Thursday, March 1, 2012

Jane Killion Seminar: Shaping

I don’t think I have ever met a trainer with better shaping skills than Jane Killion. At times, it seemed like it’s the only tool she uses (and I’ll talk more about that in my posts on her approach to reactivity, aggression, and attention), but I’ve got to admit that if I was even half as good at shaping as she is, I might not feel the need to use other approaches. Honestly, her skills are just phenomenal.

So, what is shaping? Shaping is the process of teaching a dog a new behavior in small, incremental steps. Instead of trying to teach the entire behavior all at once, shaping builds it up, one piece at a time. Jane used the analogy of a movie: if the behavior is what you would see in the final video clip, then shaping is the process of teaching that behavior one frame at a time.

More specifically, Jane does something called free shaping, which is where the dog offers each successive behavior as opposed to the trainer prompting, luring, compelling, or showing the dog what to do. Jane makes this distinction because she wants the dog to be the active participant. Since a “Pigs Fly” dog tends to be an independent thinker, free shaping harnesses the dog’s natural ability.

It also creates a dog who is “obsessed” with solving problems. Jane believes that free shaping taps into the SEEKING circuit in the brain, and since we know that anticipation is reinforcing, this means that the dog who has learned via shaping will find solving training challenges to be inherently reinforcing. This makes shaping a valuable tool for the “Pigs Fly” dog who isn’t all that interested in pleasing his person.

If this sounds wonderful, it’s because it is. But there is a catch! For free shaping to work, the dog has to be operant. This means that the dog understands that his behavior will earn a click and treat, and even more importantly, that he actively offers behaviors in order to earn the reward.

Jane teaches a dog how to be operant by playing The Box Game. To play, set a box on the floor and click/treat any interaction with the box that your dog offers. Any interaction. You aren’t teaching a behavior, so you should click anything and everything your dog does with, near, or to the box. Dog looks at the box? Click Dog sniffs at it? Click. Dog paws it? Click. Dog steps in it? Click. Dog jumps over it? Click. And if your dog will have nothing to do with the box? Click him for looking away from it, or even for getting up and walking away. That may sound weird, but remember, the goal is not teaching a behavior, it’s to teach the dog the concept of offering behaviors.

Once your dog is eagerly offering behaviors, it’s time to start shaping a behavior. Jane had working participants shape a behavior related to the box. This allowed the handlers (and the dogs!) to practice their new-found skills with something that doesn’t really matter. Choosing a "silly" behavior is important, because when people start with behaviors they care about, they tend to get tense and worried, which interferes with their ability to train.

To shape a behavior, start by thinking about what you want to teach. Jane said that the key to training lies in setting up your sessions well. You should spend more time planning than training. Have a very clear picture of what you want, and be sure to state it in terms of what the dog will do, not what he won’t do. For example, a sit-stay is defined as “sitting and holding that position” as opposed to “sitting and not changing position.”

Next, figure out what each step will be. Again, think of the final behavior as a movie, and visualize each frame. The change between each step should be very small. For example, if the behavior is teaching your dog to go lie on his bed, you might say that the first step is for the dog to turn his head towards the bed. Jane, however, would say the first step is for the neck muscles to move, or for the dog's eyes to flick in the direction of the bed. Like I said: tiny steps.

Click/treat that first step repeatedly until it is clear that the dog is deliberately offering the behavior. The amount of time you spend on this step will depend on the dog; if he’s done a lot of shaping, you might only need to click each step two or three times, but if this is his first time out, you might need to click it a hundred times.

Now, stop clicking. If your dog understood that his behavior was resulting in reinforcement, he will probably be a little confused. He will likely try the behavior again a few times. When that doesn't produce a click, he will start to offer variations on the behavior, trying to figure out what it is you want. Pick one that takes you closer to your end goal and click/treat that. Technically, your dog is going through an extinction burst, so be quick, before the dog gives up entirely. While much of training is a science, there is definitely an art to determining when to stop clicking, and what to click after that.

A lot of the working dogs were already operant, having had experience with shaping. In fact, some of them were almost frantic in their efforts to offer clickable behaviors. Jane acknowledged that shaping can be messy and chaotic at times. This is often hard for people to accept, especially if they are used to dogs who wait to be told what to do, or who really want to control their dogs.

That said, there are problems unique to shaping, although they are almost always due to timing. (Of course, poor timing is a problem no matter how you train, it’s just that poor clicker timing results in a different set of problems than in a poorly timed collar correction.) Almost every clicker problem can be solved simply by clicking earlier. Barking out of frustration during training is usually a very smart, fast dog saying “I did it! Click me already!” Dogs who throw lots of behaviors usually are not sure what the trainer wants because of her poor timing. And superstitious behaviors- those things that the dog thinks is part of the behavior but really isn’t- usually creep in due to late timing.

Timing and splitting behaviors down into tiny pieces seem to go hand in hand. For example, when Jane was working on teaching a dog to jump, she started with clicking the dog for seeing/committing to the jump versus going over it. Not only does slicing the criteria like this make it easier to shape the dog, but it also means that if your click is even a half-second late, you’re probably still clicking the goal behavior- in this case, jumping- instead of the behavior after it, like landing.

It was pretty amazing to watch Jane work with the dogs because her timing was just so good, and the lesson of clicking early was definitely something I needed. My timing is pretty good, but it is nowhere near Jane’s level of pure awesomeness. I am going to work hard to click sooner and to break down behaviors into smaller pieces. I think both things will really benefit my ability to teach the behaviors I want.

Coming up soon: using shaping to create an attentive dog and solve behavior challenges.


Natasha said...

Sounds like a fantastic seminar!

Jen said...

My timing....needs work. So does my forming of criteria. So, if I have a specific thing in mind (loose leash walking, say) it's okay. Free shaping? Less so.

Another thing to work on!

Susanna said...

I think I'm beginning to understand our different impressions of Jane's seminar and her methods. :-) Travvy and I were already doing shaping and free-shaping in class. Our instructor is better at it than I'll ever be, and so is Jane, but that's because they're pros and they do it all the time with all kinds of dogs.

My skills aren't in their league, but they've come a long way from the days when I couldn't manage a leash, a clicker, and some treats at the same time and still pay attention to what my dog was doing.

My "Pigs Fly" malamute gets very engaged in figuring things out. When he gets frustrated, I know I need to do a better job of breaking a behavior down into smaller parts. And yeah, I take some shortcuts that Jane probably wouldn't (like more coaxing and even luring when teaching something new), but the methods do work for me and my Travvy. They keep us both engaged!

Crystal Thompson said...

My biggest problem with shaping is that I find it so fun that I never finish the stupid behavior. I get so caught up in the excitement of watching Maisy figure it out that I never bother with getting it on cue. I just move on to the next stupid pet trick. And trust me, I shape stupid stuff... Go touch that chair with your nose. Go paw the couch on the middle cushion. Nothing useful.

We have been working on shaping her new trick, and I've been making a conscious effort to click earlier. It seems to be working.

Ninso said...

The dumbest thing I ever shaped was having Jun stand underneath a chair, with all of her feet inside the 4 rungs that connected the legs. Hilarious!

I love shaping! I love operant dogs! Did Jane have anything to say about cross-over dogs and how easy or hard it is to teach them to be operant? I started Lok with L/R and he got more operant, but never got anywhere near Elo's level. Does Jane think any dog can be super operant, or is a lot of it based on the individual or prior training experiences?

Andreja said...

I read somewhere that the way to make timing of the click better is to throw a ball and click as it hits the floor. To do that you need to anticipate what will happen at the moment of click. You can't wait to actually see the ball hitting the floor and then tell your thumb to move.

Well, after doing that exercise my problem was that I was clicking too early. For example, I anticipated that Ruby will hit the contact during running contacts, but in fact he missed. Since dog's feet are on the contact for mere 1/10th of the second I didn't have time to actually see the contact and click him. I had to stop using clicker because my thumb would automatically go off when Ruby made that final stride and I often misjudged whether he will hit the contact or not.

So I would say early clicking works only if you can predict what is going to happen between the time that your eyes see dog's behavior and the time when clicker sounds off. For some behaviors and some dogs that is a very short time and not much can happen. For others it makes a big difference.

Laura, Lance, and Vito said...

I see your point but I think I disagree in most practical cases. I think for most people if they try to click early they will actually be on time and it is a rare occurrence for the dog to change behaviors in that split second. I think the bigger issue is in our overeagerness we see something that isn't there.
To use your example with running contacts, I can tell if my dog is going to hit the contact and where based on that striding before and if he is accelerating or not. Yes it happens in an extremely short amount of time but you can see that collection or extension on the stride before. That being said I didn't use a clicker with my running contacts, more so because I didn't want to be late.

With a different example of a dumbbell hold, there's been many times I clicked something that I shouldn't have. But I think it's more so me getting into a rhythm where I think the dog is going to do something based on past experience versus what the dog is actually doing.

Susanna said...

Our trainer encourages us to click the intent, especially when teaching something new. Say you're teaching directed jumping. You click the dog's glance toward the jump even if s/he doesn't step toward it.

Yeah, sometimes I misread what my dog is doing, but mostly it seems to work. I'm learning how to read Travvy better as he tries to figure out what I want, and really in the long run it doesn't seem to matter that I sometimes flub it.

Crystal Thompson said...

Ninso, Jane did say that "cross over" dogs (whether it's from a more P+ stance or from lure/reward, like you mention) can be difficult for some dogs. One of the things she said specifically was that some dogs will NOT interact with the box during the box game unless told, which is why she recommended clicking even for looking away/walking away from it.

As for your last question, I think she would say that any dog can become operant, but that it might be harder/take longer for some dogs. I'm not sure what she would say about HOW operant dogs can become, although I personally would agree that that will be informed by the dog's previous experiences.

Crystal Thompson said...

Susanna, your trainer sounds like a lot like Jane; Jane also talked about clicking the intent. :)

Newdrim said...

Laura, I didn't mean to imply that dog changes the behavior in split second, but that in my case the two behaviors - hitting or missing looked so much alike to me that I couldn't tell them apart until his feet were actually on the contact. After all, the difference between a hit or miss can be mere 2 inches when you start training. I am surprised that anyone can learn to predict dog's length of stride that well.

I think teaching a hold is a different problem because there delaying the click for a split second won't make it any more accurate than clicking early.

Joanna said...

I just started shaping a couple of silly but useful tricks, so that I could improve my shaping skills and so that my dog would have something impressive to show off. One is bringing me my slippers and dropping them next to my feet, facing the correct direction (this is going to be hard!), and the other is sliding a button at the base of my lamp to turn it on or off.