WIP: Remove unused estimate learning rate at each iteration.

[ntustison]

Description

I'm guessing this is a legacy option before we switched to the conjugate gradient descent optimizers for everything but the exponential transforms. But even those don't use this flag as no scales estimator is applied in these two cases (B-spline and Gaussian).

Addresses #1405

[cookpa]

Thanks @ntustison !

I think this means we should also remove this option from antsMotionCorr.cxx, which also uses a conjugate gradient optimizer

https://github.com/ANTsX/ANTs/blob/master/Examples/antsMotionCorr.cxx#L131

[cookpa]

I don't think this should have been removed, I get a seg fault now unless I specify --collapse-output-transforms 1. The argument handling is inconsistent and probably needs to be refactored, but some args (like this) have the convention that a default is assumed. In others, there is a check for the number of functions and only if > 0 is the value queried.

[stnava]

I this going to impact ANTsR and ANTsPy? I believe these do set the option internally.

[ntustison]

Thanks for trying this. Can you tell me exactly the command you're running because I can't reproduce?

[cookpa]

bash -c "../install/bin/antsRegistration --verbose 1 --dimensionality 3 --float 0 --output [ test,testWarped.nii.gz,testInverseWarped.nii.gz ] --interpolation Linear --use-histogram-matching 0 --winsorize-image-intensities [ 0.005,0.995 ] --initial-moving-transform [ tpl-MNI152NLin2009cAsym_res-01_T1w.nii.gz,T1w.nii.gz,1 ] --transform Rigid[ 0.1 ] --metric MI[ tpl-MNI152NLin2009cAsym_res-01_T1w.nii.gz,T1w.nii.gz,1,32,Regular,0.25 ] --convergence [ 100x50x25x10,1e-6,10 ] --shrink-factors 8x4x2x1 --smoothing-sigmas 3x2x1x0vox --transform Affine[ 0.1 ] --metric MI[ tpl-MNI152NLin2009cAsym_res-01_T1w.nii.gz,T1w.nii.gz,1,32,Regular,0.25 ] --convergence [ 100x50x25x10,1e-6,10 ] --shrink-factors 8x4x2x1 --smoothing-sigmas 3x2x1x0vox --transform SyN[ 0.1,3,0 ] --metric CC[ tpl-MNI152NLin2009cAsym_res-01_T1w.nii.gz,T1w.nii.gz,1,4 ] --convergence [ 10x7x5x0,1e-6,10 ] --shrink-factors 8x4x2x1 --smoothing-sigmas 3x2x1x0vox"

This doesn't work, but it does work if I add --collapse-output-transforms 1

[ntustison]

Thanks @cookpa , I'll look at it in a few minutes.

An alternative is to just add documentation to the option. I'm fine with whatever.

[cookpa]

This works for me

[ntustison]

Thanks @cookpa , @stnava . I'll take a look at any ANTsPy/ANTsR dependencies later today before merging.

[ntustison]

Also, I'll take a look at the other programs (eg, antsMotionCorr).

[cookpa]

Thanks for sorting this @ntustison

[effigies]

@ntustison Do you know when "we switched to the conjugate gradient descent optimizers for everything but the exponential transforms"? For nipype wrapping, it would be good to know at what point this had an effect.

---

Edit: From a quick git-blame, it may have been around 5bedd97, which was released in ANTs 2.0.

[cookpa]

I think it was 2013, here

42dd9d3