This indeed doesn't look right, but if you want a different trading behavior per train cycle for some reason, simply code it inside the algo. If you only want to train 4 algos separately, use a loop over all 4.