Adapted eta interpolation:
### Issues with previous interpolation:
## Eta Calculation:
- previously assumed photon hit to be in bottom left pixel of cluster
(photon hit assumed in bottom right pixel of cluster)
- clusters are filled from top left to bottom right (previously assumed:
bottom left to top right)
## Actual Interpolation:
- photon hits are given in pixel coordinates (previous interpolation
assumed euclidean coordinates, e.g. positive distance in y coordinate
becomes negative distance in row pixels)
- removed *2 of calculated distance
## General Adaption:
- max_sum_2x2 return subcluster index relative to cluster center e.g.
bottomleft, bottomright
## Added proper test case
- simulated photon hit with normal energy distribution
- Note: Test case for 2x2 cluster fails - Think uniform photon hit
distribution cant be modeled by normalized eta distribution for 2x2
clusters
The python bindings build a shared library and I cant link against
static libraries. Apparently I have to build with
CMAKE_POSITION_INDEPENDANT_CODE=On.
- Still quite far from a state where it can be merged
- Reduce 5x5 to 3x3
- Reduce 3x3 to 2x2
Open issues:
- [ ] Can we generalize it?
- [ ] Which reductions are needed
- [ ] Naming
Vadym accidentally called view() directly on an R-value frame, which
leads to a dangling view pointer.
Adjusted code such that compiler throws an error if called on an R-value
frame.
Co-authored-by: Erik Fröjdh <erik.frojdh@psi.ch>
While using the cluster finder and saving a cluster, pixels which are
out of bounds are skipped. cluster.data should contain the pedestal
corrected ADU information of each pixel.
However, the counter "i" which keeps track of the position of
cluster.data is only incremented if the pixel was inside the bounds of
the frame.
This means that any clusters close to the frame's edges are not
construed properly. This means that if you want to extract a 3x3 from a
9x9 cluster, it can fail if the cluster data is not properly centered in
the pixel.
Fixed by moving i++ outside the bounds check.
Co-authored-by: Jonathan Mulvey <jonathan.mulvey@psi.ch>
- NDView operator()(size_t) now returns a view with one less dimension
- Apply calibration takes also a 2D array and then ignores pixels that
switch
- Calculate pedestal from a dataset which contains all three gains
- G0 variant of pedestal
- Function to count pixels switching
- helper function to initialize shape
- helper function to calculate the number of elements
- move constructor to create a NDArray<T, Ndim-1> if sizes match
Erik Fröjdh
mazzol_a
maliakal_d