The CMU professor is using the term "attribute" here

They are 1 2 3 4 5 6 attributes

to describe TABLE 2.1 in Tom M. Mitchell. Machine Learning (free)

enter image description here

usually, those 6 columns are also called features, so, there are 6 features in that table.

That professor uses the term "candidate feature" here

Every potential binary function is a candidate feature

obviously, "candidate feature" is different to "feature".

Does candidate feature means the point in feature space?

Note: There are 6 features while the number of candidate features in this example is $2^{96}$


1 Answer 1


Candidate features are all given characteristics of an instance [Sky, AirTemp, Humidity, Wind, Water, Forecast] which come into consideration to predict its class label [EnjoySport].

Model accuracy and training speed can be significantly compromised by irrelevant or redundant features which can lead to overfitting. Hence, it might be a good idea to remove superfluous features before fitting (feature selection).

The ones you end up using to train your classifier are called features.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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