Comment 1 for bug 428895

There are two issues here I think underlying this.

The more I thought about it, the more I don't like having the results property. Some models (GLM, etc.) have a fit that takes arguments, so having a results property that returns the fit with default arguments isn't that useful (and I'd like most of the models to be as streamlined as possible).

Also, I'm not sure it makes sense to have predict as a method of the model rather than the results. I tried to connect the two, hence the error that is raised, but having a model method that depends on fit already being called is a bit odd to think about. Maybe it does though for the experimental guys who know what their params should be. I don't know.

This can be fixed, but I think we should make the design decision that this problem is a symptom of before we fix it.