Which of the following is FALSE about the training process of using policy gradients?

Group of answer choices

No Answer

The normalization of the discounted rewards will fit them all in the range from $-1$ to $1.$

After played the game for some episodes, then the gradients will be used to update each trainable parameters.

The normalization is using the average and standard deviation across all discounted rewards for all episodes in each iteration.

Let the model play the game for some episodes to compute the gradients and rewards, but don't apply any update during this step.