SEC.Models.SdmEstimator

SEC.Models.SdmEstimator#

SEC.Models.SdmEstimator.py

class LognormalEnv(N, T, me, mp, N0, t0, mu, sigma)#

Bases: tuple

Named-tuple form of the lognormal env-params 8-tuple (N, T, me, mp, N0, t0, mu, sigma).

Returned by estimate_sdm_lognormal_from_monopore and accepted by optimize_sdm_lognormal_xr_decomposition as its env_params argument. Backward-compatible with positional unpacking.

Note: k (gamma shape) is not included here; it lives in model_params={'k': ...} of the lognormal optimizer (default 2.0). Access me and mp via ln_env.me / ln_env.mp when calling sdm_lognormal_model_moments(rg, N, T, N0, t0, k, mu, sigma, me=..., mp=...).

N#

Alias for field number 0

N0#

Alias for field number 4

T#

Alias for field number 1

me#

Alias for field number 2

mp#

Alias for field number 3

mu#

Alias for field number 6

sigma#

Alias for field number 7

t0#

Alias for field number 5

estimate_sdm_column_params(decomposition, **kwargs)#

Estimate column parameters from the initial curve and component curves.

N, T, me, mp, N0, t0, poresize

Parameters:

  • decomposition (Decomposition) – The decomposition containing the initial curve and component curves.

  • poresize_bounds ((lo, hi) tuple, optional) – Bounds for the column pore size in Å. Default: (70, 300). Use a narrow window when the column type is known (e.g., (75, 80) for Superdex 200).

  • N0 (float, optional) – If given, the mobile-phase plate number is fixed to this value and excluded from the optimization. Use when N0 is known from buffer-only runs.

  • N0_bounds ((lo, hi) tuple, optional) – Bounds for N0 when it is a free parameter. Default: (500, 50000). Ignored if N0 is given.

  • include_M3 (bool, optional) – If True (default), include the third central moment (cube-root skewness) in the matching objective. This is critical for asymmetric peaks; for symmetric peaks it is a near no-op. Set False to reproduce the legacy M1+M2-only behaviour.

  • M1_weight (float, optional) – Weights for the moment-matching objective. Defaults follow legacy DispersiveMonopore.guess_params_using_moments: M1=6.0, M2=2.0, M3=2.0.

  • M2_weight (float, optional) – Weights for the moment-matching objective. Defaults follow legacy DispersiveMonopore.guess_params_using_moments: M1=6.0, M2=2.0, M3=2.0.

  • M3_weight (float, optional) – Weights for the moment-matching objective. Defaults follow legacy DispersiveMonopore.guess_params_using_moments: M1=6.0, M2=2.0, M3=2.0.

  • debug (bool, optional) – If True, print diagnostics and show a debug plot.

Returns:

(N, T, me, mp, N0, t0, poresize) – Estimated parameters for the SDM column.

Return type:

tuple

estimate_sdm_lognormal_column_params(decomposition, **kwargs)#

Estimate column parameters for SDM with lognormal pore distribution.

Runs the mono-pore estimator first, then converts poresize to lognormal parameters (mu, sigma).

Parameters:

  • decomposition (Decomposition) – The decomposition containing the initial curve and component curves.

  • kwargs (dict) – Additional parameters for the estimation process.

Returns:

(N, T, me, mp, N0, t0, mu, sigma) – Estimated parameters for the SDM column with lognormal pore distribution.

Return type:

tuple

estimate_sdm_lognormal_from_monopore(mono_ccurves, xr_icurve, **kwargs)#

Estimate lognormal column parameters from converged mono-pore SDM results.

Converts the mono-pore column parameters to lognormal initial guess by: 1. Extracting converged (N, T, x0, tI, N0, k) from the mono-pore result 2. Setting mu = ln(geometric_mean(poresize_stored, 2.5*Rg_max)), sigma=0.3 3. Shifting x0/tI to align the lognormal PDF peak with the data peak 4. (Optional) Refining (t0, k, mu, sigma) by analytical moment matching

against the EGH decomposition. Enabled when decomposition is given.

Parameters:

  • mono_ccurves (list of SdmComponentCurve) – Converged mono-pore component curves.

  • xr_icurve (Curve) – The XR integrated elution curve (data).

  • decomposition (Decomposition, optional) – EGH decomposition. When provided, the heuristic init is refined by moment matching (see refine_lognormal_params_by_moments). Recommended for the standard SDM(lognormal) pipeline.

  • debug (bool, optional)

  • moment_refine (bool, optional) – If False, skip the moment-matching refinement step even when decomposition is given. Default True.

Returns:

(N, T, me, mp, N0, t0_adj, mu, sigma) – Estimated parameters for the lognormal SDM optimizer.

Return type:

tuple

refine_lognormal_params_by_moments(decomposition, N, T, N0, t0, k, mu, sigma, me=1.5, mp=1.5, debug=False)#

Refine SDM lognormal column parameters by analytical moment matching.

Given an initial guess (typically from estimate_sdm_lognormal_from_monopore which uses heuristic poresize geometric mean + sigma=0.3), this function refines (mu, sigma, k, t0) by Nelder-Mead against the per-component (M1, Var) extracted from the EGH decomposition.

The upper bound on mu is enforced as ln(2 × Rg_max) — the SDM separation criterion. Above this, K_SEC values compress across components and elution curves converge, causing a degenerate Stage-3 decomposition. The L-BFGS-B solver that was used previously could overshoot this bound, accidentally yielding a bad starting point or causing a degenerate collapse; Nelder-Mead with an explicit penalty respects the bound by construction.

Uses the fast analytical moment evaluator molass.SEC.Models.LognormalPore.sdm_lognormal_model_moments() (~50 us/call), so the full refinement costs only a few tens of ms.

Only (mu, sigma, k, t0) are refined; (N, T, N0) are held fixed because the mono-pore stage already constrains them well from M1+M2+M3 matching.

Parameters:

  • decomposition (Decomposition) – EGH decomposition providing per-component empirical moments.

  • N (float) – Initial guesses (from estimate_sdm_lognormal_from_monopore).

  • T (float) – Initial guesses (from estimate_sdm_lognormal_from_monopore).

  • N0 (float) – Initial guesses (from estimate_sdm_lognormal_from_monopore).

  • t0 (float) – Initial guesses (from estimate_sdm_lognormal_from_monopore).

  • k (float) – Initial guesses (from estimate_sdm_lognormal_from_monopore).

  • mu (float) – Initial guesses (from estimate_sdm_lognormal_from_monopore).

  • sigma (float) – Initial guesses (from estimate_sdm_lognormal_from_monopore).

  • me (float, optional) – SEC partition exponents (default 1.5).

  • mp (float, optional) – SEC partition exponents (default 1.5).

  • debug (bool, optional) – If True, print per-component moment-matching diagnostics.

Returns:

(t0, k, mu, sigma) – Refined parameters. Other inputs (N, T, N0) are returned by the caller.

Return type:

tuple of float

See also

molass.SEC.Models.LognormalPore.sdm_lognormal_model_moments