Merge pull request #920 from MilesCranmer/deprecate-parametric-expression

Deprecate ParametricExpressionSpec in favor of TemplateExpressionSpec

Author: MilesCranmer

docs/examples.md

@@ -560,37 +560,25 @@ model = PySRRegressor(
 model.fit(X, y)
 ```
 
-You can also use parameters in your template expressions, which will be optimized during the search:
-
-```python
-template = TemplateExpressionSpec(
-    expressions=["f", "g"],
-    variable_names=["x1", "x2", "x3"],
-    parameters={"p1": 2, "p2": 1},  # p1 has length 2, p2 has length 1
-    combine="p1[1] * sin(f(x1, x2)) + p1[2] * g(x3) + p2[1]",
-)
-```
-
-This will learn an equation of the form:
-
-$$ y = \alpha_1 \sin(f(x_1, x_2)) + \alpha_2 g(x_3) + \beta $$
-
-where $\alpha_1, \alpha_2$ are stored in `p1` and $\beta$ is stored in `p2`. The parameters will be optimized during the search.
-
 ### Parametric Expressions
 
 When your data has categories with shared equation structure but different parameters,
-you can use a `ParametricExpressionSpec`. Let's say we would like to learn the expression:
+you can use the `parameters` argument of `TemplateExpressionSpec` to specify learned category-specific parameters.
+
+For example, let's say we want to learn an equation of the form:
 
 $$ y = \alpha \sin(x_1) + \beta $$
 
-for three different values of $\alpha$ and $\beta$.
+where $\alpha$ and $\beta$ are different for each category.
+
+Further, let's say we have 3 categories,
+with $\alpha \in \{0.1, 1.5, -0.5\}$ and $\beta \in \{1.0, 2.0, 0.5\}$.
 
 ```python
 import numpy as np
-from pysr import PySRRegressor, ParametricExpressionSpec
+from pysr import PySRRegressor, TemplateExpressionSpec
 
-# Create data with 3 categories
+# Create data with 2 features and 3 categories
 X = np.random.uniform(-3, 3, (1000, 2))
 category = np.random.randint(0, 3, 1000)
 
@@ -603,34 +591,48 @@ y = np.array([
     scales[c] * np.sin(x1) + offsets[c]
     for x1, c in zip(X[:, 0], category)
 ])
+```
+
+Now, let's define our parametric expression:
 
+```python
+template = TemplateExpressionSpec(
+    expressions=["f"],
+    variable_names=["x1", "x2", "category"],
+    parameters={"p1": 3, "p2": 3},  # One parameter per category
+    combine="f(x1, x2, p1[category], p2[category])"
+)
+```
+
+Next, we pass the category as a _column_ in `X`
+corresponding to the index we defined in `variable_names`.
+
+**Note that because Julia is 1-indexed, we need to add 1 to the category index.**
+
+```python
+category_p_one = category + 1
+X_with_category = np.column_stack([X, category])
+```
+
+Now, we can fit our model:
+
+```python
 model = PySRRegressor(
-    expression_spec=ParametricExpressionSpec(max_parameters=2),
+    expression_spec=template,
     binary_operators=["+", "*", "-", "/"],
     unary_operators=["sin"],
     maxsize=10,
 )
-model.fit(X, y, category=category)
+model.fit(X_with_category, y)
 
-# Predicting on new data:
-# model.predict(X_test, category=category_test)
+# Predicting on new data
+# model.predict(X_test_with_category)
 ```
 
 See [Expression Specifications](/api/#expression-specifications) for more details.
 
-You can also use `TemplateExpressionSpec` in the same way, passing
-the category as a column of `X`:
-
-```python
-spec = TemplateExpressionSpec(
-    expressions=["f", "g"],
-    variable_names=["x1", "x2", "class"],
-    parameters={"p1": 3, "p2": 3},
-    combine="p1[class] * sin(f(x1, x2)) + p2[class]",
-)
-```
-
-this column will automatically be converted to integers.
+You can use this approach for more complex cases,
+where you have multiple expressions in the template and parameters that vary by category.
 
 
 ## 12. Using TensorBoard for Logging

environment.yml

@@ -9,4 +9,5 @@ dependencies:
   - scikit-learn>=1.0.0,<2.0.0
   - pyjuliacall>=0.9.22,<0.9.23
   - click>=7.0.0,<9.0.0
+  - beartype>=0.19,<0.22
   - typing-extensions>=4.0.0,<5.0.0

pyproject.toml

@@ -25,12 +25,12 @@ dependencies = [
     "juliacall>=0.9.24,<0.9.26",
     "click>=7.0.0,<9.0.0",
     "setuptools>=50.0.0",
+    "beartype>=0.19,<0.22",
     "typing-extensions>=4.0.0,<5.0.0",
 ]
 
 [project.optional-dependencies]
 dev = [
-    "beartype>=0.19,<0.21",
     "coverage>=7,<8",
     "coveralls>=4,<5",
     "ipykernel>=6,<7",

pysr/expression_specs.py

@@ -1,6 +1,8 @@
 from __future__ import annotations
 
 import copy
+import textwrap
+import warnings
 from abc import ABC, abstractmethod
 from textwrap import dedent
 from typing import TYPE_CHECKING, Any, NewType, overload
@@ -11,6 +13,7 @@
 from .export import add_export_formats
 from .julia_helpers import jl_array
 from .julia_import import AnyValue, SymbolicRegression, jl
+from .utils import ArrayLike
 
 try:
     from typing import TypeAlias
@@ -319,9 +322,41 @@ def create_exports(
         return _search_output_to_callable_expressions(equations, search_output, i)
 
 
+def parametric_expression_deprecation_warning(
+    max_parameters: int, variable_names: ArrayLike[str]
+):
+    function_name = "f"
+    var_names = list(variable_names)
+    message = dedent(
+        f"""
+        ParametricExpressionSpec is deprecated – you should switch to TemplateExpressionSpec
+        with explicit parameters indexed by category.
+
+        Since you have `max_parameters={max_parameters}` and
+        `variable_names=[{", ".join(f'"{v}"' for v in var_names)}]`, you could migrate like this:
+
+            n_categories = len(np.unique(category))  # count the number of parameters required
+            expression_spec = TemplateExpressionSpec(
+                expressions=["{function_name}"],
+                variable_names=[{", ".join(f'"{v}"' for v in var_names + ["category"])}],
+                parameters={{{", ".join(f'"p{i+1}": n_categories' for i in range(max_parameters))}}},
+                combine="{function_name}({', '.join(var_names + [f'p{i+1}[category]' for i in range(max_parameters)])})",
+            )
+            X = np.column_stack([X, category])       # add the category column
+
+        Finally, do not pass `category` when calling .fit().
+    """
+    ).strip()
+    wrapped = "\n".join(textwrap.fill(line, 88) for line in message.splitlines())
+    warnings.warn(wrapped, FutureWarning, stacklevel=3)
+
+
 class ParametricExpressionSpec(AbstractExpressionSpec):
     """Spec for parametric expressions that vary by category.
 
+    **This is deprecated in favor of the `TemplateExpressionSpec` class,
+    which now supports parameters indexed by category.**
+
     This class allows you to specify expressions with parameters that vary across different
     categories in your dataset. The expression structure remains the same, but parameters
     are optimized separately for each category.

pysr/sr.py

@@ -15,10 +15,11 @@
 from io import StringIO
 from multiprocessing import cpu_count
 from pathlib import Path
-from typing import Any, List, Literal, Tuple, Union, cast
+from typing import Any, Literal, Tuple, Union, cast
 
 import numpy as np
 import pandas as pd
+from beartype.typing import List
 from numpy import ndarray
 from numpy.typing import NDArray
 from sklearn.base import BaseEstimator, MultiOutputMixin, RegressorMixin
@@ -39,6 +40,7 @@
     AbstractExpressionSpec,
     ExpressionSpec,
     ParametricExpressionSpec,
+    parametric_expression_deprecation_warning,
 )
 from .feature_selection import run_feature_selection
 from .julia_extensions import load_required_packages
@@ -2251,6 +2253,11 @@ def fit(
         random_state = check_random_state(self.random_state)  # For np random
         seed = cast(int, random_state.randint(0, 2**31 - 1))  # For julia random
 
+        if isinstance(self.expression_spec, ParametricExpressionSpec):
+            parametric_expression_deprecation_warning(
+                self.expression_spec.max_parameters, variable_names
+            )
+
         # Pre transformations (feature selection and denoising)
         X, y, variable_names, complexity_of_variables, X_units, y_units = (
             self._pre_transform_training_data(

pysr/test/test_main.py

@@ -3,6 +3,7 @@
 import os
 import pickle as pkl
 import platform
+import re
 import tempfile
 import traceback
 import unittest
@@ -32,6 +33,7 @@
 )
 from pysr.export_latex import sympy2latex
 from pysr.export_sympy import pysr2sympy
+from pysr.expression_specs import parametric_expression_deprecation_warning
 from pysr.feature_selection import _handle_feature_selection, run_feature_selection
 from pysr.julia_helpers import init_julia
 from pysr.sr import (
@@ -1034,6 +1036,25 @@ def test_param_groupings(self):
         # Check the sets are equal:
         self.assertSetEqual(set(params), set(regressor_params))
 
+    def test_parametric_deprecation_warning(self):
+        """Test that the helpful warning message is displayed."""
+        pattern = re.compile(
+            r"ParametricExpressionSpec is deprecated.*TemplateExpressionSpec.*"
+            r"max_parameters=2.*"
+            r"variable_names=\[\"alpha\", \"beta\"\].*"
+            r"expressions=\[\"f\"\].*"
+            r"variable_names=\[\"alpha\", \"beta\", \"category\"\].*"
+            r"parameters=\{\s*\"p1\": n_categories,\s*\"p2\": n_categories\s*\}.*"
+            r"combine=\"f\(alpha, beta, p1\[category\], p2\[category\]\)\"",
+            flags=re.S,
+        )
+
+        with self.assertWarnsRegex(FutureWarning, pattern):
+            parametric_expression_deprecation_warning(
+                max_parameters=2,
+                variable_names=["alpha", "beta"],
+            )
+
     def test_load_all_packages(self):
         """Test we can load all packages at once."""
         load_all_packages()

pysr/utils.py

@@ -4,8 +4,9 @@
 import inspect
 import re
 from pathlib import Path
-from typing import Any, List, TypeVar, Union
+from typing import Any, TypeVar, Union
 
+from beartype.typing import List
 from numpy import ndarray
 from sklearn.utils.validation import _check_feature_names_in  # type: ignore