We need to change the name of the package to RegularisationToolkit and add a short description, e.g. The collection of CPU/GPU optimised regularisation modules for iterative image reconstruction or other image processing tasks

On some occasions it fails with NaNs while CPU version works fine. Will investigate.

A dedicated project:
https://github.com/vais-ral/CCPi-FISTA_Reconstruction/projects/2

Remove the MATLAB specific stuff from main repository directory and put in Wrappers to follow the directory structure described in the Developer's Guide.

Fails with cuda 8.0, fine with 9.1:

----------------------------------------------------------------------
Traceback (most recent call last):
  File "/scratch/anaconda2/conda-bld/ccpi-regulariser_1542907757837/test_tmp/run_test.py", line 663, in test_FDGdTV_CPU_vs_GPU
    self.assertLessEqual(diff_im.sum(), 1)
AssertionError: 44.039284 not less than or equal to 1

----------------------------------------------------------------------
Ran 12 tests in 15.135s

FAILED (failures=1)```

Try the test_regularizers.py and you see that the Patch based regularizers returns only zeros.

after #60 this is no longer used

While running conda-build with Python 2.7 there is a following issue:

from ccpi.filters.regularisers import ROF_TV, FGP_TV, SB_TV, TGV, LLT_ROF, DIFF4th
ImportError: No module named ccpi.filters.regularisers
Tests failed for ccpi-regulariser-0.10.1-np114py27_0.tar.bz2 - moving package to /dls_sw/apps/savu/savu_future/miniconda/conda-bld/broken
WARNING:conda_build.build:Tests failed for ccpi-regulariser-0.10.1-np114py27_0.tar.bz2 - moving package to /dls_sw/apps/savu/savu_future/miniconda/conda-bld/broken
TESTS FAILED: ccpi-regulariser-0.10.1-np114py27_0.tar.bz2

As a workaround use --no-test to skip tests, i.e.:
conda build Wrappers/Python/conda-recipe --numpy 1.12 --python 2.7 --no-test

Write a decent demo script for the regularizers based on https://github.com/vais-ral/CCPi-FISTA_Reconstruction/blob/master/Wrappers/Python/demo/test_cpu_regularizers.py and https://github.com/vais-ral/CCPi-FISTA_Reconstruction/blob/master/Wrappers/Python/test/test_gpu_regularizers.py

• CPU regularizers
• GPU regularizers

https://github.com/vais-ral/CCPi-FISTA_Reconstruction/blob/master/Wrappers/Matlab/mex_compile/regularizers_GPU/Diffus_HO/Diff4thHajiaboli_GPU.cpp#L75

for (i=0; i < N; i++) {
            for (j=0; j < M; j++) {
                if (((i > 0) && (i < N-1)) &&  ((j > 0) && (j < M-1)))   B[(i-1)*(dims[1])+(j-1)] = B_L[i*M+j];
            }}  

what does the if? i or j are always less than M-1 or N-1. It seems to me that only when i==0 or j==0 the if returns False. Actually what happens to B in that case?

https://github.com/vais-ral/CCPi-FISTA_Reconstruction/blob/master/Wrappers/Matlab/mex_compile/regularizers_GPU/Diffus_HO/Diff4thHajiaboli_GPU.cpp#L56 initializes to 0?

https://wiki.jenkins.io/display/JENKINS/Github+Plugin states 2 modes:

1. manual, you need to put URL int Settings->Webhook->Payload URL: https://anvil.softeng-support.ac.uk/jenkins/github-webhook/
   Don't have permission to do that in github repo.

2. automatic, you need to enable in Jenkins Global Config the github repo.
   Don't have permission in Jenkins to see such configuration.

Rename fista_module.cpp to regularizer_module.cpp or similar, because that's what it is related to.

• recipe for regularizers (win)
• recipe for regularizers (linux)
• recipe for Python package

Windows build seems to be invoked in a thread and returns strange values.

When running try_astra_projector.py and calling function f = Norm2sq(Aop,b,c=0.5), the error is the following:

File "/home/algol/Documents/DEV/CCPi-FISTA_Reconstruction/modular/funcs.py", line 35, in init
self.L = 2.0self.c(self.A.get_max_sing_val()**2)

File "/home/algol/Documents/DEV/CCPi-FISTA_Reconstruction/modular/ops.py", line 109, in get_max_sing_val
self.s1, sall, svec = PowerMethodNonsquare(self,10)

File "/home/algol/Documents/DEV/CCPi-FISTA_Reconstruction/modular/ops.py", line 124, in PowerMethodNonsquare
x1norm = np.sqrt((x1**2).sum())

TypeError: unsupported operand type(s) for ** or pow(): 'VolumeData' and 'int'

I'd like to create separate branches for each of the regularizers excluding FGP & ROF-TV. What would be the best way to do that?

Skip GPU tests when "code: 35, reason: CUDA driver version is insufficient for CUDA runtime version".
Currently tests fails on

Executing transaction: ...working... done
Error: /home/jenkins/workspace/CILsingle/CCPi-Regularisation-Toolkit/Core/regularisers_GPU/Diffus_4thO_GPU_core.cu:234, code: 35, reason: CUDA driver version is insufficient for CUDA runtime version
Tests failed for ccpi-regulariser-0.10.2-np111py36_0.tar.bz2 - moving package to /home/jenkins/conda-bld/broken
WARNING:conda_build.build:Tests failed for ccpi-regulariser-0.10.2-np111py36_0.tar.bz2 - moving package to /home/jenkins/conda-bld/broken
TESTS FAILED: ccpi-regulariser-0.10.2-np111py36_0.tar.bz2
Build step 'Execute shell' marked build as failure

OpenMP declarations are not used. Probably wrongly built.

After cmake build and running GPU modules in Matlab (Linux). The related issue. Can be fixed either linking to a newer libstdc++.so.6.0.24 or by setting the env variable:
export LD_PRELOAD=/home/user/anaconda3/lib/libstdc++.so.6.0.24
The latter worked in my case

running test_regularizers.py on linux FGP_TV does not return the same as the windows version. It returns NaN

After Cmake build Python can't find
ImportError: libcilreg.so: cannot open shared object file: No such file or directory
I do

cd install/python/
export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:.

I run python from install/python/ and it still cannot find libcilreg.so.

add build variants to the conda build

it seems that the package gets cythonized correctly but when imported from python the ROF_TV function does not exist

Does this line work in Matlab at the first iteration?
https://github.com/vais-ral/CCPi-FISTA_Reconstruction/blob/master/main_func/FISTA_REC.m#L675

Given that the r_old is defined only if i>0

https://github.com/vais-ral/CCPi-FISTA_Reconstruction/blob/master/main_func/FISTA_REC.m#L510

http://proximity-operator.net/

In order to keep things in one place while Cythonizing, i.e. just have regularizers.pyx
Pros:

1. Easy to manage interfaces to CPU/GPU functions and also do parameters check for both cores.
2. A possibility of adding a 'CPU' / 'GPU' switch in order to have just one name for the regularizer,
   e.g. FGP_TV(blabla, 'GPU')
3. More compact representation

Cons: A bit more challenging debugging

P.S. As a trial I can do it for ROF or FGP.

Create GPU regularizers lib

• Add Nonlocal Total Variation

free(alloc) not added. Fixing now...

FGP-TV CPU has been updated to make code more self-consistent where the main function is:
https://github.com/vais-ral/CCPi-FISTA_Reconstruction/blob/b9562bbfdf6ca2cc8775e242ae13cbfaeeb2479b/Core/regularizers_CPU/FGP_TV_core.c#L40

One needs to wrap it using Boost or what has been used recently for ROF-TV? Essentially I'm taking care most of memory allocation stuff inside the C-code so should be less work to set it up from outside.

FGP - CUDA version has been added with the same interface:
https://github.com/vais-ral/CCPi-FISTA_Reconstruction/blob/b9562bbfdf6ca2cc8775e242ae13cbfaeeb2479b/Core/regularizers_GPU/TV_FGP/FGP_TV_GPU_core.cu#L344

Remove hard coded boost numpy and boost python libraries from setup.py.

Until that's done we are stuck to using boost 1.64 and numpy 1.12

I found out that mex'ed GPU functions will not run until I run them from the folder where they're in first! Then I can go to another folders and run it from there, but until that Matlab says something like: Invalid MEX-file '/home/algol/Documents/MATLAB/rtmct/MultichannelRegularization/priors/FGP_TV/GPU/FGP_TV2D_GPU.mexa64': FGP_TV2D_GPU_kernel.o: cannot open shared object file: No such file or directory.

https://github.com/vais-ral/CCPi-FISTA_Reconstruction/blob/master/main_func/FISTA_REC.m#L502
https://github.com/vais-ral/CCPi-FISTA_Reconstruction/blob/master/main_func/FISTA_REC.m#L524

If I'm not mistaken the 2 test are like this:

if i>=3 and lambda>0
   then do A
[...]
if i < 3 and lambda >0
  then do A

which is equivalent to

if lambda >0 then do A

Right?

These lines of code are always run and could be ported outside the if-then-else control loop.

https://github.com/vais-ral/CCPi-FISTA_Reconstruction/blob/cb8ef11f00e897b6f5b3049126dd32baa3c50cf9/main_func/FISTA_REC.m#L531
https://github.com/vais-ral/CCPi-FISTA_Reconstruction/blob/cb8ef11f00e897b6f5b3049126dd32baa3c50cf9/main_func/FISTA_REC.m#L553
https://github.com/vais-ral/CCPi-FISTA_Reconstruction/blob/cb8ef11f00e897b6f5b3049126dd32baa3c50cf9/main_func/FISTA_REC.m#L579

make a choice of the names of the regularizers and their parameters

Do we use Cython exclusively for building up CPU and GPU modules? If Yes, then we need to remove all Cmake and boost dependencies? Do we actually need CMakeLists.txt ?

@paskino please have a look at problems with compilation now.

-- Divergence-based diffusion PDE’s for 2D and 3D cases
-- Possible extension to multichannel data later.

A new regulariser has been added into the master which includes C-OMP realisation of ROF-TV model
http://www.ipol.im/pub/art/2012/g-tvd/article_lr.pdf

This is the classical explicit scheme which is although slow has one of the lowest memory footprints among other TV regularisers. GPU implementation is also suitable for fairly large 3D volumes. It will follow at some point.

it worth to include it into regularisation compilation routine...

When moving things to Python it is better to look at DemoRD2.m where real data reconstruction is performed in full 3D.

-- Bowsher prior to implement or rather re-write the CCPForge version

Update the code from master into pythonize branch.

Create base Reconstructor class and OS Reconstructor.

building with cmake, complains to the line

The error:

 File "/home/kjy41806/Documents/SOFT/CCPi-Regularisation-Toolkit/build/Wrappers/Python/setup-regularisers.py", line 71
    packages = {'ccpi', 'ccpi.filters'}
                      ^
SyntaxError: invalid syntax
make[2]: *** [Wrappers/Python/build/timestamp] Error 1
make[1]: *** [Wrappers/Python/CMakeFiles/PythonWrapper.dir/all] Error 

All seems in order to me...

Some work required to do Tests and Demos.

implement some (2D+time) ideas using nltv gpu module