27 February 2009

Rapid reinforcement

Yesterday I read through some of Fanny's articles and noticed a section on high criteria - high frustration training. Since Orkan is the kind of dog that gets extremely frustrated by extinction I decided to try a session of really low criteria training with him, just to see what would happen. That was fun! I worked on heel position, and rewarded almost every go, except the really, really bad ones. I think the reward rate was at 90%, meaning I even rewarded the quite crappy attempts. He still improved, and it turned out to be a bit of a wake-up call for me, that I should probably count more of his attempts when I make my mental statistics on reinforcement rates.

When I help Sigurd with training Bris, I've told him a couple of times that he needs to count all behaviour as attempts, not just the ones that look like one. I guess I'm making the same mistake myself, I need to lower my criteria sometimes. There is such an enormous difference between training Storm and Orkan, in that Storms rate of offering behaviours is a lifetime lower. So when I let myself get stuck in rewarding every same number of seconds with the two of them (not literally, but still it is a patterning of what I consider rewarding often or seldom), I actually have a much lower reward rate with Orkan than with Storm. Not that I've ever done that conciously, but I think I've been missing out on how many attempts I have NOT rewarded, just keeping track of the ones I did reward...

Now that's too much thinking for my tired little brain right now - anyway we had a very nice session :-)