Making the impossible paw-ssible: Introducing data science for cats and dogs

Share on twitter
Share on facebook
Share on linkedin

At RapidMiner, our mission is to make data science accessible to anyone, no matter what their background or skill set. With the recent release of RapidMiner Go, we expanded our offerings to those who don’t have experience with machine learning. Since then, we’ve been thinking about how we can reach even more users with data science.

After working remotely for the past several weeks, we realized the answer was right under our noses: RapidMiner Studio’s visual coding interface is perfect for those of us without fingers to type. Today, we’re breaking the species barrier and making machine learning relevant, useful, and accessible for cats and dogs in addition to our human userbase.

Data science for cats and dogs

We’re working on a brand-new UI for Studio in the coming months that will make it even easier for our four-legged friends to interact with RapidMiner products. In the meantime, though, we worked with our pets (using a modified version of our AI Assessment) to identify high-impact use cases for them and then build models to provide them with insights now.

Here’s what we discovered.

Just like us, our pets want to be able to take what’s happened before and predict what’s likely to happen in the future. Humans and other animals have instincts to help them, but as we know, people are terrible natural statisticians, and our pets are no different in this regard. These are some of the most important use cases for both our dogs and cats.

Dog-centric use cases

From love to pets to walks, we explored a range of things that dogs care about most in an attempt to help them better forecast what’s coming.

When will my owner be home?

The sound of an engine, the jingle of keys, and the lingering scent of owners after they’ve left—how do we put these together to determine how much longer a dog has to languish at home, unloved and unpetted?

Using a combination of factors, we were able to build a simple decision tree model that was accurate in predicting owner return time to within about an hour. However, we’re going to continue to work on the model here, as the general consensus from the dogs in our lives was that, despite the model’s accuracy, the time predicted was “too long” and “unacceptable”.

Is it a good time to play with the cat?

After consulting with the dogs in our lives to build a dataset of features like cat posture and time since last attempt to play with the cat, we ran through several iterations of different model types. Fortunately, we discovered that we actually don’t need a model to answer this question—the answer is “yes”.

Walk?!

Dog owners are hard to understand. They do things like stand up from the couch. They walk in the general direction of the door. They say words that start with “w” or end with “k”. How’s a dog to know when it’s time for walkies?

To help understand inscrutable human behavior, we used a deep learning neural network. We also used natural language processing to score words that owners say for similarity to the word “walk” and included that in the model.

The initial results from this model are a bit murky, and we’re going to continue to iterate. The frequency of walks predicted seemed to be lower than expected.

Am I a good boy/girl?

Yes, absolutely—no model needed!

Cat-centric use cases

Cats can be fickle creatures, but that doesn’t mean we can’t use information about the past to help them predict the future.

Is it a good time to lay on my owner’s laptop?

We considered a number of factors to help cats decide if it was a good time to lay on a laptop. Although we initially hypothesized that ambient temperature and the temperature of the laptop would be the key predictors, it turned out that there was a single data point that was nearly deterministic in deciding if it was a good time for laptop lounging: Is my owner trying to work? If yes, laptop lounging is pretty much guaranteed.

Should this thing be on the floor?

Yep.

Is it a good time for the dog to play with me?

Nope.

In or out?

This one was more for the humans than the cats. We considered over one hundred different possible predictive data points to determine whether our cats wanted to be in or out, but no model we tried was able to predict this better than chance. We’ll continue working on this and report back.

Conclusion

Stay tuned for more from us in the coming months as we release new versions of Studio that are amenable to paw-control, letting our pets build their own models for the things they care most about, without relying on a human intermediary.

If you’d like help figuring out how models could be applied to the things you care about, sign up for a free AI Assessment, and we’ll walk you through how data science can have a strong impact on your business.

RapidMiner AI Assessment Matrix

Find out how ML can transform your business

We apply our expertise to help you identify the use cases you should tackle in your organization. The outcome is an impact-feasibility map that you can use with or without us.

Additional Reading

Chris Doty

Chris Doty

Chris Doty is RapidMiner's Content Marketing Manager and has worked on projects for companies like Google, NatGeo TV, the US Chamber of Commerce, and Virgin Pulse. In a past life he was an academic, and has a PhD in linguistics.