Support for quantitative evaluation of deconvolutions and addition of the inital argument to supply start values.
#33
Conversation
…ve evaluation of deconvolutions. Also added an example.
Codecov Report
@@ Coverage Diff @@
## master #33 +/- ##
==========================================
- Coverage 92.71% 92.15% -0.56%
==========================================
Files 14 14
Lines 535 599 +64
==========================================
+ Hits 496 552 +56
- Misses 39 47 +8
Continue to review full report at Codecov.
|
|
Thanks for the PR. I was wondering, why do we need now Also I would feel more happy if code coverage increases instead of decreasing. Can you check with CodeCov which are the missing lines and add maybe simplistic tests? We should ensure that the code is at least running. Checking for meaningful values is hard, I admit. |
src/deconvolution.jl
Outdated
| center_set!(rec0, one_arr) | ||
| rec0 .*= mean(measured) | ||
| fill!(one_arr, mean(measured)) | ||
| center_set!(rec0, ones(csize) .* initial) |
There was a problem hiding this comment.
Since I wanted the user to be able to call deconvolution also with a single number (such as the mean or 0.0) for initialization.
There was a problem hiding this comment.
Did you try without ones(csize)?
Maybe center_set! handles that correctly.
src/analysis_tools.jl
Outdated
| end | ||
| end | ||
| idx += 1 | ||
| false |
There was a problem hiding this comment.
This is specific to the callback. If it returns true the optimizer is told to stop.
There was a problem hiding this comment.
maybe add that as a comment and with an explicit return false
Before it was a bit inconsistent with the iterations (default 20 for |
|
New commit with hopefully everything fixed. Also the code coverage should have been increased. |
…re sense for unnormalized data.
In the analysis_tools.jl support for quantitative evaluation of deconvolutions was added via returning an option structure to be provided to the deconvolution routine. The new function
options_trace_deconvreturns an opt_options structure to be supplied todeconvolutionwhich then does all the bookkeeping. It also returns a query function that then provides the results.In addition the "initial" argument was added to allow starting the iterations with a value or array of the user's choice.
The tests of the new functions makes use of this
initialargument by supplying the ground truth and checking for the metrics to yield 1.0 and 0.0 respectively.