sleipnirgroup-jormungandr 0.6.7.dev81__tar.gz → 0.6.7.dev83__tar.gz

{sleipnirgroup_jormungandr-0.6.7.dev81 → sleipnirgroup_jormungandr-0.6.7.dev83}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sleipnirgroup-jormungandr
-Version: 0.6.7.dev81
+Version: 0.6.7.dev83
 Summary: Reverse mode autodiff library and NLP solver DSL
 License-Expression: BSD-3-Clause
 License-File: LICENSE.txt

{sleipnirgroup_jormungandr-0.6.7.dev81 → sleipnirgroup_jormungandr-0.6.7.dev83}/include/sleipnir/optimization/solver/ipm.hpp

@@ -634,7 +634,9 @@ ExitStatus ipm(const IPMMatrixCallbacks<Scalar>& matrix_callbacks,
             matrices.A_i(trial_x), trial_c_i, trial_s, trial_y, trial_z, μ);
 
         // If the step using αᵐᵃˣ reduced the KKT error, accept it anyway
-        if (next_kkt_error <= Scalar(0.999) * current_kkt_error) {
+        if (next_kkt_error <= Scalar(0.999) * current_kkt_error &&
+            constraint_violation(trial_c_e, trial_c_i, trial_s) <=
+                filter.max_constraint_violation) {
           // Accept step
           α = α_max;
           break;

{sleipnirgroup_jormungandr-0.6.7.dev81 → sleipnirgroup_jormungandr-0.6.7.dev83}/include/sleipnir/optimization/solver/sqp.hpp

@@ -444,7 +444,9 @@ ExitStatus sqp(const SQPMatrixCallbacks<Scalar>& matrix_callbacks,
             matrices.g(trial_x), matrices.A_e(trial_x), trial_c_e, trial_y);
 
         // If the step using αᵐᵃˣ reduced the KKT error, accept it anyway
-        if (next_kkt_error <= Scalar(0.999) * current_kkt_error) {
+        if (next_kkt_error <= Scalar(0.999) * current_kkt_error &&
+            constraint_violation(trial_c_e) <=
+                filter.max_constraint_violation) {
           // Accept step
           α = α_max;
           break;

{sleipnirgroup_jormungandr-0.6.7.dev81 → sleipnirgroup_jormungandr-0.6.7.dev83}/include/sleipnir/optimization/solver/util/dense_kkt_solver.hpp

@@ -58,18 +58,23 @@ class DenseKKTSolver {
   /// @param lhs Left-hand side of the system.
   /// @return The factorization.
   DenseKKTSolver& compute(const DenseMatrix& lhs) {
-    m_info = m_solver.compute(lhs).info();
+    // If the previous δ was small and nonzero, skip the δ = 0 attempt because
+    // it's almost always rejected for small pivots. If the previous δ was
+    // large, attempt δ = 0 so δ can drop to zero quickly.
+    if (m_prev_δ == Scalar(0) || m_prev_δ > Scalar(1e-2)) {
+      m_info = m_solver.compute(lhs).info();
 
-    if (m_info == Eigen::Success) {
-      auto D = m_solver.vectorD();
-
-      // If the inertia is ideal and D from LDLT is sufficiently far from zero,
-      // don't regularize the system
-      if (Inertia{D} == ideal_inertia &&
-          (D.cwiseAbs().array() >= Scalar(1e-4)).all()) {
-        m_prev_δ = Scalar(0);
-        m_prev_γ = Scalar(0);
-        return *this;
+      if (m_info == Eigen::Success) {
+        auto D = m_solver.vectorD();
+
+        // If the inertia is ideal and D from LDLT is sufficiently far from
+        // zero, don't regularize the system
+        if (Inertia{D} == ideal_inertia &&
+            (D.cwiseAbs().array() >= Scalar(1e-4)).all()) {
+          m_prev_δ = Scalar(0);
+          m_prev_γ = Scalar(0);
+          return *this;
+        }
       }
     }
 

{sleipnirgroup_jormungandr-0.6.7.dev81 → sleipnirgroup_jormungandr-0.6.7.dev83}/include/sleipnir/optimization/solver/util/sparse_kkt_solver.hpp

@@ -71,19 +71,24 @@ class SparseKKTSolver {
       m_analyzed_pattern = true;
     }
 
-    m_solver.factorize(unregularized_lhs);
-    m_info = m_solver.info();
-
-    if (m_info == Eigen::Success) {
-      auto D = m_solver.vectorD();
-
-      // If the inertia is ideal and D from LDLT is sufficiently far from zero,
-      // don't regularize the system
-      if (Inertia{D} == ideal_inertia &&
-          (D.cwiseAbs().array() >= Scalar(1e-4)).all()) {
-        m_prev_δ = Scalar(0);
-        m_prev_γ = Scalar(0);
-        return *this;
+    // If the previous δ was small and nonzero, skip the δ = 0 attempt because
+    // it's almost always rejected for small pivots. If the previous δ was
+    // large, attempt δ = 0 so δ can drop to zero quickly.
+    if (m_prev_δ == Scalar(0) || m_prev_δ > Scalar(1e-2)) {
+      m_solver.factorize(unregularized_lhs);
+      m_info = m_solver.info();
+
+      if (m_info == Eigen::Success) {
+        auto D = m_solver.vectorD();
+
+        // If the inertia is ideal and D from LDLT is sufficiently far from
+        // zero, don't regularize the system
+        if (Inertia{D} == ideal_inertia &&
+            (D.cwiseAbs().array() >= Scalar(1e-4)).all()) {
+          m_prev_δ = Scalar(0);
+          m_prev_γ = Scalar(0);
+          return *this;
+        }
       }
     }
 

{sleipnirgroup_jormungandr-0.6.7.dev81 → sleipnirgroup_jormungandr-0.6.7.dev83}/python/src/sleipnir/__init__.py

@@ -2,4 +2,4 @@
 
 from ._sleipnir import *
 
-__version__ = "0.6.7.dev81"
+__version__ = "0.6.7.dev83"