As a followup to this question, I'm interested in what the typical "Hello World" problem (first easy example problem) is for unsupervised learning.

A quick Google search didn't find any obvious answers for me.

  • $\begingroup$ Probably knmeans clustering or gaussian clustering $\endgroup$ – FourierFlux Sep 14 at 6:32
  • 1
    $\begingroup$ @FourierFlux Both seem to be solution methods, not problems. Do I understand correctly that problems would be: cluster $n$ training observations in a $d$-dimentsional space, then apply those clusters to new observations? $\endgroup$ – Christian Aichinger Sep 14 at 6:45

I disagree with the context that MNIST is the "hello world" of supervised learning. It is definitely, though, the "hello world" of image classification, which is a very specific sub-field of supervised learning.

I'd consider the Iris dataset a better candidate for the "hello world" of supervised learning, with other close candidates such as the Wine, Wisconsin breast cancer or Pima indians datasets. However, as an even simpler and more alternative choice, a lot of people prefer generating their own 2-dimensional datasets so that can more intuitively understand what the different algorithms are doing. An example of this is TensorFlow playground.

Equivalently, in unsupervised learning there are a lot of different tasks. I personally think that clustering is probably the task that is easier for people to understand and as such the most common intro to unsupervised learning. Here there are, as well, two options:

  • Using an already established dataset, e.g. Iris (without the labels).
  • Generating your own synthetic 2-dimensional data, to better understand how the algorithms work. An example is this.
| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.