When giving a dog a treat reward as part of training, what is more effective? A few large rewards or a lot of small rewards?
I’ve heard some people say that numerous small rewards is better for reinforcing good behavior, but if the total amount of ‘treat’ is the same in both instances does it really matter to the dog?
A small treat for every positive action, or a large treat for a few select positive actions?
I think rewarding each positive action might be more beneficial.
Large treats or medium-sized treats will fill up your dog faster than small treats. The dog may stop responding before you are ready to stop training.
Small chunks, the size of a pencil eraser (does anyone remember those?), will keep your dog interested in training for longer periods of time.
I look for “super treats” that the dog will really over-respond to. Liver treats, bits of hotdog, bits of chicken. It’s the dogs choice.
As I understand it, dogs have a fairly short memory, so waiting until the end of a training session and giving them a huge treat wouldn’t be as good as ‘paying them’ frequently when they respond correctly to a command, but I have found with my dog that if I reward him frequently he become too focused on the treat and doesn’t pay as much attention to me, so I’m trying to find a middle ground.
I don’t want my dog to learn that every time he does something good he should expect to get a treat so I sometimes just pet him or give him words of encouragement instead, but when I am working with him off leash I sometimes need to reward frequently to keep him from wandering off.
Sometimes he will do something really amazing and I want to give him a larger reward, but I don’t think he really cares how much reward he gets, but whether he gets a treat or not.
Those sniffer dogs that the police and customs use don’t get food, all they get is their favourite ball to chase.
Small, frequent but intermittent treats seem to work best in my experience. Pair each “treat” with a verbal reward and use the verbal reward all the time,the bit of liver 80% of the time.
Tiny, frequent treats, the dogs don’t care about the size, they just want a treat, and they enjoy the game.
Also, try and give them something healthy. I buy cheap fresh chicken and cook it at home - my dog loves it more than treats and it’s much better for him.
Dogs will very quickly reach the point where they don’t even care about the treat at all, for its own sake, but rather appreciate it as a token of the fact that you’re happy with them. Any size treat will work equally well for this purpose.
Dogs are social critters and will work their asses off just for the reward of their Master’s approval.
The same is known to be true of captive dolphins. I’ve seen and participated in this myself. True story: We thawed a pre-determined amount of fish, then ran the training sessions until the fish ran out. When we were down to the last fish, and gave that to the dolphins (there were two dolphins), they wouldn’t eat it. Instead, they swam around the tank for a while then came back and handed the fish back to the trainer. They continued to work, and the trainer continued to give the dolphin the same fish over and over, until the fish was totally in shreds. We never taught them to do that – they just did it of their own initiative.
This was not totally surprising. In the dolphin-training biz, there were numerous stories of dolphins working solely for the apparent reward of their trainers continuing to work with them.
Consider clicker training as well. Basically, a “click” is paired with the reward, allowing you to eventually use the click as a reward by itself.
Another example: When my mom is training her dog, the “treats” she uses are just pieces of his ordinary kibble. He gets to eat whole bowlfuls of the stuff for free, and usually isn’t even in too big of a hurry to do that, but he loves being handed them one at a time after coming when called, or sitting on command, or whatever.
In the animal training biz, there are whole bodies of theory and practice about how to do it most effectively, mostly built on the psychological theories of behaviorism.
These theories deal with ways to shape animals’ behavior, and methods to reinforce desired behaviors and discourage unwanted behaviors.
There are various specific schedules of reinforcement that may be used, perhaps at various stages in the training process. It has been found that continuous reinforcement (that is, giving some kind of treat after every instance of the desired behavior) isn’t the most effective.
Instead, the most effective reinforcement schedule seems to be random interruped reinforcement, or RIR, in which the reinforcement for each individual action is or isn’t given according to some quasi-random schedule that the animal could not learn to predict. The ratio of reinforced acts to non-reinforced acts is established and the quasi-random choice to reinforce each individual act is made (approximately) according to the desired probability.
This is known to be effective for teaching humans too. One extreme example commonly cited is gambling behavior in which the gambling acts are reinforced (that is, the gambler wins) rather infrequently (and not necessarily winning a large amount most of the time). Note how seriously “addicted” (some susceptible) people can become to performing those gambling behaviors, for only occasional rewards.
Wiki cite: Animal training.
See also (if you can find it in a library or somewhere) Pryor, Karen. Lads Before the Wind (apparently now in its 4th expanded edition, 2004, out in paperback); originally pub. 1975). TL;DR: The author was a founder and original head trainer at Sea Life Park, Honolulu; she writes of her adventures in dolphin training there.
This is also very visible in those dolphin shows at places like Sea World e.g., where a whistle is commonly used.
In the early stages of training, the whistle is paired with the reward. The animal learns that the whistle means “reward is coming!”. As training advances, the actual reward (fish!) is given less often, according to some RIR schedule. The whistle itself becomes the reward.
In dolphin training, from all that I’ve seen, the fish is given almost every time. At least, so it seems from the public exhibition shows I’ve seen. I wonder if this is just “optics” done for public “consumption”. (Good audience, fish for you!) Pryor tells of working with dolphins in open-ocean research, where giving fish is inconvenient and dangerous (attracts sharks)!
Compare with horse training. I’ve never seen food used as a reward for that. Horses can be trained to do very detailed and elaborate work (see: Dressage ) which is nothing less than full-scale horse ballet. I’m not aware that food rewards are ever used for horse training of any sort. And come to think of it, I don’t recall ever seeing any kind of “clicker” or target training used with horses either – but to be sure, I’m not really very familiar with horse training.
To clarify, constant reinforcement will get results quicker, but the results are also much less permanent. That is to say, if you reward a dog every time he does a trick, then as soon as you stop rewarding him, he’ll stop doing the trick. But if you only sometimes reward him, then when you stop rewarding him, he’ll just figure “this must be one of those times I don’t get the reward”.
EDIT: Sugar cubes are used as rewards in horse-training. They might transition away from them fairly quickly, though, just like with dogs and dolphins.
At our dolphin lab, we had a resident dog (actually, belonged to one of the directors) who could do a few simple tricks, including “Stay!”.
We’d tell him to “Stay” and put a kibble on top of the end of his nose. He’d sit there, motionless, staring cross-eyes at the kibble, until someone said “Okay!” and then he’s snatch it right out of the air.
For the animals, such training can become very rote, with the animal doing some very specific movement even if the context calls for something else. Or, with “de-contextualized” training, the animal could be trained to actually “think about” what’s going on, and adjust its behavior accordingly.
One day, just to see what would happen (although I already was pretty sure), I told the dog (who’s name was Ehu) to “Stay!” and then put a kibble on his foot. Sure enough, he sat there motionless, staring cross-eyes at the kibble on his nose that wasn’t there. When I said “Okay!” he snatched it right out of the air like always, and got . . . nothing. Then he looked around and found it on his foot.
ETA: Photo of the dolphins. (No, that’s not me in the background.)
Typically, I think the plan would be to use constant reinforcement in the early stages of training, with the results like Chronos says here, and then move on to RIR, making the reinforcement less and less frequent gradually.
Another indispensable training technique is “target training”.
See this photo, from kayaker’s link to Karen Pryor’s web site. It shows a stick with a plastic ball on the end, and a dog with his nose glued to that ball.
Train an animal to glue his nose to a target like this, and you can then use that to lead the animal through a lot of complex behaviors. It’s definitely a standard technique in dolphin training, and sea lion training too.
I volunteered for a while at the California Marine Mammal Center near Sausalito, where we had an adule male California sea lion who was epileptic. Of course, he could never be rehabilitated and released. Some of the staff had trained him (sort-of) to follow a target. It was possible to take him out of his pen from time to time and take him for a walk around the premises by having him follow the target.
We kept him doped up on dilantin to control his seizures. One of the volunteers was a nurse at a local hospital, and she managed to get all the supplies of expired dilantin that they would otherwise have had to toss. The sea lion’s name was Zonker, presumably after the Doonesbury character (this being in the early 1980’s).
ETA: Eventually, it didn’t end well. Some time after I left, Zonker was placed in a mini-zoo-like place (Children’s Fairyland, at Lake Merritt in Oakland) where, I heard years later, he eventually had a seizure and drowned.
If you want to see RIR in action, watch someone pulling the slot machine’s handle hour after hour.
Seen. See post #12 above.
BTW, see also Melissa Shyan-Norwalt, certified animal behaviorist. She was a grad student at the dolphin lab while I was there, and has gone on to become a certified animal problem solver.
