Add a verbose option to BFGS. (#95)

* adding a verbose option to BFGS * adding default values for verbose option to custom BFGS solvers --------- Co-authored-by: Johanna Haffner <[email protected]>
patrick-kidger · Dec 21, 2024 · eb8d50e · eb8d50e
1 parent 1d8e79b
commit eb8d50e
Show file tree

Hide file tree

Showing 2 changed files with 28 additions and 0 deletions.
diff --git a/optimistix/_solver/bfgs.py b/optimistix/_solver/bfgs.py
@@ -19,6 +19,7 @@
     max_norm,
     tree_dot,
     tree_full_like,
+    verbose_print,
 )
 from .._search import (
     AbstractDescent,
@@ -162,6 +163,7 @@ class AbstractBFGS(
     use_inverse: AbstractVar[bool]
     descent: AbstractVar[AbstractDescent[Y, _Hessian, Any]]
     search: AbstractVar[AbstractSearch[Y, _Hessian, FunctionInfo.Eval, Any]]
+    verbose: AbstractVar[frozenset[str]]
 
     def init(
         self,
@@ -252,6 +254,20 @@ def rejected(descent_state):
             accept, accepted, rejected, state.descent_state
         )
 
+        if len(self.verbose) > 0:
+            verbose_loss = "loss" in self.verbose
+            verbose_step_size = "step_size" in self.verbose
+            verbose_y = "y" in self.verbose
+            loss_eval = f_eval
+            loss = state.f_info.f
+            verbose_print(
+                (verbose_loss, "Loss on this step", loss_eval),
+                (verbose_loss, "Loss on the last accepted step", loss),
+                (verbose_step_size, "Step size", step_size),
+                (verbose_y, "y", state.y_eval),
+                (verbose_y, "y on the last accepted step", y),
+            )
+
         y_descent, descent_result = self.descent.step(step_size, descent_state)
         y_eval = (y**ω + y_descent**ω).ω
         result = RESULTS.where(
@@ -310,13 +326,15 @@ class BFGS(AbstractBFGS[Y, Aux, _Hessian], strict=True):
     use_inverse: bool
     descent: NewtonDescent
     search: BacktrackingArmijo
+    verbose: frozenset[str]
 
     def __init__(
         self,
         rtol: float,
         atol: float,
         norm: Callable[[PyTree], Scalar] = max_norm,
         use_inverse: bool = True,
+        verbose: frozenset[str] = frozenset(),
     ):
         self.rtol = rtol
         self.atol = atol
@@ -325,6 +343,7 @@ def __init__(
         self.descent = NewtonDescent(linear_solver=lx.Cholesky())
         # TODO(raderj): switch out `BacktrackingArmijo` with a better line search.
         self.search = BacktrackingArmijo()
+        self.verbose = verbose
 
 
 BFGS.__init__.__doc__ = """**Arguments:**
@@ -345,4 +364,8 @@ def __init__(
     default is (b), denoted via `use_inverse=True`. Note that this is incompatible with
     line search methods like [`optimistix.ClassicalTrustRegion`][], which use the
     Hessian approximation `B` as part of their own computations.
+- `verbose`: Whether to print out extra information about how the solve is
+    proceeding. Should be a frozenset of strings, specifying what information to print.
+    Valid entries are `step_size`, `loss`, `y`. For example 
+    `verbose=frozenset({"step_size", "loss"})`.
 """
diff --git a/tests/helpers.py b/tests/helpers.py
@@ -120,6 +120,7 @@ class BFGSDampedNewton(optx.AbstractBFGS):
     use_inverse: bool = False
     search: optx.AbstractSearch = optx.ClassicalTrustRegion()
     descent: optx.AbstractDescent = optx.DampedNewtonDescent()
+    verbose: frozenset[str] = frozenset()
 
 
 class BFGSIndirectDampedNewton(optx.AbstractBFGS):
@@ -131,6 +132,7 @@ class BFGSIndirectDampedNewton(optx.AbstractBFGS):
     use_inverse: bool = False
     search: optx.AbstractSearch = optx.ClassicalTrustRegion()
     descent: optx.AbstractDescent = optx.IndirectDampedNewtonDescent()
+    verbose: frozenset[str] = frozenset()
 
 
 class BFGSDogleg(optx.AbstractBFGS):
@@ -142,6 +144,7 @@ class BFGSDogleg(optx.AbstractBFGS):
     use_inverse: bool = False
     search: optx.AbstractSearch = optx.ClassicalTrustRegion()
     descent: optx.AbstractDescent = optx.DoglegDescent(linear_solver=lx.SVD())
+    verbose: frozenset[str] = frozenset()
 
 
 class BFGSBacktracking(optx.AbstractBFGS):
@@ -153,6 +156,7 @@ class BFGSBacktracking(optx.AbstractBFGS):
     use_inverse: bool = False
     search: optx.AbstractSearch = optx.BacktrackingArmijo()
     descent: optx.AbstractDescent = optx.NewtonDescent()
+    verbose: frozenset[str] = frozenset()
 
 
 class BFGSTrustRegion(optx.AbstractBFGS):
@@ -164,6 +168,7 @@ class BFGSTrustRegion(optx.AbstractBFGS):
     use_inverse: bool = False
     search: optx.AbstractSearch = optx.LinearTrustRegion()
     descent: optx.AbstractDescent = optx.NewtonDescent()
+    verbose: frozenset[str] = frozenset()
 
 
 atol = rtol = 1e-8