CHANGELOG.md

Reaction Plane Fit Changelog

Changelog based on the format here.

[4.0] - 21 October 2019

Added

• Additional documentation. See: c49fa753201250734f6dbc4bdce6a64959a277cb.
• Integration test for differential signal fit. See: efcfc40fdf662d5fb0297ed1d02c2f6b538e2485.

Changed

• Switch emphasis to fitting before normalizing by the number of triggers, including updating default arguments and documentation. See: 17a5d24397c313c3c5a39964e9a8bd471161153e, e04a1f3b1736c8baef41872680bd175f3dadd977, 73f5f0d8dcbf7f86f1742e57279da43fd62faa1f.
• When reading the RP fit object from YAML, setup the entire fit object. See: 1b38175ae421729a1933bae63bcc6652262927d3.
• Improve Minuit fit strategy. See: d8dfb49d0599f564597461270b73bb67d1818bdf.
• Rename _fit to cost_func for clarity - it represents the cost function, not the fit function. And it's sometimes useful to access from outside the object (such as when calculating effective chi squared). See: 3dc750119be1f52f05134b11f24883b2e5c73b07.
• Make the Minuit verbosity a user settable option. See: 22e67c85b630ef09a3105bcee4587e8d7091b293.
• Allow a user to skip HESSE when necessary. This should be rare, but it is required at times. See: 98104086ca505a25778c75a0c63053085a836368.
• Updated the signal differential fit procedure. The inclusive case is just treated as a background only fit. This is required because adding coupling gaussian distributions is not a straightforward process (unlike if they were uncoupled). Consequently, extracting the inclusive width relies on fitting to information already included in the RP fit (included via the differential signal fit components), which means that the covariance of the fit is neglected. Use care when using this fit approach! (A warning is thrown when using it). See: 7ba221d8effcda48d3a5245776aefcb971d7748e.
• Sample data now acts as if it is the raw correlation rather than scaled by the number of triggers. See: 5c613d2f5d21882fe9d61d6d0067bd602cb74ff2, 9484b457012f6647283e3fc1ebf43a719a0fd7e5, 5c613d2f5d21882fe9d61d6d0067bd602cb74ff2, and 7db8ea4e4201879c609c0cebeb211ea67570e7ba.
• Improve testing procedure by loading stored fits from YAML rather than defining them by hand in the test file. This makes any future updates much easier. See: 5c613d2f5d21882fe9d61d6d0067bd602cb74ff2.

Fixed

• Log likelihood fits by updating the underlying pachyderm.fit package. See: 33bdcb41cd0f2dcc6e76f8760b8e5eddd24456dd.
• Minor typos in code, documentation. See: 150afbba7f065f54895f1bd30409c66df14990b9 and 56af2759823cf2ff027c2b89d7047fdcb2bfc98f.
• Create output directory when writing if necessary. See: 7ca7f01d12a897907427972c6a9f6402cba6c616.
• Allow specification of v_{1} values. See: ed037ba57f72dc24257dd6a2dbe1351e85e6c73d and a fix in 63cb49906eabc115f58b6847de756346b61b08c3.
• Updated typing. See: db79b4134f57ff76727356654489b9a823cb6b67.
• Integration tests run successfully on TravisCI again. See: 330c3bc6a7f90f1a5e0773c03170a8b27b8cf4ca.

[3.1] - 24 August 2019

Changed

• Added a method to create a full set of fit components for each EP orientation, regardless of which were used for the fit. This allows for downstream code to not need to differentiate between the different cases for most tasks, easing the user experience. See: Nearly all of the commits between 3.0 and 3.1.
• Use the pachyderm.fit.BaseFitResult for components in order to avoid needing to calculate unnecessary values for component fit results. It's more natural. See: bbbcac45c41490072f345b6aebbeb6f49d427a1d.
• Updated the included test data for better test consistently with external sources. See: 0bf5e1ceaede9cd578a20c12db81455f38326a2d.
• Updated pre-commit hooks. See: 6d780545b5547f42c7c3a684f11234d069973c10.

Fixed

• Warning about registering pytest marks. See: 8f6f5386fddea1e540a15de99ce8852dfc2319fd.
• Removed final probfit dependence. See: b94ea052246269e454f3df1c7afbb7ac96f35cec.

[3.0] - 8 August 2019

Changed

• Move to using the pachyderm fit package for general fit functionality, cost functions, and fit results. This move was instigated by the recognition that probfit cost functions didn't actually work properly with pre-binned data. It also allows for the centralization of basic fit code to make it generally available. Some code from this package was refactored into pachyderm. See: 9e62f6777afed290709b74d4ee18284d61c2bab9 and 0f66cff8264ff06f5138d78ba0213f40a195dd5e. Note that developed was handled in a private branch in this repository, so some older commits have references to a fit package in the RPF package before it was moved to pachyderm.
• No longer rely on probfit. All functionality and more is now provided in pachyderm.fit. Plus, the package wasn't super actively maintained. See: 9e62f6777afed290709b74d4ee18284d61c2bab9.
• Updated pre-commit hooks. See: 03844495a12f4269b14dbd4972c09fb203a88c7c.

Fixed

• Log likelihood and chi squared cost functions properly take binned data into account. See: 0f66cff8264ff06f5138d78ba0213f40a195dd5e.

[2.1.2] - 4 May 2019

Added

• Enabled strict mypy checks. See: 1585c7398201319bac49f6881f678be3f9582ee5.

Changed

• Improved typing. 341330a5e48f1526848de27ceaba4ee10499de67.

[2.1.1] - 3 May 2019

Changed

• Refactor function error calculation so that it can used by other packages. See: 04dff6cc7cbd1fd4ad1f2f26ecc528dfeebbe3f7.

Fixed

• Enable fixing parameters via user arguments. See: cf9de684245217d4c7583407940d5a3864b192b1.
• Properly mark the package as typed. See: fe43d5368bb8fdd1defa5dbff8a39ac45d2424a3.
• Improved code formatting and typing information, and fixed typos.
• Fix mypy typing in third party packages by adding zip_safe = False in the setup.py. See: 79eb4e821bed8d5290c18c779c2bce6c8a71e5c0.

[2.1] - 29 March 2019

Added

• Calculate the correlation matrix in the fit result. See: ee94cb80e774c38848f31592b8df765c333fba7a.
• Read and write fit results with YAML, including integration tests. See: 983a2519d9b882a34d6bf44be312b45a3e051316.
• Added test for invalid fit arguments. See: 009a974dc038d79ad2890e9c3f935094a88cb3ee.

Changed

• Modify the ReactionPlaneFit.fit(...) to return the minuit object for advanced use cases. The fit result stores almost all of the minuit information, but it can be useful in limited circumstances. For now, it is only used to check the correlation matrix calculation. See: 7a1efd908203abf1a1b8cb8d2f9d76afdfb1442d.
• Initialize the component fit functions when defining the fit objects. This allows fit components to be evaluated using stored fit results. See: e2673f25e7fa952087bb45ab5e35204850138b69.
• Updated typing information. See: 0f284c79ccfb6ee04715b33e4500b2c96832f7ff and 60af85240d84adf5e011ee0b4b2f7a0c8cef154e.
• Updated documentation. See: b94fc955ce15034490888ec75368853f61ad9244.
• Updated pre-commit hooks with newer versions of the packages. See: 6d54231e4707502fec9811c0507daa5a37358076.

[2.0] - 22 February 2019

Note the API changes introduced by moving the fit results to the components!

Added

• Store errors on fit parameters. See: fbb9bd36cf7f4b010377cb1c1d27cb5d09ccf57d.
• Evaluate the background function of a component. For a background fit component, this is the same as the fit function, while for a signal fit component, this will describe only the background contributions. See: 8836a38cbdb44f6db1508a23f295ff4872433938.
• Preliminary instructions on running the fit via c++. They are not yet tested. See: 8f6d57ca160e9500b4fa5c2747faa9e4b23c5a12.

Changed

• Moved component fit results into the fit components. See: a08c42a483d51ba5c599986ae8f5cdec6a4505f0 and 51ef75007d2a765bd03bf01958517d3109140067.
• Moved fit evaluation into the fit component. This was enabled by moving the fit results into the fit components. See: e3c13956b80ad31b7232f46fe0784b0179462175.
• Moved FitType to the base module so that it is more easily accessible. See: f3513c9f3d01d51c46e6b776fad005c32864001d.
• Moved error calculation to fit component. See: 674c8cf0f3bff1ebe95a8d5b76a7c2917d3169b7.
• Refactor error calculation to allow it to calculate errors for both the fit and background functions (or any other function). See: 674c8cf0f3bff1ebe95a8d5b76a7c2917d3169b7.
• Improved example plot styling. See: eb9fe16bd19e1e0106a822ee7e6a31f8a9be7e51 (included with a few other minor changes).
• Determine and plot errors on residual. See: 74b7f14b34b37b3563dcebc1031730a9343b5252.
• Large improvements to typing information and documentation in a variety of commits.

Fixed

• Background fit example legend placement was accidentally moved. See: 1869eb0650c9b26157c6fed103e80e5ad78f7f08.
• Baseline test images, which started failing after a number of small aesthetic changes. See: 795d9d228a5cde6d5477dfd1c697efcabec7547c and 33f9dd83f5918ba5f5eabd86b78759df1197a9ac.
• Failing tests due to floating point variations in calculating parameter errors. The comparison tolerances had to be loosened. See: e668599966df447e99de7c145ad6c97ce7d2d814 and f2729a4c5ed517b25fd99019f71ccabf65bd2ac5.
• Indicate that the package is typed. The typing information is already there - it just didn't know it propagate it when running mypy on other packages. See: b113e40be09705fdc3f4cce1a6ea7bbcaa14442a.
• Updated call to the old pachyderm.Histogram1D API. See: f23f31473b77203e3c2d5cf9d2617c4f3f53a6db.

Removed

• evaluate_fit_component(...) from the ReactionPlaneFit object. It is unnecessary given the API changes. See 5e2048eb1bace3d34dd428aea79b7ca002e61156.

[1.2] - 30 January 2019

Added

• Added mypy to Travis CI checks. See: 3fe44b42773b7b6d4fa184cf8db82a5e529713e0.
• Pre-commit hooks configuration to the repository based on pre-commit. This should generally improve the quality of commits. Currently includes flake8, mypy, yamllint, and some additional minor python checks. See: e0218dffc0829ac742790497b5f0494bd41c0035 and a40c72e42cb50bf0658c871e8ea8d3f0aa5b393e.
• Ability to run Minos errors. They aren't stored (because it doesn't make sense to do so), but they can be compared to Hesse errors. See: b0149099bc4e05a0e4f37d78680abe5ed4fa7ff3.
• Allow for user arguments to override the fit arguments. See: bf570c37163b57c02d890aa5047f99e8bb82077c.

Changed

• Adapted to new pachyderm histogram API. See: c6b17fe6555d7baab3be9cb53c4f238bd8e23b65.
• To implement the new API, it required some relaxation of array comparison tolerances (but they are still quite stringent). See: 631db84754b1d03076acc24ea5ab9bdae88bcb38.
• Renamed "inPlane" -> "in_plane" (and the other angles accordingly) for better matching of conventions (which makes it easier to use and integrate with other packages). See: 300258f77f2ebd9f83754ab1d0f41e028e95032f.
• Updated RP plot labeling for new RP orientation naming convention. See: 7a9e939f442a73e96349583553228d923341109a.

Fixed

• A number of type annotation issues identified by mypy. See: 8b71c77f99f8d95fd364bc7a700e2a32ecbdb499.
• Test discovery failed in some cases. See: 85e5ad0f2a6ab4d55393167c9289018061826f37

[1.1] - 11 December 2018

Added

• Test the FitComponentResult objects in the integration tests. See: dcb21966.
• Updated documentation.
• mypy to the dev requirements. Although not run in the CI (because not all of the identified issues are actually issues), it is good to have available and run from time to time. See: bdc216a7.

Changed

• Refactored Histogram class to pachyderm, where it was renamed Histogram1D. See: 923c3847.

[1.0.1] - 30 November 2018

Fixed

• Resolved a tagging problem related to a rebase.

[1.0] - 30 November 2018

Added

• Zenodo DOI information.
• Chi2/ndf calculation.
• Plot module (with tests) to draw the fit and the residual.
  • The tests take advantage of image comparion.
• Additional fit quality checks.

Fixed

• Parameter argument ordering in the signal inclusive fit. See ba4e833a.
• Fixed error propagation by calculating the error for each individual component rather than for the simultaneous fit (which was the wrong approach). This has the side effect of substantially speeding up the error calculation.
• Specify all initial parameters, limits, and steps, which can substantially improve the chi2.
• Various type annotations.

[0.9] - 27 November 2018

• This is essentially a preview release.
• It can perform the RP fit using background only data, including the reaction plane orientation inclusive signal, or the reaction plane orientation dependent signal. Full error propagation is performed for each fit component.