Handle NaN or infinite feature values about cesium

[bnaul]

In case anyone's curious why a feature would be infinite: a few features assume that the data are log-scale like magnitudes, and in converting large values back to linear scale we can over- or under-flow. So I think these features are behaving as expected, but just don't make sense for every application.

[bnaul]

Another issue: sometimes NaN or infinite values appear during the computation of a feature and this currently breaks the whole pipeline. Need to handle such errors more gracefully (how?).

[stefanv]

What is the current status here, Brett?

[bnaul]

Seems like this should be easy to fix, just need to decide what to do exactly:

• Notify user as soon as invalid feature values are detected? i.e., mid- or post-featurization
• Throw a (human-readable) exception informing the user which features are invalid?
• ...?

[stefanv]

How about both? An error message that pops up, and then a little report when you click on the failed result?