Tuesday, May 3, 2011

Clicker Expo 2011 (Chicago): Kay Laurence- Raising Criteria

I really, really enjoyed Kay Laurence's session on raising criteria. One of the most difficult parts of shaping is knowing when, and how much, to raise criteria. Or at least, I think so. At any rate, there is a definite art to it, and I loved seeing Kay's take on it.

Although each presenter used slightly different words, the theme of “think-plan-do” recurred over and over at Clicker Expo, so it should come as no surprise that Kay emphasized the importance of planning your sessions in advance. Kay explained that planning helps you develop clear criteria. If you don't know exactly what you want, how is your dog supposed to figure it out? At best, unclear criteria slows down learning. At worst, it creates messy, inconsistent behaviors.

Therefore, you need to decide what your criteria is, and then click only the behaviors which meet it. If in doubt, don't click. It is more confusing for the dog if you click several variations of the same behavior than if you don't click several perfect versions. Of course, this is easier said than done. I know I often have the tendency to want to help my dog by clicking good tries, but Kay said that it is nearly impossible to tidy up messy behaviors later on. As a result, it becomes clear that knowing when to withhold the click is as much of a skill as knowing when to click.

So choosing your criteria and sticking with it is important, but it's also important to make sure that you are picking small, easily achievable steps. Kay said that a clicker-savvy dog should be able to figure out what the criteria is in two clicks. If it takes longer, your criteria is probably too high. Although you want quality from the outset, you can't set the bar so high that the dog can't find it. You will get further with lots of quick, small steps than with fewer, bigger steps.

As for how quickly those steps should be taken, Kay recommended increasing the criteria once your dog is able to do five repetitions of the behavior without error and without hesitation. It is important to get five solid reps; raising criteria too quickly means that you're building the behavior on a shaky foundation.

While raising criteria too fast, too soon is more problematic than too slow, a failure to raise criteria at the correct time can be detrimental, too. If you stick at a lower level too long, you run the risk of “locking in” that level of behavior, and making it harder to raise criteria as a result.

Kay believes that you will make the most progress if you take frequent, short breaks. She has found that alternating 20 clicks and treats with a 10-20 second break ten times will yield more progress than 200 clicks and treats in a row. This is partially because it gives you a chance to quickly evaluate how your dog is performing. It's much harder to make this decision on the fly, while training, so she advised us to take a break in order to have time to think.

Another thing to consider during your breaks is your location. Kay said that dogs are very geographically aware, and recommended playing with the environment as part of the criteria. Change the direction you're facing, or introduce small objects as distractions. Use props that help your dog be successful and fade them out later on. Explore how these things affect your dog's behavior.

Breaks can also help you evaluate if your treats are helping or hurting your progress. Tossing a dark colored treat on a dark colored surface is going to slow your dog down. So is throwing treats that break into little pieces when they land, causing the dog to sniff around to find them all. You might also find that your dog is anticipating where or how you're going to give him the treat, so you might need to vary the delivery. If you're tossing treats to reset the behavior, randomly alternate which direction you're throwing it, as well as if you're throwing it overhanded, underhanded, or like a frisbee.

On the subject of tossing treats to reset the behavior, Kay said that it's okay to toss a treat even if you withheld a click. Tossing the treat lets them know they were on the right track, but didn't quite meet the criteria. As odd as it sounds, she said dogs really do seem to understand that the absence of a click means they need to try harder next time.

Like Cecilie Koste, Kay said you should make breaks clear to the dog. Either you're working with him or you're not. Do your best to avoid confusing him. Although she didn't recommend using stations, she did say that when you're on a break, you shouldn't look at, talk to, or otherwise engage your dog.

Overall, I really enjoyed Kay's session. It was a lab, which meant I got to see her interact with a wide variety of dogs. She often chose criteria points that I wouldn't have thought of, which was interesting. The biggest thing is that I tend to shape forwards, towards something, and she seemed to be shaping backwards. For example, when teaching a dog to go around a cone, I would probably start by clicking orientation towards the cone, then steps toward the cone, etc. Kay started with trotting towards the trainer. Then she had the dog trot towards the trainer with the cone visible next to her. She gradually moved the cone into the dog's path so that the dog had to choose to go around it to get to her.

I also really appreciated her emphasis on quality; I'm not very good at being clear with my criteria, and often accept “close enough.” This, of course, prevents me from getting precise behaviors. Sometimes this is okay, but sometimes I would like more consistent results. Watching her train really helped me understand the importance of selecting a clear clickable moment, and I'm looking forward to implementing this in the future.

But what about you guys? How do you choose what your criteria will be? When do you know your dog is ready for the next step? How quickly do you move? What problems have you had, and what do you think you might do differently? Let me know!


Ninso said...

This is such an issue for me and I think a few points in this post will be hugely helpful for me! I struggle with breaking complex behaviors down into small enough criteria. I think that's why I struggle with teaching complex tricks, like limp. And I am right there with you, clicking to reward "good tries"! Good to know that I can treat without clicking for good tries. And I have always been of the "when in doubt, click" mindset, but I see that the opposite is more effective.

I have a recent example of the futility of trying to clean up behaviors when the criteria weren't clear enough to begin with. Elo has recently learned "wave." I did it all through free shaping. I knew I wanted two waves of the paw, but that was as specifically as I defined it, and it turned out not to be specific enough. I got my two waves easily enough. I worked from a shift of weight to a slight paw raise, to a higher raise, and eventually got my two waves. Well, at that point I realized there are multiple ways to wave twice. Sometimes they are nice and clear and high and crisp, which is what I want. Sometimes I get a more "reachy" wave, where he is stretching his paw out forwards. Sometimes he lifts his other paw off the ground. I have clicked all of these at some point or another. And when I withheld clicks, naturally he tried "harder" and I got more reachy-ness, more waves, more second-paw stuff. I didn't want him to try harder, I just wanted him to try differently!! Over the past few weeks (once I realized what I actually wanted), I have tried to clean it up, but haven't gotten very far. I finally decided "good enough" and started putting it on cue.

Crystal Thompson said...

Hooray! I'm always glad when something I've written is useful to someone!

Your example of trying to clean up messy behaviors is REALLY good. Not only have I struggled to clean up behaviors (especially those with superstitious behaviors attached, like the damn paw lift when she retrieves, sigh), but I also imagine it's confusing for a dog to get clicked for several versions of the same thing. "Just WHAT do you want, lady??"

Kristine said...

Thanks for posting all of this. I have already seen several places where I need to clean up my act. I am definitely guilt of raising criteria much too fast. Sometimes I just want to see how far I can push. Last night, in fact, I was working on teaching her to close the door of her crate. She was doing so well, exceeding my earlier expectations, that I think I started expecting more than she was capable of and eventually we had to go all the way back to the beginning. It was frustrating and all entirely my fault.

I also admit that many of her behaviours are "messy". It doesn't bother me too much but I can see where it has created a few problems.

It looks like I have a lot of work to do!

Laura, Lance, and Vito said...

Really good post. I don't see myself becoming more analytical and planning of my training sessions but can make myself take more breaks. I have noticed that it helps sometimes when I put the dogs in a down while trying to think what I'm doing but I haven't purposefully planned to take little breaks within each behavior.

Really interesting about rewarding certain attempts but still witholding the click! I know there are many times the dogs get frustrated and in order to keep a high ROR I reward the next effort. I guess I need to be be more conscious of clicking vs just rewarding.

Crystal Thompson said...

Kristine- progress is addicting! It's fun to see how far our dogs can go. I totally get why people (myself included) might raise criteria too quickly.

Laura- I don't see you planning any more than you do, either (you've been pretty vocal about that), but taking breaks may be a good compromise. :)

I have started experimenting with tossing treats to reset/reward good attempts, but not clicking. No conclusions yet, but it's nice to have permission to try it!