulfgard]

It is different implementations, therefore the result can be different. Also the whole algorithm is stochastic, meaning every run will lead to different results. How different is it? Are you sure, all parameters are the same?

…

________________________________ From: cydcydcyd [[email protected]] Sent: Tuesday, February 26, 2019 9:14 AM To: Shark-ML/Shark Cc: Oswin Krause; Mention Subject: Re: [Shark-ML/Shark] argument list for class template "shark::RFClassifier" is missing (#264) hello，I try to use RFTrainer as regression, but the result is different from sklearn. here is some of my codes: data.shuffle(); data.makeIndependent(); RFTrainer<RealVector> trainer(false,false); trainer.setMaxDepth(10); trainer.setMinSplit(20); RFClassifier<RealVector> model; trainer.train(model, data); Data<RealVector> testInputs = createTestInputs(fMat); Data<RealVector> pred = model(testInputs); — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<#264 (comment)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AOWTBq0mSlRjOBwSZzSmvixpQL-3ZgFVks5vROzlgaJpZM4bIzmc>.

[cydcydcyd]

The results are quite different and I'm not sure if I missed the key steps.
I need to use OpenCV to calculate the feature of the images and the features in cv::Mat will be converted into shark::RealVector.
LabeledData<RealVector, RealVector> labeleddata = createLabeledDataFromRange(inputs, outputs);
WeightedLabeledData<RealVector, RealVector> data(labeleddata, 1.0f);
I set weight of each feature to 1.

[Ulfgard]

so what are the results? just for comparison. You don't have to use wieghts, you can just use the labeleddata

…

________________________________ From: cydcydcyd [[email protected]] Sent: Tuesday, February 26, 2019 9:38 AM To: Shark-ML/Shark Cc: Oswin Krause; Mention Subject: Re: [Shark-ML/Shark] argument list for class template "shark::RFClassifier" is missing (#264) The results are quite different and I'm not sure if I missed the key steps. I need to use OpenCV to calculate the feature of the images and the features in cv::Mat will be converted into shark::RealVector. LabeledData<RealVector, RealVector> labeleddata = createLabeledDataFromRange(inputs, outputs); WeightedLabeledData<RealVector, RealVector> data(labeleddata, 1.0f); I set the all weights of features to 1. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<#264 (comment)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AOWTBiqBqa6F4yR2myAJngGToxd_GV0Oks5vRPKhgaJpZM4bIzmc>.

[cydcydcyd]

My input feature is 637 dimensions, and the output tag is 49 dimensions. There are 2000 data in total. It takes 15sec to train with RFTrainer. Is this normal speed? I use release mode.

The parameter type of the train function is WeightedLabeledData < RealVector, RealVector >. I don't know how to do without weights.