- June 2017
- May 2017
- April 2017
- March 2017
- February 2017
- January 2017
- December 2016
- September 2016
- July 2016
- June 2016
- March 2016
- February 2016
- December 2015
- May 2013
- March 2013
- December 2012
- July 2012
- June 2012
- May 2012
- April 2012
- March 2012
- February 2012
- January 2012
- December 2011
- November 2011
- October 2011
- September 2011
- June 2011
- March 2011
- February 2011
- January 2011
- December 2010
- November 2010
- October 2010
- September 2010
- August 2010
- July 2010
- May 2010
- April 2010
- March 2010
- February 2010
- December 2009
- October 2009
- August 2009
- July 2009
- June 2009
- May 2009
- April 2009
- March 2009
- February 2009
- December 2008
- November 2008
- October 2008
- August 2008
- July 2008
- June 2008
- May 2008
- April 2008
- March 2008
- February 2008
- January 2008
- December 2007
- November 2007
- August 2007
- July 2007
- April 2007
- March 2007
- February 2007
- January 2007
- 4 hour body
- American Express Global Business Travel
- animal communication
- Anne McCaffrey
- body fat
- book reviews
- clicker training
- Doubting River
- fat loss
- general news
- genre conventions
- house for sale
- jack kruse
- jury duty
- leptin reset
- Leslie Peeples
- low carb
- monthly challenge
- monthly results
- New Orleans
- new year's resolutions
- paula deen
- plotters guide
- Rainbow Bridge
- San Miguel de Allende
- Slow Carb
- social media
- story elements
- writing projects
- year in review
- year-long challenge
Tag Archives: training
This entry grew from a discussion on one of the clicker training mailing lists I frequent. The subject was whether one can lure and shape at the same time, or if they were two separate (and exclusive) techniques.
Shaping is “successive approximation.”
Usually when shaping is discussed, people associate it with an entire behavior taught through successive approximation, start to finish. For example, to shape a spin, you might first click for a glance to the right. Then a slight head turn. Then a weight shift. And so on until you have the complete spin.
Shaping can also be used to teach just certain parts of a behavior. For example, say you taught a sit through luring. Now you’re ready to add duration. First one second. Then two. Then four. That’s using shaping to add duration.
Shaping — successive approximation — *can* be combined with other methods, however. Let’s look in detail at a specific example: a fold-back down.
The trainer places the treat between Dog A’s front paws. He ducks his head down and back to get it, and within a moment, drops his entire body. Click! That’s NOT shaping. That’s pure luring. The lure was used to get the *complete* behavior.
The trainer places the treat between Dog B’s front paws. He ducks his head down and back to get it, but unlike Dog A, he doesn’t drop his entire body.
There was a dog like this in River’s class this week. These were not clicker trainers, and they could not get the dog to down. I had to chew on my fist to keep from interrupting. Not my class, not my class, not my class….
So the trainer MIXES luring and shaping: She puts the treat between his front paws. He ducks his head down and back — click and reinforce. Next rep he shifts his weight back — click and reinforce. Next rep she waits for the elbows to bend — click and reinforce. And so on. Each time she lured, but also she worked with successive approximations.
It’s precisely because successive approximation can be used in so many varied ways that I prefer to use the term “free-shaping” when talking about starting a behavior from the beginning and “shaping” when I’m referring to other uses of successive approximation.
One of my Facebook friends linked to a post I wrote on Karen Pryor’s site back in 2006. I reread it for the first time in a long time and thought it would be good to repost it here, because it’s just as true today as it was then. It is, perhaps, more true now, because there are are so many people drawing arbitrary lines in the sand over labels.
Earlier this month on the ClickerSolutions mailing list, a list member used the term “purely positive,” and another member asked what that meant. That began a lively discussion about the myths and misconceptions inherent in this term.
The meaning of “purely positive” tends to vary according to who is using it. Some clicker trainers use it as a sort of marketing tool, perhaps to indicate that they eschew corrections and attempt to stick with positive reinforcement as much as possible. Traditional trainers use the term as a slur, similarly to how clicker trainers use the terms “punishment trainers” or “pain trainers.”
How, you might ask, can “purely positive” be a slur? It sounds like a wonderful label! It would be, except for two minor complications: “Purely positive” does not exist, and the term is laden with mistaken, half-true, and untrue connotations.
First, the term implies that clicker trainers use no aversives. Extinction and negative punishment are both used by clicker trainers, and BOTH are aversive. Extinction is every bit as aversive as punishment, sometimes even more so. So even trainers who try to avoid negative punishment still have an aversive element to their training if they’re using extinction. All aversives are not created equal. Some are mild and some are severe. Whether the aversive is due to something being added, something being removed, or something just not paying off does not determine the severity of the consequence.
In the class Pax took in Nov/Dec, the instructor wanted to teach dogs to recall instantly, even if another person was playing with/distracting the dog. She taught this traditionally through collar corrections. She set the dogs up, and if they didn’t respond to the recall cue, the owner was to give the dogs a sharp correction. I taught it differently. I instructed people to pet my dog, and then one second after the recall cue… no matter what Pax did… to stand and turn away. We could even practice that without the dog!! There was no decision making there; they heard the cue, counted to one, and then turned away. If Pax chose to stay with the person when I called, he found that all the fun attention went away. No point in that! My solution was just as punishment-based as the instructor’s was, but there was no fear, pain or intimidation. Instead, the reinforcer the dog wanted (attention) was tied to his behavior.
Second, the term “purely positive” suggests that clicker trainers are permissive, that we just ignore unwanted behavior and pretend it doesn’t exist. That is blatantly untrue, at least with any trainer with any skill and knowledge. There are many, many ways to effect behavior. Clicker trainers eschew methods that rely on pain, fear, or intimidation. That still leaves a whole world of possibilities open to us.
Third, the term implies some black and white dichotomy that simply doesn’t exist. Training is a whole lot of gray. It’s incorrect to assume that because clicker trainers concentrate on positive solutions that all clicker trainers stick to only positive solutions. All clicker trainers are not of one mind. Each trainer has made his own decisions about what is and isn’t acceptable to him. Some use NRMs; some don’t. Some say “No” or make “buzzer” sounds; some don’t. Some use mild physical punishers like sprays of water or citronella or noise-related booby traps; some don’t. Some use negative reinforcement in various fashions; some don’t. Some use some of the above in real life but not in training.
As one list member eloquently noted, clicker trainers must stop using the term, because it is, due to these misconceptions, hurting our credibility. As she said, “No one trains by positive reinforcement alone. No one always, in all their dealings with a dog, avoids all possible aversive experiences. Minimizing them is one thing. Not directly employing them to instruct is one thing. But implying there is some kind of purity turns this into a religion, and a pissing contest, consciously or not.”
Before you start training a specific behavior, you need to know, in detail, what the finished behavior will look like. The more detail you put into that description, the easier it is to plan the road map for getting there. Some questions to ask include:
- What will the finished behavior look like?
You must be able to picture the behavior in perfect, precise detail. Don’t just focus on the obvious. Think about each part of the dog’s body — what must it be doing during each part of the behavior? Want a dog to win the heart of the judge? Include a wagging tail and pricked ears as requirements of the behavior. In clicker training, it’s all possible! By the way, don’t forget the dog’s mouth. So often people ask me how to stop a dog from whining or barking during the behavior. If silence is part of the behavior, plan it, and train it from the start!
- How will this behavior be cued?
Verbally? Physically? Environmentally? A combination? Remember that part of teaching a cue is making sure that only the cues you want become lasting cues — and that dogs are master discriminators. Include plenty of time for generalizing the behavior.
- What kind of latency is required?
Latency is speed of response — the time that elapses between the cue and the behavior. Zero latency is an immediate response. Fast latency is habitual, meaning if you train it for some behaviors, the dog will likely adopt it for all behaviors.
- Does this behavior have duration? Distance?
How long should the behavior last? If there’s a specific time requirement, plan to train fifty percent beyond that. For example, if you need a two minute sit-stay for competition obedience, plan to train at least a three minute sit-stay.
Distance should be trained similarly. Distance includes behaviors where the dog is sent to work at a distance, behaviors where the dog must respond to a cue when he is at a distance from the owner, and behaviors where the dog must maintain a behavior even when the owner moves away from him. Distance is challenging because the further the handler is from the dog, the stronger environmental stimuli become.
- Does your dog have to be in a particular place relative to you to perform this behavior?
Should the dog always be in front of you or perhaps always within a certain radius of you? If not — and especially if you specifically don’t want the dog to restrict his position relative to you — you should plan on spending time generalizing this element.
- Are you always going to be sitting, standing, or lying down when you give the cue?
Again, this is a generalization issue. Your body position can easily become a secondary cue for the behavior. This may work for you in competition heeling, but it can sabotage you for a household sit.
- In what locations will the behavior be cued?
Steve White trains every behavior in twenty different locations to ensure that his police dogs truly generalize their behaviors. You may not need quite that much generalization. For some behaviors, you don’t need any!
- What distractions might the dog face in those locations when performing the behavior?
List them, rank them, train them.
- How reliable does this behavior have to be?
Reliability is a number. You may need only 9 out of 10, or you may need 99 out 100 — or 999 out of 1000.
The definition of the behavior is a detailed description of where you want to go. The second step is to evaluate where you currently are. If this is a brand new behavior, that’s easy! You’re starting from scratch. If this is an in-progress behavior, evaluate the behavior for all of the above criteria. Keep records and let the data tell you exactly what your dog is capable of doing reliably.
The final step is to make a plan to get from where you are to where you want to be. Start with the behavior. Break it into responses, and shape it to perfection. When it’s exactly right, add the cue. Then one by one add elements like duration, distance, and distractions.
As you train, keep your training plans firmly in mind. Track your progress. Periodically review your training plan, and revise the definition of the final behavior, if necessary. Don’t stop working on the behavior until the behavior your dog performs is a reliable mirror image of the behavior you described.
Here is the initial training plan I put together for “Sit.” It isn’t complete – the steps here don’t give exactly the behavior described – but it’s well on its way. Note that the steps in the training plan are mid-level goals. It may require many, many micro-level steps to get from one mid-level goal to the next.
Cue: Verbal “sit”, hand signal, single blast on whistle
Description: On cue, dog will immediately drop into a competition-quality sit, no matter where he is in relation to me, and remain in the sit until released.
- Behavior specifics: Tucked, square.
- Duration: Up to five minutes.
- Distance: Respond to cue up to 400 yards away.
- Latency: Immediate.
- Position: Assume position from stand or while moving. Dog should turn to face me, except in specific situations where an alternative is specifically trained. I should be able to be in any physical position.
- Locations: Everywhere.
- Distractions: Anything and everything. Especially distractions common in a dog show or performance environment. Must maintain the sit when being touched by strangers or sniffed by strange dogs. Must hold the sit even if another dog is sent for a retrieve right next to him.
Other: Must maintain even when I’m out of sight.
- Capture the behavior
- Shape for tucked and square
- Add the verbal cue
- Add duration (up to 30 seconds)
- Generalize my position
- Add hand signal
- Add distance (up to 20 ft, 1 minute away)
- Handler goes out of sight for up to 15 seconds
- Distractions: dropped toy, tossed toy, rolling toy, jumping, running, door/gate, dropped food, dogs passing
- Take into field – all prior criteria
- Sit at distance (up to 20 ft)
- Add whistle cue
- Sit while moving
- Combine whistle-sit at distance with go out
- Increase duration (3 minutes)
- Handler goes out of sight for up to 3 minutes
- Handler goes up to 50 yards away
- Dog responds to cue from up to 50 yards away
Now to do this for every behavior I want to teach…. (See why I’m tired?)
A friend of mine on Facebook is having his dog snake-proofed today. The dog works in the Oregon wilderness where there are rattlesnakes, and snake-proofing ensures that the dog, upon smelling or seeing one, will go in the other direction.
The training is done with an electric collar. The collar is put on the dog, and the dog is taken on a walk where he will encounter a snake. If he goes to investigate — and my friend is certain that’s exactly what his dog will do — he is zapped with the collar. The process is repeated until the dog avoids the area as soon as he smells the snake. It’s classic avoidance training.
Okay, for the record, I support my friend’s choice. Working with rattlesnakes is potentially lethal for both handler and dog. Although a positive trainer has suggested a non-e-collar-based solution, that solution has not been tested. If it fails, the dog could die. Who wants to volunteer their dog? Hands?
I’m lucky. I live in western Washington state, and the only snake we have is the garter snake. I don’t have to make this choice. There are people in some states who find poisonous snakes in their back yards. For them, this training is much less optional.
One of the commenters to my friend’s Facebook blurb tried to assure my friend that e-collars don’t really hurt. He uses them on his dog, and he tried it on himself first. It’s just the surprise, not pain.
To which I say: balderdash. (I’d say something different, but there are children reading.)
First, lots of people do exactly what this person did: Before putting the collar on their dogs, they try it on themselves. (Around their arms. For some reason, no one ever seems to want to strap it to their necks.) Their experiences vary. Some people have to go up several levels before the collar becomes truly unpleasant. Others cannot tolerate it on the lowest settings. Dogs are the SAME way. Some clearly find it aversive even at very low levels. Some, though, don’t register even mild discomfort until several levels up.
Regardless, this is the takeaway: Your experience is not your dog’s experience. You do not get to say what is and isn’t aversive to your dog.
(Folks, I have a mouth full of cavaties, I’m sorry to say. All but one in the last ten years was filled without the use of Novocaine or other pain killer. The pain just doesn’t bother me that much. So… should the dental industry base their pain-management on my experience? I’m betting not many people would go for that. People vary. Dogs vary.)
Second, let’s be real about what an aversive is and how it works. An aversive is a stimulus that suppresses behavior. It has to be strong enough to suppress any natural desire that’s encouraging the animal to do something different. It’s highly unlikely that “surprise” would be enough to stop a dog, more or less permanently — or at least for a good long time — from doing something he really truly wants to do.
Let’s take an example of a field dog — a hunting dog — being trained for retriever field trials. A common “factor” that these dogs have to face is brush filled with thorns. However, they are supposed to persist THROUGH it, not cheat around it. (I think it’s nuts, but they didn’t ask my opinion.) The field trainer sets the dog up on a line through some brush. If the dog veers around it, he is zapped with the e-collar.
The stimulus of that e-collar has to be strong enough that, with just a few applications, the dog would rather run through blackberries than veer around them.
That ain’t surprise.
The snake breaking ain’t surprise.
I’m not going to say that the collar is, simply because of what it does, a horrible cruel thing. I don’t like them, and I don’t plan to ever use one. BUT, as I said above, I haven’t been put in a situation where my dog’s life might rely on it either.
I just want you to understand this: If the collar is effective at stopping behavior, chances are the dog isn’t having a good time. Don’t whitewash that. You may decide that the payoff is worth the choice, but don’t whitewash how the collar works. “Discomfort” and “surprise” don’t suppress behavior in a high drive dog. Period.
I’ve logged back on to some key clicker lists to reacquaint myself with clicker techniques and theory before River comes home. (One week!) I quickly found myself in a situation that reminded me why I took a break from the lists in the first place. A trainer — a good one, I might add — had a dog who had a problem with his recall in the field and mentioned instinctive drift. I said the problem wasn’t instinctive drift, and explained why. This trainer locked on to the idea that I was saying training a highly reliable recall in the field is easy. I never said nor implied that.
Getting behavior is relatively easy. Depends on the behavior and the animal and the skill of the trainer, of course. But in general, getting behavior is relatively easy. Let me be specific: Getting a dog to recall, on cue, between me and my husband in my house is easy. Getting a dog to recall, reliably, from a distance of 400 yards, in a field with all sorts of distractions, while carrying a freshly shot bird that the dog REALLY wants to keep is HARD. Incredibly hard.
(But the difficulty level still has nothing to do with instinctive drift. Just sayin’.)
Training is based on the principle that reinforcing operant behavior makes it more likely to be repeated. Simple example:
Getting a dog to sit for his food bowl is easy. Every time you feed the dog a meal, you’ll be reinforcing that behavior. That means that every successful rep will make it more likely that the dog will sit next time you put down the food bowl.
Does that mean that you could take the dog into Disneyland and expect him to plop his butt down for his food bowl? Nope. When you go to Disneyland, you’ve added a whole new element: competing reinforcers (and punishers).
In a lab (or a home!), it’s pretty easy for a trainer to control the reinforcers and punishers because they control the environment. In the “real world,” trainers have a LOT less control over what’s happening, which means there are other factors influencing the dog’s behavior.
If you took that dog to Disneyland, he would be influenced by the adrenaline coursing through his system, by the strange scents, by children running and yelling, and by the unusual sights — like someone dressed up as a giant mouse! Unless the dog is starving, he won’t give a flip about his food bowl, nor about your sit cue. Frankly, he probably won’t even HEAR your sit cue.
So does that mean you can’t get a dog to sit for his food bowl in Disneyland? Of course not. It means you have to break down the problem, and address the factors systematically and gradually. It is HARD WORK to do this, particularly when you can’t control all of the factors during training and can’t predict all of the factors.
Training isn’t easy. Getting competition-quality behaviors, consistently and reliably, in a competition venue is VERY hard. Getting those same behaviors at great distance is harder still. I am impressed to no end by trainers who do accomplish this — but there are trainers out there proving every day that it CAN be done. I’m even more impressed when they do it without relying on aversives.