beta distribution in PPO
조회 수: 1 (최근 30일)
이전 댓글 표시
I want to confine the actions of my PPO algorithm and I was thinking whether or not I can implement beta distribution for my PPO algorithm to confine my action space somehow.
heres the script of networks i am using
----------
commonPath = [
featureInputLayer(prod(obsInfo.Dimension),Name="comPathIn")
fullyConnectedLayer(120)
tanhLayer
fullyConnectedLayer(1,Name="comPathOut")
];
% Define mean value path
meanPath = [
fullyConnectedLayer(64,Name="meanPathIn")
tanhLayer
fullyConnectedLayer(64,Name="fc_2")
tanhLayer
fullyConnectedLayer(prod(actInfo.Dimension))
leakyReluLayer(0.1,Name="meanPathOut")
];
% Define standard deviation path
sdevPath = [
fullyConnectedLayer(64,"Name","stdPathIn")
tanhLayer
fullyConnectedLayer(64)
tanhLayer
fullyConnectedLayer(prod(actInfo.Dimension));
softmaxLayer(Name="stdPathOut")
];
% Add layers to layerGraph object
actorNet = layerGraph(commonPath);
actorNet = addLayers(actorNet,meanPath);
actorNet = addLayers(actorNet,sdevPath);
% Connect paths
actorNet = connectLayers(actorNet,"comPathOut","meanPathIn/in");
actorNet = connectLayers(actorNet,"comPathOut","stdPathIn/in");
actorNetwork = dlnetwork(actorNet);
댓글 수: 1
Kautuk Raj
2024년 2월 15일
To implement a Beta distribution for the action outputs in the PPO algorithm, I think we would need to modify the network architecture to output the parameters (alpha and beta) of the Beta distribution. These parameters must be positive, so one would typically use an activation function that ensures positivity, such as the softplus function.
답변 (0개)
참고 항목
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!