Add PIR (Physics Intermediate Representation) method

experiment/methods/PIRRegressor.py

@@ -0,0 +1,136 @@
+"""
+SRBench (dev branch) method definition: PIR -- classical engine.
+
+Format per dev CONTRIBUTING.md / experiment/methods/AFPRegressor.py. This file
+lives at experiment/methods/PIRRegressor.py and defines the four objects the
+harness looks for:
+    est           a sklearn-compatible Regressor instance
+    hyper_params  hyperparameter search space (empty -> no tuning)
+    complexity    complexity(est) -> int model size
+    model         model(est, X) -> sympy-compatible string
+
+The classical engine is installed via experiment/methods/src/PIR_install.sh,
+which pip-installs the PUBLIC, torch-free engine:
+    git+https://github.com/Qazi-pk/physics-engine@v0.1.0
+
+Config below is byte-for-byte the BLIND sweep that produced the archived
+results_tierA_blind/ (7/44 stable, SEED=0): enforce_dimensions=False,
+allowed_powers=[1,2], pairwise on, OT off, no physics features. All other
+params left at adapter defaults (the blind sweep did not set them).
+"""
+
+import signal
+import inspect
+import numpy as np
+import sympy as sp
+from sklearn.base import BaseEstimator, RegressorMixin
+
+from physics_engine.sklearn_adapter import PIRRegressor
+
+
+# Exact blind-sweep config (matches sweep_tierA_blind.py, SEED = 0).
+_PIR_VANILLA_CONFIG = dict(
+    enforce_dimensions=False,
+    allowed_powers=[1, 2],
+    include_pairwise_products=True,
+    use_ransac=True,
+    use_residual=True,
+    use_sparse=True,
+    use_ot_loss=False,
+    add_physics_features=False,
+)
+
+
+class _Timeout(Exception):
+    pass
+
+
+def _on_alarm(signum, frame):
+    raise _Timeout()
+
+
+class PIRClassicRegressor(BaseEstimator, RegressorMixin):
+    """sklearn wrapper around classical PIRRegressor with a SIGALRM time budget
+    and a guaranteed-valid fallback model."""
+
+    def __init__(self, max_time=3600, random_state=0, **pir_kwargs):
+        self.max_time = max_time
+        self.random_state = random_state
+        self.pir_kwargs = {**_PIR_VANILLA_CONFIG, **pir_kwargs}
+
+    def _build(self):
+        kw = dict(self.pir_kwargs)
+        try:
+            return PIRRegressor(random_state=self.random_state, **kw)
+        except TypeError:
+            return PIRRegressor(**kw)
+
+    def fit(self, X, y):
+        y_arr = np.asarray(y, dtype=float).ravel()
+        self.expr_ = repr(float(np.mean(y_arr))) if y_arr.size else "0.0"
+        self.n_features_in_ = np.asarray(X).shape[1] if np.ndim(X) > 1 else 1
+        self._inner = self._build()
+
+        use_alarm = hasattr(signal, "SIGALRM") and self.max_time and self.max_time > 0
+        old = None
+        if use_alarm:
+            old = signal.signal(signal.SIGALRM, _on_alarm)
+            signal.alarm(int(self.max_time))
+        try:
+            self._inner.fit(X, y)
+            if hasattr(self._inner, "model"):
+                self.expr_ = self._inner.model()
+            elif hasattr(self._inner, "expr_"):
+                self.expr_ = str(self._inner.expr_)
+        except _Timeout:
+            pass
+        finally:
+            if use_alarm:
+                signal.alarm(0)
+                if old is not None:
+                    signal.signal(signal.SIGALRM, old)
+        self.is_fitted_ = True
+        return self
+
+    def predict(self, X):
+        if hasattr(self, "_inner") and hasattr(self._inner, "predict"):
+            try:
+                return self._inner.predict(X)
+            except Exception:
+                pass
+        n = X.shape[0] if hasattr(X, "shape") else len(X)
+        try:
+            val = float(self.expr_)
+        except (TypeError, ValueError):
+            val = 0.0
+        return np.full(n, val, dtype=float)
+
+
+est = PIRClassicRegressor(max_time=3600, random_state=0)
+
+# No hyperparameter tuning: the blind sweep used a single fixed config.
+# Empty list -> harness skips the grid search (matches skip_tuning on sym track).
+hyper_params = {}
+
+
+def complexity(est):
+    """Model size = number of nodes in the sympy expression."""
+    try:
+        expr = sp.sympify(str(getattr(est, "expr_", "0")))
+        return int(sp.count_ops(expr)) + len(expr.free_symbols)
+    except Exception:
+        return 0
+
+
+def model(est, X=None):
+    """Return a sympy-parseable model string with symbols matching the dataset
+    feature columns.
+
+    SRBench passes X_train_scaled (a numpy array, no column names) when the
+    signature accepts X. The engine emits symbols using whatever names it saw
+    at fit time. If the engine already produced dataset-style names, return as
+    is; otherwise leave generic positional names for SRBench's symbolic_utils
+    mapping. (Confirm naming against one real dataset in local CI -- see notes.)
+    """
+    expr = str(getattr(est, "expr_", "0.0"))
+    return expr

experiment/methods/src/PIR_install.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+# Installs the classical PIR engine for SRBench (dev branch).
+# Auto-detected by the GitHub Actions workflow via the *_install.sh name.
+# No sudo; installs into the active conda env ($CONDA_PREFIX).
+#
+# Pulls the PUBLIC, torch-free engine pinned at an immutable tag.
+set -euo pipefail
+
+PIR_REPO_URL="${PIR_REPO_URL:-https://github.com/Qazi-pk/physics-engine}"
+PIR_REF="${PIR_REF:-v0.1.0}"
+
+echo "[PIR install] python: $(which python)"
+echo "[PIR install] installing ${PIR_REPO_URL}@${PIR_REF}"
+
+pip install "git+${PIR_REPO_URL}@${PIR_REF}"
+
+# Sanity: import must resolve and expose PIRRegressor.
+python -c "from physics_engine.sklearn_adapter import PIRRegressor; print('[PIR install] import OK')"