- Fix
PowerLaw.from_simulationfor the new format ofSimulatedEventsInformation, it was broken since splitting the singleviewconeintoviewcone_minandviewcone_max. [#258]
This release contains an important bug fix for the energy dispersion computation, it was wrongly normalized before.
In prior versions of pyirf, the energy dispersion matrix was normalized to a sum of 1 over the migration axis. This is wrong, the correct normalization is to an integral of 1, which is fixed now.
The internal API of the interpolation functions had to be adapted to take in additional keywords, mainly the bin edges and the kind of normalization (standard or solid angle cone sections). [#250]
Replace single
viewconeargument ofSimulationInfowithviewcone_minandviewcone_max, e.g. to correctly enable ring wobble simulations. [#239]
- See above on the energy dispersion change.
- Add option to specify which containment to use for angular resolution. [#234]
Change the interpolation API to top-level estimator classes that instantiate inter- and extrapolator objects. Drops the
interpolate_xyzfunctions originally used to interpolate a xyz IRF component in favour of aXYZEstimatorclass. Moves data checks from intepolator to estimator classes.Direct usage of interpolator objects is now discuraged, use estimator objects instead. [#228]
- Correctly fill n_events in
angular_resolution, was always 0 before. [#231] - Remove condition that relative sensitivity must be > 1. This condition was added by error and resulted in returning nan if the flux needed to fulfill the conditions is larger than the reference flux used to weight the events. [#241]
- Add moment morphing as second interpolation method able to handle discretized PDF components of IRFs. [#229]
- Add a base structure for extrapolators similar to the interpolation case as well as a first extrapolator for parametrized components, extrapolating from the nearest simplex in one or two dimensions. [#236]
- Add an extrapolator for discretized PDF components, extrapolating from the nearest simplex in one or two dimensions utilizing the same approach moment morphing interpolation uses. [#237]
- Add a
DiscretePDFNearestNeighborSearcherand aParametrizedNearestNeighborSearcherto support nearest neighbor approaches as alternatives to inter-/ and extrapolation [#232]
- Migrating the interpolation methods from
pyirf.interpolationto interpolator objects, allowing for later inheritance for new algorithms and reusability. [#210]
Add and enable
towncrierin CI. [#207]Add a fixture containing three IRFs from the prod5 IRF data-release for unit testing. Specifically the fixture contains the contents of:
- Prod5-North-20deg-AverageAz-4LSTs.180000s-v0.1.fits.gz.
- Prod5-North-40deg-AverageAz-4LSTs.180000s-v0.1.fits.gz
- Prod5-North-60deg-AverageAz-4LSTs.180000s-v0.1.fits.gz
The user has to download these irfs to
irfs/usingdownload_irfs.py, github's CI does so automatically and caches them for convenience. [#211]
For releases between v0.4.1 and v0.8.1, please refer to the GitHub releases page.
0.4.1 (2021-03-22)
- Released March 22nd, 2021
- 1 Contributors
- Maximilian Nöthe
- #135 Add functions to convert pyirf results to the corresponding gammapy classes
- #137 Add example notebook for calculating point-lile IRFs from the FACT open data
0.4.0 (2020-11-09)
- Released November 11th, 2020
- 2 Contributors
In order of number of commits:
- Maximilian Nöthe
- Michele Peresano
This release is an important update that introduces three changes in the cut optimization, background estimation and sensitivity calculation.
Together, these changes bring the calculated sensitivities much closer to the ones calculated by EventDisplay.
- Scale the relative flux calculated to reach the target sensitivity
up if the requirements on the minimum number of signal events are not met.
Essentially, instead of always calculating the flux that
yields
target_sensitivityand then checking if the two other conditions are met, we increase the required flux to meet the other requirements. This can result in new sensitivities where before pyirf would report no sensitivities, and report better sensitivities everywhere where the event number conditions where not met before at the target significance. The best sensitivity now is the lowest flux that just barely satisfies all requirements (so is at the minimum requirement of one of the three). - Differentiate between reco_source_fov_offset and true_source_fov_offset, using the former for background rates and the latter for everything concerning signal events.
- Change
optimize_gh_cutto do the optimization in terms of efficiency and limit this efficiency to max. 80 % in the EventDisplay comparison.
Smaller improvements also include:
- It is now possible to include a
particle_typecolumn in the event lists, which will result in additionally reporting all event counts also perparticle_type. E.g. ifparticle_typeis included in the background table consisting of both electrons and protons,estimate_backgroundwill not only reportn_background(_weighted)but alson_electron(_weighted)andn_proton(_weighted) relative_sensitivitynow supports vectorized application and broadcasting of inputs, as previously wrongly advertized in the docstring.
GammaPy 0.18.0 was released and includes fixes for IRF axis orders.
The output of pyirf in GADF fits format can now be read by gammapy without
problems.
The workarounds for installing GammaPy is also no longer needed.
- #110 Optimize cuts in efficiency steps with maximum efficiency of 80% for EventDisplay comparison
- #104 Scale flux for conditions, differenatiate reco and true source_fov_offset
- #108 Add counts / weighted counts per particle type
- #107 Small update to installation instructions
- #106 Use vectorize for relative_sensitivity
0.3.0 (2020-10-05)
- Released October 5th, 2020
- 5 Contributors
In order of number of commits:
- Maximilian Nöthe
- Michele Peresano
- Noah Biederbeck
- Lukas Nickel
- Gaia Verna
This release is the result of the IRF sprint week in September 2020. Many bug fixes and improvements were made to the code.
As the target for the sprint week was to reproduce the approach of EventDisplay and
the resulting IRFs, one scheme of cut optimization is implemented.
The examples/calculate_eventdisplay_irfs.py should follow the approach
of EventDisplay closely and shows what is currently implemented in pyirf.
In the central and upper energy range, pyirf now reproduces the EventDisplay sensitivity
exactly, the lower energy bins still show some disagreement.
The cut optimization seems not yet to be the same as EventDisplay's and will be further investigated.
This example could be used as a starting point if you also want to do cut optimization for best sensitivity.
At least one version of each IRF is now implemented and can be stored in the GADF format. Computation of full-enclosure IRFs should be possible but is of now not yet tested on a reference dataset.
- #97 Store correct signal amount, store information on which checks failed for sensitivity bins (Maximilian Nöthe)
- #96 Add integration test (Michele Peresano)
- #98 Remove option point_like for psf (Maximilian Nöthe)
- #95 Cut updates (Maximilian Nöthe)
- #91 Fix conditions to take relative sensitivity into account, fixes #90 (Maximilian Nöthe)
- #89 Fix brentq returning the lower bound of 0 for flat li ma function (Maximilian Nöthe)
- #85 Improve comparison to EventDisplay (Maximilian Nöthe)
- #75 Add a function to check a table for required cols / units (Maximilian Nöthe)
- #86 Fix Li & Ma significance for n_off = 0 (Maximilian Nöthe)
- #76 Feature resample histogram (Noah Biederbeck, Lukas Nickel)
- #79 Fix integration of power law pdf in simulations.py (Gaia Verna)
- #80 Estimate unique runs taking pointing pos into account (Maximilian Nöthe)
- #71 Background estimation (Maximilian Nöthe)
- #78 Change argument order in create_rad_max_hdu (Lukas Nickel)
- #77 Calculate optimized cut on only the events surviving gh separation (Maximilian Nöthe)
- #68 Effective area 2d (Maximilian Nöthe)
- #67 Add method integrating sim. events in FOV bins (Maximilian Nöthe)
- #63 Verify hdus using ogadf-schema (Maximilian Nöthe)
- #58 Implement Background2d (Maximilian Nöthe)
- #52 Add sections about tests, coverage and building docs to docs (Maximilian Nöthe)
- #46 Add PyPI deploy and metadata (Maximilian Nöthe)
0.2.0 (2020-09-27)
- Released September 27th, 2020
- 4 Contributors
In order of number of commits:
- Maximilian Nöthe
- Michele Peresano
- Lukas Nickel
- Hugo van Kemenade
For this version, pyirf's API was completely rewritten from scratch, merging code from several projects (pyirf, pyfact, fact-project/irf) to provide a library to compute IACT IRFs and sensitivity and store them in the GADF data format.
The class based API using a configuration file was replaced by a finer grained function based API.
Implemented are point-like IRFs and sensitivity.
This release was the starting point for the IRF sprint week in September 2020, where the refactoring continued.
- #36 Start refactoring pyirf (Maximilian Nöthe, Michele Peresano, Lukas Nickel)
- #35 Cleanup example notebook (Maximilian Nöthe, Michele Peresano, Lukas Nickel)
- #37 Move to python >= 3.6 (Hugo van Kemenade)
0.1.0 (2020-09-16)
This is a pre-release.
- Released September 16th, 2020
0.1.0-alpha (2020-05-27)
This is a pre-release.
- Released May 27th, 2020
- 3 contributors
- Started basic maintenance
- Started refactoring
- First tests with CTA-LST data
In alphabetical order by last name:
- Lea Jouvin
- Michele Peresano
- Thomas Vuillaume