Question about reinforcement-learning-with-tensorflow HOT 3 CLOSED

Hi, think in the following way may help. If we have a positive value for q target, the q table will also add some value at the end. So the values in q table will somehow explode.

from reinforcement-learning-with-tensorflow.

Do you mean large positive values or just positive values. If the values are between 0 to 1, then it may not.
But later I saw that you confirmed to Qdash(S,A) - Q(S,A).

In the very example q_table.ix[S, A] += ALPHA * (q_target) is working well and converges faster. Would be interesting to understand when you could end up in blow out.

Thanks for responding to the question though.

from reinforcement-learning-with-tensorflow.

It will show the right behaviours in this example, but it will never coverge. Actrally, no matter the value's sign, any value will give you an unconverged but right behaviour result.

If you keep running the script in your way, you will find your q table will exceed its capacity to hold one values, it may show NaN at the end

from reinforcement-learning-with-tensorflow.