Machine learning with variable features

I'm relatively new to the machine learning field and I've been experimenting with sklearn under Python. What I'm wondering, is there any way to create an incrementally trained model that can handle data sets with highly variable features?

This is mainly for analyzing web logs, with features that vary between numerical, text and categorical.

From what I've seen so far, most of the models are quite sensitive to feature drift over time and require inputs consisting of fixed, predictably sized vectors.