Uses https://github.com/tehZevo/rnd for the RND implementation
PORT
: the port to listen onINPUT_SIZE
: input (observation) vector sizeOUTPUT_SIZE
: RND model output vector sizePREDICTOR_HIDDEN
: hidden layer sizes for the predictor network, separated by spaces e.g.,128 64
(defaults to32
)TARGET_HIDDEN
: as above, but for the target network, defaults to32 32
ACTIVATION
: hidden layer activation function for both models, defaults to "swish"LR
: learning rate, defaults to 1e-3BUFFER_SIZE
: training buffer size, defaults to 10000BATCH_SIZE
: training batch size, defaults to 32SAVE_PATH
: path to save/load models from, defaults tomodels
AUTOSAVE_STEPS
: save every steps, defaults to 1000