Performance in a midsession reversal task with different probabilities of reinforcement