Finish colocalization

[etadobson]

• merge Colocalize command for global methods
• finish SACA code (see below comment)
• wrap SACA code in Ops
• add SACA Op to Colocalize command
• MTKT needs some TLC !!! (see issue MTKT issues: Correctness & Performance are suspect #619)
• send to Ming/Shulei for maths tests
• contact Nate Sherer for bio-dataset testing
• announce!

[etadobson]

• make Shulei's java implementation stable (ie - use random(seed))
• add a unit test validating behavior of the algorithm (using small synthetic double[][]'s as input)
• propagate random(seed) update to Imglib2 version (make sure unit tests still pass)
• code style / clean-up pass (no public fields, minimal public API, non-public helper classes...)
• consider getting rid of OO design (constructor/execute... in favor of static method) (have static method that takes same arguments as constructor and pass to execute())
• address new performance bottlenecks...
• ensure better Random behavior... don't create a new one with each call to WtKendallTau
• convert to Ops computer style where result image is preallocated and passed to the algorithm
• image creation factory needs to support creating buffers of a different type... RAI<T> to RAI<U> and result should be RAI<V> ... should use input image type <I>, intermediate type <T>, and output type <O>
  • when testing, make sure to pass factory that produces DoubleType images always for consistency with Shulei's original code
• contact Shulei... ask if other outputs should be made available to users (ie - double-check p-values / confidence scores / threshold cutoffs)
• noticed intermediate and result NaNs - why?

[etadobson]

imglib/imglib2#273

[etadobson]

The output from the SACA op needs to be signed!!! "The pixel-wise z-score output from SACA not only reflects colocalization or anti-colocalization at each pixel by its sign, but also shows the degree of colocalization or anti-colocalization through its absolute values."