feat(trapping): Create trapping constraints automatically

Previously, TrappingSR3 could only use constraints passed to it, and only then a
limited set of constraints.  It also didn't apply the trapping constraints
automatically, because constraints were required at __init__, and actually
shaping them requires knowledge about the number of features, typically not
known until fit (unless the user is a developer who knows how the feature
libraries work internally 😉)

WIP, Spawned issue dynamicslab#452

pysindy/optimizers/constrained_sr3.py

@@ -72,8 +72,8 @@ class ConstrainedSR3(SR3):
 
     constraint_lhs : numpy ndarray, optional (default None)
         Shape should be (n_constraints, n_features * n_targets),
-        The left hand side matrix C of Cw <= d.
-        There should be one row per constraint.
+        The left hand side matrix C of Cw <= d (Or Cw = d for equality
+        constraints). There should be one row per constraint.
 
     constraint_rhs : numpy ndarray, shape (n_constraints,), optional (default None)
         The right hand side vector d of Cw <= d.

pysindy/optimizers/trapping_sr3.py

@@ -71,10 +71,6 @@ class TrappingSR3(ConstrainedSR3):
         If relax_optim = True, use the relax-and-split method. If False,
         try a direct minimization on the largest eigenvalue.
 
-    inequality_constraints : bool, optional (default False)
-        If True, relax_optim must be false or relax_optim = True
-        AND threshold != 0, so that the CVXPY methods are used.
-
     alpha_A :
         Determines the step size in the prox-gradient descent over A.
         For convergence, need alpha_A <= eta, so default
@@ -166,10 +162,11 @@ class TrappingSR3(ConstrainedSR3):
     def __init__(
         self,
         *,
+        _n_tgts: int = None,
+        _include_bias: bool = True,
         eta: Union[float, None] = None,
         eps_solver: float = 1e-7,
         relax_optim: bool = True,
-        inequality_constraints=False,
         alpha_A: Union[float, None] = None,
         alpha_m: Union[float, None] = None,
         gamma: float = -0.1,
@@ -180,10 +177,46 @@ def __init__(
         A0: Union[NDArray, None] = None,
         **kwargs,
     ):
-        super().__init__(thresholder=thresholder, **kwargs)
+        # n_tgts, constraints, etc are data-dependent parameters and belong in
+        # _reduce/fit ().  The following is a hack until we refactor how
+        # constraints are applied in ConstrainedSR3 and MIOSR
+        self._include_bias = _include_bias
+        self._n_tgts = _n_tgts
+        if _n_tgts is None:
+            warnings.warn(
+                "Trapping Optimizer initialized without _n_tgts.  It will likely"
+                " be unable to fit data"
+            )
+            _n_tgts = 1
+        constraint_separation_index = kwargs.get("constraint_separation_index", 0)
+        constraint_rhs, constraint_lhs = _make_constraints(
+            _n_tgts, include_bias=_include_bias
+        )
+        constraint_order = kwargs.get("constraint_order", "feature")
+        if constraint_order == "target":
+            constraint_lhs = np.reshape(np.transpose(constraint_lhs, [1, 0, 2]))
+        constraint_lhs = np.reshape(constraint_lhs, (constraint_lhs.shape[0], -1))
+        try:
+            constraint_lhs = np.concatenate(
+                (kwargs.pop("constraint_lhs"), constraint_lhs), 0
+            )
+            constraint_rhs = np.concatenate(
+                (kwargs.pop("constraint_rhs"), constraint_rhs), 0
+            )
+        except KeyError:
+            pass
+
+        super().__init__(
+            constraint_lhs=constraint_lhs,
+            constraint_rhs=constraint_rhs,
+            constraint_separation_index=constraint_separation_index,
+            constraint_order=constraint_order,
+            equality_constraints=True,
+            thresholder=thresholder,
+            **kwargs,
+        )
         self.eps_solver = eps_solver
         self.relax_optim = relax_optim
-        self.inequality_constraints = inequality_constraints
         self.m0 = m0
         self.A0 = A0
         self.alpha_A = alpha_A

test/conftest.py

@@ -410,7 +410,8 @@ def data_linear_oscillator_corrupted():
 @pytest.fixture(scope="session")
 def data_linear_combination():
     t = np.linspace(0, 5, 100)
-    x = np.stack((np.exp(t), np.sin(t), np.cos(t)), axis=-1)
+    lib = PolynomialLibrary(2)
+    x = lib.fit_transform(t)
     y = np.stack((x[:, 0] + x[:, 1], x[:, 1] + x[:, 2]), axis=-1)
 
     return x, y

test/test_optimizers.py

@@ -4,6 +4,7 @@
 import numpy as np
 import pytest
 from numpy.linalg import norm
+from numpy.typing import NDArray
 from scipy.integrate import solve_ivp
 from sklearn.base import BaseEstimator
 from sklearn.exceptions import ConvergenceWarning
@@ -18,6 +19,7 @@
 from pysindy import SINDy
 from pysindy.feature_library import CustomLibrary
 from pysindy.feature_library import SINDyPILibrary
+from pysindy.optimizers import BaseOptimizer
 from pysindy.optimizers import ConstrainedSR3
 from pysindy.optimizers import EnsembleOptimizer
 from pysindy.optimizers import FROLS
@@ -67,6 +69,18 @@ def predict(self, x):
         return x
 
 
+def _align_optimizer_and_1dfeatures(
+    opt: BaseOptimizer, features: NDArray
+) -> tuple[BaseOptimizer, NDArray]:
+    # This is a hack until constraints are moved from init to fit
+    if isinstance(opt, TrappingSR3):
+        opt = TrappingSR3(_n_tgts=1, _include_bias=True)
+        features = np.hstack([features, features, features])
+    else:
+        features = features
+    return opt, features
+
+
 @pytest.mark.parametrize(
     "cls, support",
     [
@@ -100,17 +114,19 @@ def data(request):
         SR3(),
         ConstrainedSR3(),
         StableLinearSR3(),
-        TrappingSR3(),
+        TrappingSR3(_n_tgts=1),
         Lasso(fit_intercept=False),
         ElasticNet(fit_intercept=False),
         DummyLinearModel(),
         MIOSR(),
     ],
+    ids=lambda param: type(param),
 )
 def test_fit(data_derivative_1d, optimizer):
     x, x_dot = data_derivative_1d
     if len(x.shape) == 1:
         x = x.reshape(-1, 1)
+    optimizer, x = _align_optimizer_and_1dfeatures(optimizer, x)
     opt = WrappedOptimizer(optimizer, unbias=False)
     opt.fit(x, x_dot)
 
@@ -167,12 +183,12 @@ def test_alternate_parameters(data_derivative_1d, kwargs):
     ],
 )
 def test_sample_weight_optimizers(data_1d, optimizer):
-    x, t = data_1d
-
+    y, t = data_1d
+    opt = optimizer()
+    opt, x = _align_optimizer_and_1dfeatures(opt, y)
     sample_weight = np.ones(x[:, 0].shape)
     sample_weight[::2] = 0
-    opt = optimizer()
-    opt.fit(x, x, sample_weight=sample_weight)
+    opt.fit(x, y, sample_weight=sample_weight)
     check_is_fitted(opt)
 
 
@@ -222,12 +238,12 @@ def test_sr3_bad_parameters(optimizer, params):
 )
 def test_trapping_bad_parameters(params):
     with pytest.raises(ValueError):
-        TrappingSR3(**params)
+        TrappingSR3(_n_tgts=1, **params)
 
 
 def test_trapping_objective_print():
     # test error in verbose print logic when max_iter < 10
-    opt = TrappingSR3(max_iter=2, verbose=True)
+    opt = TrappingSR3(_n_tgts=1, max_iter=2, verbose=True)
     arr = np.ones(1)
     opt._objective(arr, arr, arr, arr, arr, 1)
 
@@ -481,7 +497,7 @@ def test_trapping_sr3_quadratic_library(params, trapping_sr3_params):
 
     params.update(trapping_sr3_params)
 
-    opt = TrappingSR3(**params)
+    opt = TrappingSR3(_n_tgts=1, _include_bias=False, **params)
     opt.fit(features, x_dot)
     assert opt.PL_.shape == (1, 1, 1, 2)
     assert opt.PQ_.shape == (1, 1, 1, 1, 2)
@@ -494,7 +510,7 @@ def test_trapping_sr3_quadratic_library(params, trapping_sr3_params):
     params["constraint_rhs"] = np.zeros(p)
     params["constraint_lhs"] = np.eye(p, r * N)
 
-    opt = TrappingSR3(**params)
+    opt = TrappingSR3(_n_tgts=1, _include_bias=False, **params)
     opt.fit(features, x_dot)
     assert opt.PL_.shape == (1, 1, 1, 2)
     assert opt.PQ_.shape == (1, 1, 1, 1, 2)
@@ -631,7 +647,7 @@ def test_constrained_sr3_prox_functions(data_derivative_1d, thresholder):
         (SR3, {"trimming_fraction": 0.1}),
         (ConstrainedSR3, {"constraint_lhs": [1], "constraint_rhs": [1]}),
         (ConstrainedSR3, {"trimming_fraction": 0.1}),
-        (TrappingSR3, {"constraint_lhs": [1], "constraint_rhs": [1]}),
+        (TrappingSR3, {"_n_tgts": 1, "constraint_lhs": [1], "constraint_rhs": [1]}),
         (StableLinearSR3, {"constraint_lhs": [1], "constraint_rhs": [1]}),
         (StableLinearSR3, {"trimming_fraction": 0.1}),
         (SINDyPI, {}),
@@ -741,7 +757,7 @@ def test_sr3_enable_trimming(optimizer, data_linear_oscillator_corrupted):
         SR3(max_iter=1),
         ConstrainedSR3(max_iter=1),
         StableLinearSR3(max_iter=1),
-        TrappingSR3(max_iter=1),
+        TrappingSR3(_n_tgts=1, max_iter=1),
     ],
 )
 def test_fit_warn(data_derivative_1d, optimizer):
@@ -755,7 +771,11 @@ def test_fit_warn(data_derivative_1d, optimizer):
 
 @pytest.mark.parametrize(
     "optimizer",
-    [(ConstrainedSR3, {"max_iter": 80}), (TrappingSR3, {"max_iter": 100}), (MIOSR, {})],
+    [
+        (ConstrainedSR3, {"max_iter": 80}),
+        (TrappingSR3, {"_n_tgts": 5, "max_iter": 100}),
+        (MIOSR, {}),
+    ],
 )
 @pytest.mark.parametrize("target_value", [0, -1, 3])
 def test_row_format_constraints(data_linear_combination, optimizer, target_value):
@@ -966,6 +986,7 @@ def test_normalize_columns(data_derivative_1d, optimizer):
     if len(x.shape) == 1:
         x = x.reshape(-1, 1)
     opt = optimizer(normalize_columns=True)
+    opt, x = _align_optimizer_and_1dfeatures(opt, x)
     opt.fit(x, x_dot)
     check_is_fitted(opt)
     assert opt.complexity >= 0
@@ -1027,9 +1048,11 @@ def test_ssr_criteria(data_lorenz):
     ],
 )
 def test_optimizers_verbose(data_1d, optimizer):
-    x, _ = data_1d
+    y, _ = data_1d
     opt = optimizer(verbose=True)
-    opt.fit(x, x)
+    opt, x = _align_optimizer_and_1dfeatures(opt, y)
+    opt.verbose = True
+    opt.fit(x, y)
     check_is_fitted(opt)
 
 
@@ -1043,10 +1066,10 @@ def test_optimizers_verbose(data_1d, optimizer):
     ],
 )
 def test_optimizers_verbose_cvxpy(data_1d, optimizer):
-    x, _ = data_1d
-
+    y, _ = data_1d
     opt = optimizer(verbose_cvxpy=True)
-    opt.fit(x, x)
+    opt, x = _align_optimizer_and_1dfeatures(opt, y)
+    opt.fit(x, y)
     check_is_fitted(opt)