Polymorphism with classifiers about grt HOT 3 CLOSED

In what way are you trying to run the classifiers at the same time?

e.g., are you trying to create an ensemble (where all the classifiers are trying to recognize the same gestures, but you are combining multiple classifiers together to improve the accuracy), or are you using different classifiers at the same time to recognize different types of gestures (e.g., a basic classifier to detect that a user's hands are in a specific interaction area - such as in front of their torso - and a more complex classifier to recognize specific gestures that occur in the interaction area)?

from grt.

Hello Nick,

They are all trying to recognise the same gesture. I am very much at a
testing/building stage so I am not sure what performs better. The idea was
to have an array of Classifiers and go through all of them and output the
predictions to compare accuracies while in use with live data.

On 28 January 2016 at 06:15, ngillian-google [email protected]
wrote:

In what way are you trying to run the classifiers at the same time?

e.g., are you trying to create an ensemble (where all the classifiers are
trying to recognize the same gestures, but you are combining multiple
classifiers together to improve the accuracy), or are you using different
classifiers at the same time to recognize different types of gestures
(e.g., a basic classifier to detect that a user's hands are in a specific
interaction area - such as in front of their torso - and a more complex
classifier to recognize specific gestures that occur in the interaction
area)?

—
Reply to this email directly or view it on GitHub
#44 (comment).

from grt.

Ahh, so if you are trying to run multiple classifiers in parallel for testing purposes, then I would advice setting up a vector of pipelines, with each pipeline holding one of the classifiers. This also gives you the option to add slightly different preprocessing/feature extraction modules or settings for each pipeline if you need it.

Then for each pipeline you would call:

for(size_t i = 0; i<pipelines.size(); i++){
pipelines[ i ].train( trainingData );
}

and

for(size_t i = 0; i<pipelines.size(); i++){
pipelines[ i ].predict( inputData );
}

for training and predicting.

from grt.