flexible discriminant analysis computational completixy I'm using mda package + caret infrastucture to perform a flexible discriminant analysis for a classification problems. I have 26 features of mixed type. I found little if any guidance on the computational time. How does it vary as a function of both predictors and dataset size?
 A: From this document:

"2.6 Execution time
For a given set of input data, the following can increase the speed of the forward pass:
(i) decreasing degree (because there are fewer combinations of terms to consider),
(ii) decreasing nk (because there are fewer forward pass terms),
(iii) increasing minspan (because fewer knots need to be considered),
(iv) decreasing fast.k (because there are fewer potential parents to consider at each forward step),
(v) increasing thresh (faster if there are fewer forward pass terms).
The backward pass is normally much faster than the forward pass, unless method = "exhaustive". Reducing prune reduces exhaustive search time. One strategy is to first build a large model and then adjust pruning parameters such as prune using update.earth."

(Note that train uses the strategy with update.earth when train's method = "earth" but not when method = "fda" so technical reasons related to how MARS is bundled into FDA.)
Otherwise, it depends on your data size, number and types of predictors etc. The document linked above does have formulas for memory usage and that might be an acceptable proxy.
Max
