gamrs
Generalised Additive Models in Rust — clean-room reimplementation built on
six composable trait layers (Basis, BasisTransform, Loss/Link/VarianceFn,
InnerSolver, ScoreDerivatives, OuterSolver). Designed for parity with
R's mgcv.
Status: beta (v0.14). Fastest where wall time matters — large n and
high basis dimension (up to ~2.3× at n=1M and ~15× at k=50 vs
mgcv_rust 0.23), competitive-to-slower on tiny fits, with a known
multi-smooth-NegBin gap; see Performance for the honest
breakdown. mgcv R-parity on µ across all ten families. Multi-smooth
additive (y ~ s(x0) + s(x1)),
n-margin tensor products (te(x0, x1, …) / ti(…)) and thin-plate splines
(s(x0, x1, bs="tp")) all ship. NegBin and Tweedie fit multi-smooth, with
Tweedie offering both profile-p (tw()) and fixed-p (Tweedie(p)).
Install
pip install gamrs # base wheel
pip install gamrs[quantile] # + scipy for SHASH-calibrated quantile fits
Quickstart
from gamrs import Gam, CrTerm, TeTerm
# Single 1-D smooth, Gaussian
g = Gam(family="gaussian").fit(X, y)
mu = g.predict(X)
mu, lo, hi = g.predict_ci(X, level=0.95)
# Multi-smooth additive
g = Gam(terms=[CrTerm("x0", k=10), CrTerm("x1", k=15)]).fit(df, df["y"])
# Tensor product
g = Gam(terms=[TeTerm(cols=("x0", "x1"), k=(5, 5))]).fit(df, df["y"])
# Large-n GLM — switch to the bam()-style fREML optimiser. Ported to the
# GLM envelope driver only; elsewhere it warns and fits on REML instead.
g = Gam(family="poisson", method="fREML").fit(X_big, y_big)
Full walkthrough: docs/quickstart.md.
Optimiser & large-n notes: docs/perf.md.
Families
All ten families land 1-D parity against mgcv:
| Family | Link | Inner solver | Outer optimiser | Parity (µ rel-err) |
|---|---|---|---|---|
| Gaussian | identity | one-Cholesky | 1-D Newton | ~3e-6 |
| Bernoulli | logit | PIRLS | 1-D Newton | ~1e-3 |
| Poisson | log | PIRLS | 1-D Newton | ~8e-5 |
| QuasiPoisson | log | PIRLS | 1-D Newton (prof φ) | ~2e-4 |
| QuasiBinomial | logit | PIRLS | 1-D Newton (prof φ) | ~7e-5 |
| Gamma | log | PIRLS | 1-D Newton (prof φ) | ~2e-2 |
| InverseGaussian | log | PIRLS | 1-D Newton (prof φ) | ~3e-4 |
| NegBin | log | PIRLS | ρ-Newton + profile-θ | ~9e-7 |
| Tweedie | log | PIRLS | 3-D joint Newton | ~5e-3 |
TDist (scat) |
identity | PIRLS | 3-D joint Newton | ~4e-4 |
| Ocat | logit | gam.fit5 | joint β + threshold | smoke |
| Quantile (ELF) | identity | Armijo BT | ρ-Newton (per term) | qgam OOS ~1.00× |
Multi-smooth (s(x0) + s(x1) + …) ships with mgcv R parity tests for
Gaussian / Bernoulli / Poisson / QuasiPoisson / QuasiBinomial / Gamma /
InvGauss / NegBin / Tweedie / scat. scat / TDist multi-smooth now has mgcv
reference parity tests too — as of 0.14.0, 2-D µ rel-err ~5.5e-6 and 3-D
~8.3e-5, with σ̂² matching mgcv to ~0.1% (tests/parity_additive_scat.rs,
scripts/r/gen_scat_multismooth_fixtures.R). Those bounds tightened 8-16× in
0.13.0 when a spurious rank adjustment came out of scat's REML score, and
another 30-200× in 0.14.0 when scat's REML criterion was corrected to
mgcv's observed-curvature log|H|; the same
release adds tests/parity_scat_tf9963.rs, a saturated-basis lock (5 distinct
x against k = 5, the geometry that exposed the defect) against all three mgcv
arms (gam+REML, bam+REML, bam+fREML). Quantile/ELF now fits
multi-smooth additive too (y ~ s(x0) + s(x1) + … via the terms= arg of
fit_quantile): on a 2-D additive heteroskedastic split its out-of-sample
pinball loss matches qgam to within ±0.6% at τ ∈ {0.1, 0.5, 0.9}
(scripts/r/gen_quantile_multismooth_fixture.R,
test_parity_multismooth.py::test_additive_quantile_oos_parity).
For a coherent set of quantiles, fit_quantile_lss fits the conditional
distribution by its location μ(x) and scale σ(x) and derives every quantile
as q_τ(x) = μ(x) + σ(x)·z_τ — the mgcv gaulss/shash view. One fit serves
all τ, the bands never cross, and shape="shash" captures skew/kurtosis.
Matches mgcv gaulss OOS pinball to within ~1% on a heteroskedastic 2-D split
(scripts/r/gen_quantile_lss_fixture.R).
GAMLSS — Gaussian location-scale (gaulss)
fit_gaulss(X, y, mu_terms=…, sigma_terms=…) is the first GAMLSS
(multi-linear-predictor) family: it fits y ~ N(μ(x), σ(x)²) with smooth μ(x)
and σ(x) jointly. Because the Gaussian location-scale Fisher information
is block-diagonal (μ ⟂ log σ), the joint MLE is an orthogonal alternation
of two single-predictor weighted-Gaussian REML fits — reusing the existing fit
stack rather than a dense block-Newton. One fit gives every quantile
(q_τ = μ + σ·Φ⁻¹(τ), no crossing). It recovers mgcv gaulss's μ̂/σ̂ to
RMSE ~3e-4 / ~1e-3 and its OOS pinball to ~0.05%, at ~70× the speed (n=800).
Unlike the two-stage fit_quantile_lss, μ is reweighted by 1/σ²(x) each pass —
the joint-MLE efficiency gain. This is the seam for the wider GAMLSS class
(shash, gevlss): non-orthogonal families extend the same alternation.
GAMLSS — sinh-arcsinh (shash)
fit_shash(X, y, mu_terms=…, tau_terms=…, eps_terms=…, phi_terms=…) fits the
first non-orthogonal GAMLSS family — all four sinh-arcsinh parameters as
smooths, jointly: location μ(x), log-scale τ(x) (σ = exp τ), skewness ε(x),
and log-kurtosis φ(x). Because shash's Fisher information couples the predictors
(unlike gaulss), β for all four is solved together by a dense penalised
block-Newton, under outer REML/LAML with an analytic gradient (third
derivatives → d log|Hp|/dρ). One fit yields every quantile
(predict_quantile, the R .shashQf inverse-CDF) plus predict_params →
(μ, σ, ε, δ). It recovers mgcv gam(…, family=shash, bs="cr") to ~1e-6 (η,
smoothing params, EDF, quantiles) and fits Gaussian, skewed, and heavy-tailed
data alike. Distinct from the two-stage fit_quantile_lss(shape="shash") (one
global shape) — this is the genuine joint GAMLSS. (v1: one smooth term per
predictor.) Term specs use the same form as fit_additive, e.g.
mu_terms=[CrTerm(0, k=10)], tau_terms=[CrTerm(1, k=10)], eps_terms=[].
Multi-smooth Ocat now has an mgcv reference parity test
(tests/parity_additive_scat.rs's ocat sibling in
test_parity_multismooth.py::test_additive_ocat_parity,
scripts/r/gen_ocat_multismooth_fixtures.R): on a well-posed
noisy-latent DGP gamrs and mgcv ocat(R=4) both converge cleanly and
agree on predict_proba to ~1.8e-3 mean abs / 98% class agreement.
v0.10 ports the full mgcv R outer-Newton stabilisation stack (smart
θ-init from category frequencies, diagonal Hessian preconditioning,
Gill-Murray-Wright eigen-fix, subset Newton, rank-deficient KKT
convergence check). After the ports, single- and (well-posed)
multi-smooth ocat converge cleanly. The residual converged_=False
appears only on near-separable fixtures (noiseless quantile-cut
categories) where the latent scale wants to blow up — the exact regime
mgcv itself either converges to a degenerate θ≈181 solution or aborts
with "inner loop 1; can't correct step size". There gamrs's θ∈(−3,3)
bound keeps it stable (99% accuracy) and the conservative flag is
correct. See ~/ObsidianVault/Projects/gamrs/gamrs - mgcv outer-Newton stabilisation techniques (port catalogue) 2026-06-03.md for the
full port story.
Smooths
- Single 1-D —
s(x0)viaCrTerm(cubic regression spline default). - Additive multi —
s(x0) + s(x1) + s(x2). - Tensor product —
te(x0, x1, …)viaTeTerm,ti(…)viaTiTerm, any n-margin. - Thin-plate —
s(x0, x1, bs="tp")viaTpsTerm. - Random effects —
s(g, bs="re")viaReTerm. - Point constraint —
s(x, pc=v)viaCrTerm("x", pc=v)orterm_pc_mapping={"x": v}: the smooth passes through zero atx = vand the intercept takes the level. Predictions, λ̂ and edf are unchanged — only the intercept-versus-smooth split moves, which is what a caller reading a partial curve sees. CR bases (cr,cr_stable) only. - Numeric
by=variable —s(x, by=w)viaCrTerm("x", by="w")orterm_by_mapping={"x": "w"}: the term contributesw * f(x), exactly 0 wherew = 0. As in mgcv it is not centred unlesswis constant. Thewcolumn feeds the smooth but is not a term of its own, because an uncentreds(x, by=w)already spansw * constant. Name it inpredictor_basis_mapto add it anyway; it is then aliased and fitted as 0. An unusableterm_by_mappingentry raises. Not combinable withpc. On a 2-margin tensor,te(x0, x1, by=w)viaTeTerm(cols=("x0", "x1"), by="w");wmay be one of the margins. On a random effect,s(g, by=w, bs="re")viaReTerm("g", by="w")orpredictor_basis_map={"g": "re"}withterm_by_mapping={"g": "w"}. A random effect is fully penalised, so namingwas a parametric term is not aliased:golf + s(course, by=golf, bs="re")fits the golf mean plus per-course deviations, and a placeholder course seen only on non-golf rows is fitted as 0. CR smooths,TeTermandReTermonly. - Parametric (linear) — unsmoothed raw column via
ParametricTermorpredictor_basis_map={"x": "parametric"}(alias"linear"). Use for 0/1 indicators, counts, or anything you want unpenalised. mgcv R's "pterms" block.
Performance
gamrs vs mgcv_rust 0.23.2, best-of-7 median wall time after 3 warmup
iters (>1× = gamrs faster). Reproduce with scripts/bench_matters.py.
Numbers below are an i7-8565U (8th-gen, AVX2); the ratios are
same-box-comparable but absolute times differ on your hardware.
gamrs wins at scale. Single-smooth, k=20 — as n grows the constant-factor setup overhead amortises and gamrs pulls ahead:
| family | n=10K | n=100K | n=1M |
|---|---|---|---|
| Gaussian | 0.58× | 1.49× | 2.27× |
| Poisson | 0.24× | 1.01× | 1.89× |
| Bernoulli | 0.39× | 1.12× | 1.84× |
It also wins as the basis dimension k grows (Gaussian, n=2K): k=10 →
1.7×, k=20 → 4.0×, k=50 → 15×. (Those fits are <2ms — read them as
above-the-noise ratios, not headline wall-time claims.)
Shape-aware families (n=2K, k=10):
| family | speedup | note |
|---|---|---|
| Tweedie 1-D | 1.93× | |
| ocat 1-D | 14.5× | mgcv_rust's ocat path is slow by construction |
| ocat 2-D | 0.69× | |
| NegBin 1-D | 0.64× | |
| NegBin 2-D | 0.06× | multi-smooth profile-θ scaling gap (known) |
| scat 1-D | 0.58× | <2ms — sub-noise |
Honest aggregate (16 above-20ms cells across all sweeps): gamrs is faster in 8 — median 0.89×, geomean 0.83×, range 0.06×–14.5×. The wins concentrate where wall time actually matters — large n, large k, and Tweedie/ocat. Tiny fits (<20ms) and multi-smooth NegBin still trail mgcv_rust; the NegBin multi-smooth slowdown is a known profile-θ scaling issue under investigation.
scat climbed from v0.10.0's 0.07× to v0.11's ~0.77× (at the 6K-row profiling fixture) via analytic gradient + Level-2 analytic Hessian + observed-W PIRLS + warm-start (v0.11.0) and broadcast-expression conversion + batched-h_diag matmul (v0.11.1). The residual gap is per-pair Hessian-assembly work at small p.
For GLM families at large n, set method="fREML" (mgcv R's bam()
equivalent — Wood & Fasiolo 2017 Fellner-Schall multiplicative updates
with single-step IRLS per outer iteration). The defaults are sensible
at small/medium n; the perf guide covers when to switch.
Fellner-Schall reached the GLM envelope driver only — on the gaussian
closed-form path, the quantile path and the shape-parameter families
(scat / negbin / tweedie / ocat), method="fREML" warns and fits on REML.
Both optimise the same REML criterion and damped Newton is the stronger
route to it, so the substitution costs nothing; it is warned about because
until 0.13.0 it happened silently.
Parallel fits across threads
As of v0.11.7 the fit releases the GIL (PyO3 py.detach) for the entire
solve, so independent Gam.fit(...) / fit_quantile(...) calls run truly
concurrently on a ThreadPoolExecutor — no process pool, no pickling of
inputs. When fanning many fits across a thread pool, set
OPENBLAS_NUM_THREADS=1 so the BLAS backend doesn't oversubscribe cores
against your own threads: on a 6K-row scat fit that turns the pre-0.11.7
0.58× (GIL-bound, serialised) into ~1.95× on 4 worker threads.
Rust API
use gamrs::{TermSpec, MarginKind, DesignStrategy};
use ndarray::Array2;
let x: Array2<f64> = /* (n, n_input_dims) */;
let y = /* Array1<f64> */;
let fit = gamrs::fit(gamrs::family::gaussian_identity(), x.view(), y.view(), None, 10)?;
let fit = gamrs::fit_with_design(
gamrs::family::gaussian_identity(),
DesignStrategy::Additive { terms: vec![
TermSpec::Cr { col: 0, k: 10 },
TermSpec::Cr { col: 1, k: 15 },
]},
x.view(), y.view(), None,
)?;
let fit = gamrs::fit_with_design(
gamrs::family::gaussian_identity(),
DesignStrategy::Additive { terms: vec![
TermSpec::Tensor { col_a: 0, col_b: 1, k_a: 5, k_b: 5,
bs_a: MarginKind::Cr, bs_b: MarginKind::Cr },
]},
x.view(), y.view(), None,
)?;
let mu = fit.predict(x.view())?;
Python API
PyO3 bindings + numpy. sklearn-like surface: fit / predict /
predict_ci / predict_diff / vcov_ / coef_ / lambda_ /
edf_ / fit_stats_, plus serialize / deserialize and GamPredictor
for inference-only deployment.
Architecture
Trait layering (src/traits.rs):
Layer 1 Basis ← CrBasis, RandomEffectsBasis, TensorProductBasis<A, B>
Layer 1.5 BasisTransform ← SumToZero, StableReparam
Layer 2 Loss/Link/Variance ← 10 families (see table above)
Layer 3 InnerSolver ← GaussianClosedFormInner, PirlsInner, GamFit5Inner, ArmijoInner
Layer 4 ScoreDerivatives ← EnvelopeScore, ShapeAwareEnvelopeScore
Layer 5 OuterSolver ← NewtonWithHalving, FellnerSchall
Layer 6 FittedGam ← predict, predict_ci, predict_diff, vcov, serialize
Outer optimisers (Newton, Fellner-Schall) and per-family tolerances are
selected through Loss::outer_tuning() and Loss::allows_no_refresh(),
so adding a family is a Loss impl, not a fork of the optimiser.
Versioning
Beta (0.14.x). The API is stabilising; minor bumps may carry breaking
changes until the 1.0 surface is locked. All ten families plus Ocat and
Quantile/ELF now fit multi-smooth additive designs.
License
MIT.
Metadata
Release files for gamrs 0.16.3
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| gamrs-0.16.3.tar.gz | 4.4 MB | Without provenance | Details for gamrs-0.16.3.tar.gz |
Built distributions (wheels)
Total release size: 147.4 MB
Release files / gamrs-0.16.3.tar.gz
| Download URL | gamrs-0.16.3.tar.gz |
|---|---|
| Size | 4.4 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
51f7d2679555d9b229454973c979bc76d330ea4a05b6a75d3869d64edbf0dd13
|
|
BLAKE2b-256 checksum How to use checksums |
cb6dc8c949c6b6b564c2ba463be4573705bdbd6bc1974579f3833e3a297d01ec
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-pp311-pypy311_pp73-manylinux2014_x86_64.manylinux_2_17_x86_64.whl
| Download URL | gamrs-0.16.3-pp311-pypy311_pp73-manylinux2014_x86_64.manylinux_2_17_x86_64.whl |
|---|---|
| Size | 6.9 MB |
| Tags | Linux glibc 2.17+ x86-64 PyPy 3.11 PyPy 3.11 7.3 |
|
SHA-256 checksum How to use checksums |
61b99c802dd6c5be9b4e0f83d6c069df975dd25b8b39ecd8dc370b9e5357f638
|
|
BLAKE2b-256 checksum How to use checksums |
0398ef3e470d4ab093332242005a35b6262bcd8f40462bd76e8f6d81621ffb39
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-cp315-cp315-win_amd64.whl
| Download URL | gamrs-0.16.3-cp315-cp315-win_amd64.whl |
|---|---|
| Size | 4.2 MB |
| Tags | CPython 3.15 Windows x86-64 |
|
SHA-256 checksum How to use checksums |
7177adbb50b4567f6e4854848a5abc36bb2d8fd5c7227bffb0dbebe56e5964f7
|
|
BLAKE2b-256 checksum How to use checksums |
93ca1d83bb397add6ad5bff97b1ac31b9aeaebe7e7da0d285f2f4a7df016bbc2
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.whl
| Download URL | gamrs-0.16.3-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.whl |
|---|---|
| Size | 6.9 MB |
| Tags | CPython 3.14 CPython 3.14 free-threading Linux glibc 2.17+ x86-64 |
|
SHA-256 checksum How to use checksums |
4690c500b7da48bd4122553522b5fb065f40c758ef0ea3891b79800a6971491b
|
|
BLAKE2b-256 checksum How to use checksums |
24efd5b418398d048b7abeb80d9f561ce8da37caf12da46b61c4bf62ec2e8625
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-cp314-cp314-win_amd64.whl
| Download URL | gamrs-0.16.3-cp314-cp314-win_amd64.whl |
|---|---|
| Size | 4.2 MB |
| Tags | CPython 3.14 Windows x86-64 |
|
SHA-256 checksum How to use checksums |
d439c1825a3f1092adcc6a27eefb43f7bf6c2e14a07285b68f8ba24e19897bee
|
|
BLAKE2b-256 checksum How to use checksums |
ae28120aa43d1d5498cc63498d67fc2834bfe3741a0257d35fb6d9593f5ad6ad
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.whl
| Download URL | gamrs-0.16.3-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.whl |
|---|---|
| Size | 6.9 MB |
| Tags | CPython 3.14 Linux glibc 2.17+ x86-64 |
|
SHA-256 checksum How to use checksums |
7b7ae0cbc87bea91a7565dd052244482571842eedf05164658ef09f0047246d6
|
|
BLAKE2b-256 checksum How to use checksums |
bcbe4bcec7b726e62bca267b19ade939855159d690c5d05146f230434d5ddb13
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-cp314-cp314-macosx_11_0_arm64.whl
| Download URL | gamrs-0.16.3-cp314-cp314-macosx_11_0_arm64.whl |
|---|---|
| Size | 5.7 MB |
| Tags | CPython 3.14 macOS 11.0+ ARM64 |
|
SHA-256 checksum How to use checksums |
efdf8a501b9d3d726b91b1e33bf0900230bbc6b628df87a310ae4898af4d3afd
|
|
BLAKE2b-256 checksum How to use checksums |
739864cb6ac1561c4c8c62a57e098f922eecbd17e5b55ff8877384b5b4ed6f64
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-cp314-cp314-macosx_10_13_x86_64.whl
| Download URL | gamrs-0.16.3-cp314-cp314-macosx_10_13_x86_64.whl |
|---|---|
| Size | 7.4 MB |
| Tags | CPython 3.14 macOS 10.13+ x86-64 |
|
SHA-256 checksum How to use checksums |
cbbfed8c3484544b2b00d58a6b39f151cdb1e46f7ecea38e21593ede5e545de4
|
|
BLAKE2b-256 checksum How to use checksums |
245a1b6c7512923aea601ba6dfe5c04b33acc24ecad6fa5f03bd544af4c09931
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-cp313-cp313-win_amd64.whl
| Download URL | gamrs-0.16.3-cp313-cp313-win_amd64.whl |
|---|---|
| Size | 4.2 MB |
| Tags | CPython 3.13 Windows x86-64 |
|
SHA-256 checksum How to use checksums |
b51363c97c32997e953ce85d604d8b0e4b2c98a17e09555e09a233bf9d250ca0
|
|
BLAKE2b-256 checksum How to use checksums |
40361b6abcf0332c7395d0d61662ac1920176ef89b018347afeeb3d3b9e341ef
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.whl
| Download URL | gamrs-0.16.3-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.whl |
|---|---|
| Size | 6.9 MB |
| Tags | CPython 3.13 Linux glibc 2.17+ x86-64 |
|
SHA-256 checksum How to use checksums |
1a282c0901430629a4e6db2d9b802182d3f10ec51d4cba9a299029224dd1d0a3
|
|
BLAKE2b-256 checksum How to use checksums |
895446440aa8b5042ed7c4980c3271cc901d44987c5b370ce21b7a1d34c83db0
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-cp313-cp313-macosx_11_0_arm64.whl
| Download URL | gamrs-0.16.3-cp313-cp313-macosx_11_0_arm64.whl |
|---|---|
| Size | 5.7 MB |
| Tags | CPython 3.13 macOS 11.0+ ARM64 |
|
SHA-256 checksum How to use checksums |
10ab424c375454bbb25fbebf3152637bc4debd17ae9b5077329fdfddd0727257
|
|
BLAKE2b-256 checksum How to use checksums |
6b924b98945a7edb6e2247f982628325a92d3a1e92759093704d34336350e746
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-cp313-cp313-macosx_10_13_x86_64.whl
| Download URL | gamrs-0.16.3-cp313-cp313-macosx_10_13_x86_64.whl |
|---|---|
| Size | 7.4 MB |
| Tags | CPython 3.13 macOS 10.13+ x86-64 |
|
SHA-256 checksum How to use checksums |
987e38f1438b0ac54ce4159c14435223d896e160c2296299dedda945151af4ef
|
|
BLAKE2b-256 checksum How to use checksums |
e2c243677f8867a11c2e2c96e50dc344605e34f727bb17c064e0df568e4848ff
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-cp312-cp312-win_amd64.whl
| Download URL | gamrs-0.16.3-cp312-cp312-win_amd64.whl |
|---|---|
| Size | 4.2 MB |
| Tags | CPython 3.12 Windows x86-64 |
|
SHA-256 checksum How to use checksums |
0e48bc26b57e823e9fd24fe3d7b53acfb1b8b00865235054a4fb4c462895dbdb
|
|
BLAKE2b-256 checksum How to use checksums |
6c14b5f93678a332807be5f2a8f0448596e9727a93cb6f0e711c833bd27c822b
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.whl
| Download URL | gamrs-0.16.3-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.whl |
|---|---|
| Size | 6.9 MB |
| Tags | CPython 3.12 Linux glibc 2.17+ x86-64 |
|
SHA-256 checksum How to use checksums |
52eecfb097eed5dc0e198e9a872741647cc90a11dee1b654d7642ef960704d4d
|
|
BLAKE2b-256 checksum How to use checksums |
ba3eb9635a74fabe3ea3dc33c3d62686cb1dc52e58f8c4646dba79bd055d3c78
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-cp312-cp312-macosx_11_0_arm64.whl
| Download URL | gamrs-0.16.3-cp312-cp312-macosx_11_0_arm64.whl |
|---|---|
| Size | 5.7 MB |
| Tags | CPython 3.12 macOS 11.0+ ARM64 |
|
SHA-256 checksum How to use checksums |
c2b972b0de08e24ce7a82dbf1329c92a90a2e559b06d8db5f4aa900e2cdbcee9
|
|
BLAKE2b-256 checksum How to use checksums |
6519b67cad945df9dfb7121d39ff4ef297e66b84c28f6c8b6cc8d6ce076a3fc2
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-cp312-cp312-macosx_10_13_x86_64.whl
| Download URL | gamrs-0.16.3-cp312-cp312-macosx_10_13_x86_64.whl |
|---|---|
| Size | 7.4 MB |
| Tags | CPython 3.12 macOS 10.13+ x86-64 |
|
SHA-256 checksum How to use checksums |
c88a6135ce0d77944cc775f6f21d2d97ba039f1217cc5011a54fad7bf1c8c80c
|
|
BLAKE2b-256 checksum How to use checksums |
65edab9dffdcbeab7e2caf8a234709c3a2cd2acd8085a0f28a83e18b970eae8f
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-cp311-cp311-win_amd64.whl
| Download URL | gamrs-0.16.3-cp311-cp311-win_amd64.whl |
|---|---|
| Size | 4.2 MB |
| Tags | CPython 3.11 Windows x86-64 |
|
SHA-256 checksum How to use checksums |
b67c4be6350bc2cac53ee913e8eb4b8227fee1e2af115468b0a1a4a2d110e34e
|
|
BLAKE2b-256 checksum How to use checksums |
c0d95e5f820467cce1f54fd4bb647bd5c44e0846c1c208e6eeb21285f8005ad5
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.whl
| Download URL | gamrs-0.16.3-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.whl |
|---|---|
| Size | 6.9 MB |
| Tags | CPython 3.11 Linux glibc 2.17+ x86-64 |
|
SHA-256 checksum How to use checksums |
cb89a55796b43ddf0c4b2712a0eb16a6279cd37276021c9bc7869618799fe54f
|
|
BLAKE2b-256 checksum How to use checksums |
2ce5d72cb3f881fa9064ce887929da268d3314e8967c8f543fef3db9f9cfb739
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-cp311-cp311-macosx_11_0_arm64.whl
| Download URL | gamrs-0.16.3-cp311-cp311-macosx_11_0_arm64.whl |
|---|---|
| Size | 5.7 MB |
| Tags | CPython 3.11 macOS 11.0+ ARM64 |
|
SHA-256 checksum How to use checksums |
29f3af8470c6f6e3b6f4babb39840f800d8564f8f97b2267e649f8b1c32d3633
|
|
BLAKE2b-256 checksum How to use checksums |
d70a5850b13ce8631359fc55160929bc80b7357604649ba13b504b214609d98e
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-cp311-cp311-macosx_10_13_x86_64.whl
| Download URL | gamrs-0.16.3-cp311-cp311-macosx_10_13_x86_64.whl |
|---|---|
| Size | 7.4 MB |
| Tags | CPython 3.11 macOS 10.13+ x86-64 |
|
SHA-256 checksum How to use checksums |
1ab72f6c2f5e1586ae2f435af51e7ab8b74ea9c2b2a9ff69b73de023fbc237db
|
|
BLAKE2b-256 checksum How to use checksums |
825306968c64ea01e72f1367ca09903e4d594690dbf2165e959779bd370be7c1
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.whl
| Download URL | gamrs-0.16.3-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.whl |
|---|---|
| Size | 6.9 MB |
| Tags | CPython 3.10 Linux glibc 2.17+ x86-64 |
|
SHA-256 checksum How to use checksums |
80366fa46ee65b634f8df9e668b10c2a7e25e2db0963a877b7163a4f2dcd90c5
|
|
BLAKE2b-256 checksum How to use checksums |
8047b7dc4592922e367722e007572a9189c2847ca3f549b38bd635a850ff9850
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-cp310-cp310-macosx_10_13_x86_64.whl
| Download URL | gamrs-0.16.3-cp310-cp310-macosx_10_13_x86_64.whl |
|---|---|
| Size | 7.4 MB |
| Tags | CPython 3.10 macOS 10.13+ x86-64 |
|
SHA-256 checksum How to use checksums |
b59b464de8f63a254385c7b09b7457e61a4725711de9b06daf4712a75ba740d0
|
|
BLAKE2b-256 checksum How to use checksums |
3cfe899e3c8bbcbc8c89ff9aafa7e8b588c91d0c969a1067d8c22b5c03a3f857
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-cp39-cp39-manylinux2014_x86_64.manylinux_2_17_x86_64.whl
| Download URL | gamrs-0.16.3-cp39-cp39-manylinux2014_x86_64.manylinux_2_17_x86_64.whl |
|---|---|
| Size | 6.9 MB |
| Tags | CPython 3.9 Linux glibc 2.17+ x86-64 |
|
SHA-256 checksum How to use checksums |
32084a3d3ab61e35769cef9e2dd26687077a9a889b7f8f0acbe9686ae051b7d4
|
|
BLAKE2b-256 checksum How to use checksums |
c82ec31d63e53bb1b66ede6ef87c20a8a4309feb61ab520640a264b006383678
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Release files / gamrs-0.16.3-cp39-cp39-macosx_10_13_x86_64.whl
| Download URL | gamrs-0.16.3-cp39-cp39-macosx_10_13_x86_64.whl |
|---|---|
| Size | 7.4 MB |
| Tags | CPython 3.9 macOS 10.13+ x86-64 |
|
SHA-256 checksum How to use checksums |
aec3e34f4bbca2edaeffc80a8024fcf1e4d25f8cbf69bb732a2928dd1d079dd2
|
|
BLAKE2b-256 checksum How to use checksums |
e0fcc470aac80299c100601ad1c085927f9a2211c4d139bb412167b29f809019
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
maturin/1.16.0
|
Security for gamrs 0.16.3
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.