Azzera filtri
Azzera filtri

Applying SHAP on a Reinforcement Learning Algorithm

14 visualizzazioni (ultimi 30 giorni)
I am trying to Apply SHAP on a reinforcement Learning Algorithm and I am not sure if MATLAB has the required SHAP packages such as shap.DeepExplainer() which is a python package.
If anyone has any further information on how to apply SHAP on the Neural Network agent of a Reinforcement Learning model, please let me know.
  6 Commenti
Mahsa Raeisinezhad
Mahsa Raeisinezhad il 25 Nov 2023
function myAct_ = predict_01(tbl, env)
obsInfo = getObservationInfo(env);
actInfo = getActionInfo(env);
agent = rlPPOAgent(obsInfo,actInfo);
actor = getActor(agent);
actorNet = getModel(actor);
% Get the i-th slice of transposedArray
dataArray = table2array(tbl);
dlObservationsF = dlarray(dataArray, 'C');
% dlObservationsF = dlarray(dataArray(i,:), 'C');
% Predict using the actor network
myAct = predict(actorNet, dlObservationsF);
myAct_ = extractdata(myAct);
myAct_ = max(myAct);
end
Mahsa Raeisinezhad
Mahsa Raeisinezhad il 25 Nov 2023
The above is how tried to use SHAP in Matlab, I created a function handle of my Neural Network (agent) predictions using the Environment and applied SHAP on each decision, but I still highly recommend transfering to Python and using Python shap packages.

Accedi per commentare.

Risposte (2)

Mahsa Raeisinezhad
Mahsa Raeisinezhad il 23 Nov 2023
I decided to transfer everything in Python and use python packages. I used ONNX and Tensorflow for transferring everything. Hopefully if I have time in the future I write my own code to create same outcomes in Matlab.

Ive J
Ive J il 23 Nov 2023
Document and refs are clear enough if you're aware enough what your intentions are. Follow this example if you're interested to still stick to MATLAB.

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by