T_train is defined as the number of games an algorithm needs to play in order to reach 80% of its final performance according to both M_opt and M_rand.
The final performance of M_opt is negative however, so what constitutes 80% of it's final performance? I am currently doing the following:
80% of final M_rand is: M_rand_final * 0.8 ==> 0.96 * 0.8
80% of final M_opt is: (1+M_opt_final) * 0.8 ==> (1 - 0.79) * 0.8
Is this correct?