Add reproducible random sampling and many new pmag.py and ipmag.py tests by Swanson-Hysell · Pull Request #826 · PmagPy/PmagPy

Swanson-Hysell · 2026-03-01T17:49:28Z

Summary

Add an optional random_seed parameter to all functions in pmag.py, ipmag.py, and rockmag.py that use random number generation, using the modern numpy.random.default_rng() API. Closes include the ability to set a random.seed() in each function that uses random or np.random() #447.
Build a pytest-based test suite (369 tests) covering core scientific functions across coordinate transforms, Fisher/Kent/Bingham statistics, PCA, VGP calculations, field models, anisotropy, E/I corrections, bootstrap statistical tests, and data I/O.

Reproducible random sampling

All functions that use random number generation now accept an optional random_seed parameter (approach 1 from #447 — default behavior unchanged, but reproducibility is available when needed). This will close #447 :

# Default: non-reproducible (fresh entropy each call)
directions = ipmag.fishrot(k=50, n=100, dec=0, inc=45)

# Reproducible: pass an integer seed
directions = ipmag.fishrot(k=50, n=100, dec=0, inc=45, random_seed=42)

# Thread a Generator through call chains
rng = numpy.random.default_rng(42)
directions = ipmag.fishrot(k=50, n=100, dec=0, inc=45, random_seed=rng)

Implementation uses a _resolve_rng() helper in pmag.py that accepts None (fresh entropy), int (seeded Generator), or a Generator instance (passed through). All legacy np.random.* and random.* calls within converted functions are replaced with rng.* calls on the local Generator.

Converted functions

pmag.py (11 functions): fshdev, kentdev, gaussdev, get_unf, pseudosample, apseudo, mktk03, pseudo, di_boot, dir_df_boot, s_boot

ipmag.py (14 functions): fishrot, fisher_mean_resample, kentrot, tk03, bootstrap_fold_test, mean_bootstrap_confidence, common_mean_bootstrap, common_mean_bootstrap_H23, common_mean_watson, reversal_test_bootstrap, reversal_test_bootstrap_H23, find_svei_kent, plate_rate_mc, MADcrit

rockmag.py (1 function): backfield_MaxUnmix

Out of scope (cosmetic/plotting random, not scientific)

Site.eq_plot / Site.eq_plot_everything (random RGB colors for plot styling)

Test suite

369 tests across 24 test files, all passing. Organized by scientific topic with one pytest class per function.

Area	Files	Tests	Key functions covered
Coordinate transforms	`test_pmag_directions.py`, `test_pmag_transforms.py`, `test_pmag_projection.py`	64	`dir2cart`, `cart2dir`, `angle`, `dotilt`, `dogeo`, `doflip`, `flip`, `dimap`
Statistics	`test_pmag_statistics.py`, `test_ipmag_statistics.py`, `test_pmag_kent_bingham.py`	46	`fisher_mean`, `doprinc`, `Tmatrix`, `tauV`, `dokent`, `dobingham`, `bingham_mean`, `kent_mean`
PCA / specimen fitting	`test_pmag_domean.py`	17	`domean` (DE-FM, DE-BFL, DE-BFL-A, DE-BFP, DE-BFL-O)
Random sampling	`test_pmag_random.py`, `test_ipmag_random.py`	27	`fshdev`, `fishrot`, `fisher_mean_resample`, `kentrot`, `tk03`
VGP / paleogeography	`test_pmag_vgp.py`, `test_ipmag_vgp.py`	35	`pinc`, `plat`, `dia_vgp`, `vgp_di`, `b_vdm`, `vdm_b`, `vgp_calc`, `sb_vgp_calc`, `lat_from_inc`
Field models	`test_igrf.py`, `test_pmag_field_models.py`	24	`ipmag.igrf` (IGRF14 reference values), `pmag.doigrf` (property-based + paleo models)
E/I corrections	`test_ipmag_ei.py`	16	`squish`, `unsquish`, `f_factor_calc`, `find_f`, `find_ei`
Statistical tests	`test_ipmag_field_tests.py`, `test_ipmag_common_mean.py`	22	`bootstrap_fold_test`, `conglomerate_test_Watson`, reversal tests, common mean tests
Anisotropy	`test_pmag_anisotropy.py`	13	`s2a`, `a2s`, `doseigs`, `dosgeo`, `dostilt`
Data handling / I/O	`test_pmag_data_handling.py`, `test_pmag_io.py`, `test_pmag_utilities.py`, `test_ipmag_data_handling.py`	64	`magic_read`, `magic_write`, `get_dictitem`, `fillkeys`, `parse_site`, `make_di_block`
Plotting	`test_ipmag_plotting.py`	3	Smoke tests for `plot_net`, `plot_di`, `fishqq`
Correlation	`test_ipmag_correlation.py`	4	`simul_correlation_prob`, `rand_correlation_prob`

Test design highlights

Analytical tests where possible (cardinal directions, dipole formula, equal-area projection formula, roundtrips at 1e-10 tolerance)
IGRF14 reference values verified against the NOAA magnetic field calculator across 14 date/location combinations spanning 1999-2027.5
Cross-validation between pmag.py and ipmag.py implementations (e.g., lat_from_inc vs plat, vgp_calc vs dia_vgp)
Reproducible stochastic tests using random_seed parameter rather than global np.random.seed()
pytest infrastructure: conftest.py with Agg backend fixture, pytest.ini configuration

… (2023) functions np.matrix is deprecated and broken in modern numpy, causing failures in common_mean_bootstrap_H23, mean_bootstrap_confidence, and reversal_test_bootstrap_H23. Changes in pmag.py: - form_Mhat, form_Ghat, form_Q, find_CMDT_CR, find_T, find_CR: replace np.matrix() with np.asarray(), .getT() with .T, .getH() with .conj().T, and matrix * with @ operator - find_T: add condition number guard for singular covariance matrices that arise from degenerate bootstrap samples Changes in ipmag.py: - common_mean_bootstrap_H23: replace .getH()* with .conj().T @ - mean_bootstrap_confidence: filter degenerate bootstrap samples before computing quantile - common_mean_watson: fix '%.1f' % array formatting (use angle[0]) - lat_from_pole: fix float() on 1-element array (use paleo_lat[0])

* Plot map fix (#810) * develop some more tests * update plot_map to return axes * implement axes fix to show two continents * new tests removed from this plot_map_fix branch * use elope_path instead of rewriting Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * fix orthographic typo Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * edits to PR --------- Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> Co-authored-by: apivarunas <apiv93@gmail.com> * change from predetermined index for percentiles to direct percentile calculation in find_ei and find_ei_kent (#795) After reviewing this PR and concurring that it fixes the issue, merging. * remove duplication, fix docs, readability (#807) * simple style fixes via ruff (#812) * Working through errors / warnings in ipmag (#813) * slowly fixing errors and warnings * additional warning fixes * additional Exception clarity * clearing bare excepts * addressing review comments * adding ipmag tests * add to same file * added watson conglomerate tests * renamed file --------- Co-authored-by: Nick Swanson-Hysell <nicks-h@umn.edu> Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

…ag.py and ipmag.py

…ctions

Copilot

Pull request overview

This PR addresses issue #447 by adding an optional random_seed parameter (using the modern numpy.random.default_rng() API) to all stochastic functions in pmag.py, ipmag.py, and rockmag.py, and adds a comprehensive 369-test pytest suite covering core paleomagnetic functions. A _resolve_rng() helper is introduced in pmag.py to standardize seed handling across call chains.

Changes:

Add _resolve_rng() helper to pmag.py and convert 11 functions in pmag.py, 14 in ipmag.py, and 1 in rockmag.py to use it for reproducible random sampling
Add 369 tests across 24 new test files covering coordinate transforms, statistics, VGP, field models, E/I corrections, bootstrap tests, anisotropy, and data I/O
Remove the old test_fisher_sample.py (superseded by test_ipmag_random.py)

Reviewed changes

Copilot reviewed 28 out of 28 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Description
`pmagpy/pmag.py`	Adds `_resolve_rng()` and converts `fshdev`, `kentdev`, `gaussdev`, `get_unf`, `pseudosample`, `apseudo`, `mktk03`, `pseudo`, `di_boot`, `dir_df_boot`, `s_boot` to use it
`pmagpy/ipmag.py`	Converts `fishrot`, `fisher_mean_resample`, `kentrot`, `tk03`, `bootstrap_fold_test`, `mean_bootstrap_confidence`, `common_mean_bootstrap`, `common_mean_bootstrap_H23`, `common_mean_watson`, `reversal_test_bootstrap`, `reversal_test_bootstrap_H23`, `find_svei_kent`, `plate_rate_mc`, `MADcrit` (signature only) to accept `random_seed`
`pmagpy/rockmag.py`	Converts `backfield_MaxUnmix` bootstrap loop to use seeded RNG
`pytest.ini`	Configures test discovery and warning filters
`pmagpy/test/conftest.py`	Session-scoped fixture to force Agg matplotlib backend
`pmagpy/test/test_fisher_sample.py`	Removed (superseded by `test_ipmag_random.py`)
`pmagpy/test/test_pmag_*.py` (10 files)	New tests for pmag.py functions
`pmagpy/test/test_ipmag_*.py` (10 files)	New tests for ipmag.py functions

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

… TypeError on None

…py and ipmag.py Add random_seed parameter to scalc_vgp_df, find_ei, find_ei_kent, find_compilation_kent, reversal_test_MM1990, simul_correlation_prob, and rand_correlation_prob. Thread random_seed through internal calls to pseudo, fishrot, fisher_mean_resample, and common_mean_watson. Also fix verbose parameter not being forwarded in reversal_test_bootstrap and correct 'indicies' typos.

Swanson-Hysell · 2026-03-02T11:49:14Z

These four commits address the Copilot review feedback and extend random_seed coverage to functions that were missed in the initial pass:

Fix assert ordering bug in rockmag.py (a035e40)
Make correlation tests deterministic by seeding simul_correlation_prob and rand_correlation_prob (341fcd7)
Use tmp_path fixture instead of hardcoded /tmp in test_pmag_io.py (f6e16c8)
Add random_seed to 7 additional functions in pmag.py/ipmag.py: scalc_vgp_df, find_ei, find_ei_kent, find_compilation_kent, reversal_test_MM1990, simul_correlation_prob, rand_correlation_prob; also fix verbose not being forwarded in reversal_test_bootstrap and correct "indicies" typos (6e2aa47)

Remaining unseeded stochastic functions in svei.py (GGPrand, GGP_vMF_cdfs) are tracked separately in #825 .

Swanson-Hysell · 2026-03-02T11:50:39Z

@apivarunas The tests you wrote in that previous 5 tests file are now spread across the new tests files.

apivarunas · 2026-03-03T07:27:35Z

@Swanson-Hysell I've given this PR a once-over. Nice comprehensive random_seed implementation and test coverage! The tests ran in about 14.82 seconds on my machine. I made 2 commits to it:

Removing a legacy parameter (seed) in pmag.mtk03 which was explained via like 4 lines in a docstring but basically unnecessary. Instead, I had tk03.py use the same logic it always did, just to random_seed parameter. However, I think the tk03.py implementation is sort of odd. Something to check on, but not really in the scope of this PR.
Adjusting tolerance in the test test_north_pole_vgp_matches_pinc. I ran the test suite a few times and that test failed (max difference of 1.7e-06). Don't know why exactly it failed for me and not you, maybe some little floating-point differences? I boosted the tolerance to 2e-06 for that test.

If that's okay with you, I think you can go ahead and merge. Having this test coverage will be nice.

Swanson-Hysell · 2026-03-07T11:27:03Z

Thanks for the review and commits on this @apivarunas!

It looks like 8861e12 removed the legacy seed option from the docstring for the pmag.mktk03 function, but left it as a required position parameter in the function. Which I think that means that G2 would be received as seed in programs/tk03.py. I will dig in.

…ll of it

…mGGP

Swanson-Hysell and others added 17 commits February 28, 2026 06:38

develop some more tests

87adeed

Add pytest configuration and shared fixtures for testing

2f65d64

tiny test tweaks

ae79c7c

Merge branch 'master' into test_exploration

11620b3

Add tests for statistical and PCA functions in pmag.py

89e163c

Add tests for VGP, paleolatitude, and field intensity functions in pm…

b55aa55

…ag.py and ipmag.py

Add tests for data structure utility functions in pmag.py

e657944

Add tests for bootstrap fold test functionality in ipmag.py

bbbcede

Add random seed support for reproducibility across pmag and ipmag fun…

814e115

…ctions

Add random seed parameter to bootstrap_fold_test for reproducibility

dbdbc79

Add random seed parameter to backfield_MaxUnmix for reproducibility

90dde86

Major increase in test coverage across pmag.py and ipmag.py

02b7483

Merge branch 'master' into test_exploration

8a5e534

remove the IGRF-13 test as there are detailed IGRF-14 checks

416c4d4

small tweak to communicate that the IGRF test aren't using model 13

2210c2f

Swanson-Hysell requested review from apivarunas and Copilot March 1, 2026 17:49

Copilot started reviewing on behalf of Swanson-Hysell March 1, 2026 17:50 View session

Copilot AI reviewed Mar 1, 2026

View reviewed changes

Swanson-Hysell added 4 commits March 1, 2026 14:08

Moved assert parameters is not None before len(parameters) to prevent…

a035e40

… TypeError on None

Updated reference value tests to use seeded, deterministic assertions

341fcd7

Replaced hardcoded /tmp/empty.txt with tmp_path fixture

f6e16c8

apivarunas added 2 commits March 2, 2026 22:55

removing unnecessary legacy option

8861e12

tolerance adjustment

f791633

Swanson-Hysell added 2 commits March 7, 2026 05:42

remove the legacy seed parameter from pmag.tk03 and the ipmag.tk03 ca…

5eb0c2a

…ll of it

update docstring to reflect removal of legacy seed parameter from s_l…

e2b5ca2

…mGGP

Swanson-Hysell merged commit 81b7647 into master Mar 7, 2026
2 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add reproducible random sampling and many new pmag.py and ipmag.py tests#826

Add reproducible random sampling and many new pmag.py and ipmag.py tests#826
Swanson-Hysell merged 25 commits into
masterfrom
test_exploration

Swanson-Hysell commented Mar 1, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Swanson-Hysell commented Mar 2, 2026

Uh oh!

Swanson-Hysell commented Mar 2, 2026

Uh oh!

apivarunas commented Mar 3, 2026

Uh oh!

Swanson-Hysell commented Mar 7, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

Swanson-Hysell commented Mar 1, 2026

Summary

Reproducible random sampling

Converted functions

Out of scope (cosmetic/plotting random, not scientific)

Test suite

Test design highlights

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Swanson-Hysell commented Mar 2, 2026

Uh oh!

Swanson-Hysell commented Mar 2, 2026

Uh oh!

apivarunas commented Mar 3, 2026

Uh oh!

Swanson-Hysell commented Mar 7, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants