CosmoSim is a simulator for gravitational lensing.
Test/roulettegen
Given a dataset as CSV, this test
montage/
directory.cd Test/roulettegen
make clean
sh test.sh
```
The test script compares the original roulette simulation
with the resimulation.
Output image collages is found in the `montage` directory.
### Makefile
cd Test/roulettegen make montage.log
The makefile compares the roulette resimulation
with raytrace simulation.
Output image collages is found in the `montage` directory.
### Old notes
The output is a mess, but it is standardised for many different tests.
If one uses the text log, one should look for the string
`maskedCompare` which appears once for each data point,
like this.
maskedCompare (600, 600, 3) diff/image-s003.png(10): 0.049611466296889374 (47.18050444834179); size=951.0 diff/image-s003.png(20): 0.02648499649355367 (99.87492177719089); size=3771.0
Each row is the comparison of one neigbourhood in the form
imagefile(radius): relative distance (distance): size ``` Here the size is the size of the neighbourhood, the distance is the Euclidean distance between the original and the reconstructed neighbourhood, and relative distance is disance divided by size.
The current version (develop branch 3 July 2024) uses only the --xireference
option,
which is also what has been used in the machine learning tests.
As of March 2025 --xireference
is the default; it can be turned off with
--no-reference
.
The role and importance of this option needs documentation.
If --no-xireference
is used without --centred
, the roulette amplitude
output is predominantly NaN values. I do not know why. This should also
be documented, and it may be a good idea to remove options.