MATLAB: Can I use NN built with Fitnet for Reinforcement Learning toolbox with DQN agent

deep learningDeep Learning Toolboxdeep neural networkdqnfitnetneural networknnreinforcement learningReinforcement Learning Toolbox

Hello ,
Can I reuse the neural network built with Fitnet fuction (Function fitting neural network), inside Reinforcement Learning toolbox with DQN agent?
The neural network does not look like layers as in the Deep Network Designer. I tried using them but not able to provide the 'ImageInputLayer' to the fitnet Network.
critic = rlQValueRepresentation(net1,observationInfo,actionInfo,'Observation',{net1.inputs{1}.name});
I tried above also but still not working.
Anybody tried this method or any suggestion for a work around?

Best Answer

  • Hi Abhay,
    Reinforcement Learning Toolbox currently supports the layers supported by Deep Learning Toolbox only. You could try converting your network using Deep Learning Toolbox layers after extracting the trained parameters.