Optimization Maps

[nonlinear5]

It's been some time since I've added optimization maps to JBT, and I
am improving it for the next release.

Here is a problem with the way it's currently implemented. Suppose you
have a strategy with 3 parameters, and optimizer comes up with these
results for the following three combinations:

(10, 20, 29) --> -$2,000 P&L
(10, 20, 30) --> +$10,000 P&L
(10, 20, 31) --> -$5,000 P&L

If you look at the optimization map of paramer1 vs parameter2, you
would see +10,000 P&L for a pair (10, 20). This is because the map is
constructed by picking the best result for the pair of parameters.
This can be very misleading, though, as this example demonstrates,
since very slight deviations in parameter3 caused an overall loss.

The way I am going to fix it is this: for strategies with 3 or more
parameters, each optimization map will have an additional combo box
selection between "best case", "worst case", and "average case". Using
the example above, you would then see:

(10, 20, best case) -> +$10,000 P&L
(10, 20, worst case) -> -$5,000 P&L
(10, 20, average case) -> +$1,000 P&L

With this improved mapping, it becomes easier to interpret
optimization results and to conclude that the combination (10, 20, X)
is probably not a good choice.

Comments are welcome.

[dyno]

that's a great improvement compared to the previous version. After the
optimization, people want to know how to get the best robust
parameters. Robust means no sharp decrease in performance for small
change of the optimal parameters.

[duo737]

great improvement indeed .
From an ease-of-use standpoint, can I suggest to add not only the
combo box but also a visual measure of the variance between best-
worst and average. It could be displayed e.g. as an immediate ( i.e.
no click needed) color bar , or as a mono-color progress bar . This
way the robustness could be seen at the flick of the eye.