Version History

2.1.0

  • Bug Fix

    • Breaking numerical change: Fix missing gamma(1/n) factor in the Stretched Exponential (SE) closed-form cumulative and EUR. SE._Nfn returned qi*tau/n * P(1/n, (t/tau)^n) using the regularised lower incomplete gamma (scipy gammainc = P), but the integral of qi*exp(-(t/tau)^n) is qi*tau/n * gamma(1/n) * P(1/n, (t/tau)^n). Cumulative volume and EUR were wrong by a factor of gamma(1/n) for any n != 0.5 — understated up to ~64% at n=0.3, exact at n=0.5, overstated ~10% for n > 0.5. Any SE forecast with n != 0.5 will now report different cumulative/EUR values. For very small n, gamma(1/n) overflows (the closed-form EUR genuinely diverges there); SE now falls back to the bounded numerical integrator.

  • Performance

    • Expose n_grid in the numerical integrator via cum(t, n_grid=...) (and the other volume methods). Default 10,000 is unchanged; a smaller value (e.g. 2,000, ~5e-5 relative error) trades accuracy for a proportional speed-up on the numerically-integrated path (PLE, associated yields). n_grid < 2 raises ValueError.

  • Other changes

    • Remove unreachable pass (and commented-out lines) after a raise in THM._validate; flagged as unreachable by mypy. No behaviour change.

    • Add test_SE_cum_matches_integral and test_PLE_cum_matches_integral, which compare cum against adaptive quadrature of the rate — the previous check_model assertions only verified finiteness, so a constant-factor error in a closed-form cumulative passed silently.

2.0.0

  • Breaking: Minimum dependency versions raised

    • Require numpy >= 2.1 (was >= 1.21.1)

    • Require scipy >= 1.13 (was >= 1.7.1)

    • Require Python >= 3.10 (was >= 3.7)

    • ParamDesc.naive_gen type changed from RandomState to Generator

  • Deprecation fixes

    • Replace np.bool_ with bool in dtype specification

    • Migrate numpy.random.RandomState to numpy.random.Generator

1.3.0

  • Performance

    • Vectorize _integrate_with using cumulative_trapezoid on a log-spaced grid — eliminates per-interval Python loop (~3x faster, comparable accuracy)

    • Vectorize bourdet() using searchsorted — eliminates per-point Python helper calls

  • Bug Fix

    • Fix bourdet() producing NaN at right-edge points when smoothing window exceeds array boundary

    • Fix bourdet() incorrect backward-difference at right edge (_get_R_der selected point i itself, causing division by zero)

  • Numerical stability

    • Replace np.log(1 + x) with np.log1p(x) in _qcheck and _Ncheck for small decline rates

    • Replace np.log(1 + D*b) with log1p(D*b) in secant_from_nominal

    • Use np.expm1 for t^(1-m) - 1 in Duong model for precision near t=1

1.2.0

  • Build system

    • Migrate from setup.py / setup.cfg to pyproject.toml

    • Replace flake8 with ruff

    • Version now single-sourced in pyproject.toml, resolved at runtime via importlib.metadata

  • Bug Fix

    • Fix overflow in MultisegmentHyperbolic._Ncheck when decline rate is near-zero (subnormal), causing cum(0) to return NaN

  • Other changes

    • Fix mypy strict-mode type errors in bourdet.py and base.py

    • Update CI and test scripts to use ruff

1.1.0

  • Bug Fix

    • Fix bug in sign in MultisegmentHyperbolic.secant_from_nominal

  • Other changes

    • Add mpmath to handle precision requires of THM transient functions (only required to use the functions)

    • Adjust default degree of THM transient function quadrature integration from 50 to 10 (scipy default is 5)

    • Update package versions for docs and builds

    • Address various floating point errors, suppress numpy warnings for those which are mostly unavoidable

    • Add test/doc_exapmles.py and update figures (not sure what happened to the old file)

    • Adjust range of values in tests to avoid numerical errors in numpy and scipy functions… these were near-epsilon impractical values anyway

1.0.8

  • New functions

    • Added WaterPhase.wgr method

  • Other changes

    • Adjust yield model rate function to return consistent units if primary phase is oil or gas

    • Update to numpy v1.20 typing

1.0.7

  • Allow disabling of parameter checks by passing an interable of booleans, each indicating a check

    to each model parameter.

  • Explicitly handle floating point overflow errors rather than relying on numpy.

1.0.6

  • New functions

    • Added WaterPhase class

    • Added WaterPhase.wor method

    • Added PrimaryPhase.add_water method

  • Other changes

    • A yield model may inherit both SecondaryPhase and WaterPhase, with the respective methods removed upon attachment to a PrimaryPhase.

1.0.5

  • New functions

    • Bourdet algorithm

  • Other changes

    • Update docstrings

    • Add bourdet data derivatives to detailed use examples

1.0.4

  • Fix typos in docs

1.0.3

  • Add documentation

  • Genericize numerical integration

  • Various refactoring

0.0.1 - 1.0.2

  • Internal releases