New answers tagged


Yes, I believe you can. Assume that you want to upper bound your difference to $k$. Use the following function: $$ y_{t}^{\pi} = \frac{k}{2}*\tanh(Q_{t}^{\pi}(s,a)) $$ Here, $y_{t}^{\pi} \in [-\frac{k}{2}, \frac{k}{2}]$. Hence the upper bound would be $k$. Checkout this tanh graph. A practical suggestion - Try to soften out the sharp edjes of tanh by using $$...


But, I am not sure if that is necessarily the best way to fix the problem of having repetitive sets of fixed action in the replay process. The best way to handle this is using a Multi Discrete Action Space. You don't need a neural network for every action on its own.

Top 50 recent answers are included