RiemannianStaircaseParams configures the Burer–Monteiro rank schedule, SDP certificate verification, saddle escape, and the augmented-Lagrangian local solver used at every staircase level.
import gtsamParameter groups¶
pMinandpMaxbound the Burer–Monteiro factorization rank.alphascales the new column used to escape a certified saddle direction.etais the PSD certificate tolerance.verificationMethodselects sparse Spectra verification or dense Eigen verification.maxSpectraIters,numLanczosVectors, andspectraToltune the sparse eigensolver fallback.getAlmParams()andsetAlmParams()copy the inner augmented-Lagrangian parameters across the Python boundary.
alm = gtsam.AugmentedLagrangianParams()
alm.maxIterations = 80
alm.absoluteViolationTolerance = 1e-8
params = gtsam.RiemannianStaircaseParams()
params.pMin = 2
params.pMax = 5
params.alpha = 1e-2
params.eta = 1e-3
params.verificationMethod = gtsam.RiemannianStaircaseParams.VerificationMethod.Spectra
params.setAlmParams(alm)
copied_alm = params.getAlmParams()
print("rank range:", params.pMin, "to", params.pMax)
print("inner iterations:", copied_alm.maxIterations)
print("verification:", params.verificationMethod)rank range: 2 to 5
inner iterations: 80
verification: VerificationMethod.Spectra
Rot2 matrix problems require pMin >= 2; Rot3 requires pMin >= 3. DenseEigen is useful for small validation problems, while Spectra avoids a dense cubic eigendecomposition and is the normal scalable choice.