Spaces:
Sleeping
Sleeping
MilesCranmer
commited on
Commit
•
a06bfc4
1
Parent(s):
5b978f9
Create torch export function
Browse files- pysr/__init__.py +1 -0
- pysr/export_torch.py +172 -0
- pysr/sr.py +1 -0
pysr/__init__.py
CHANGED
@@ -1,3 +1,4 @@
|
|
1 |
from .sr import pysr, get_hof, best, best_tex, best_callable, best_row
|
2 |
from .feynman_problems import Problem, FeynmanProblem
|
3 |
from .export_jax import sympy2jax
|
|
|
|
1 |
from .sr import pysr, get_hof, best, best_tex, best_callable, best_row
|
2 |
from .feynman_problems import Problem, FeynmanProblem
|
3 |
from .export_jax import sympy2jax
|
4 |
+
from .export_torch import sympy2torch
|
pysr/export_torch.py
ADDED
@@ -0,0 +1,172 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#####
|
2 |
+
# From https://github.com/patrick-kidger/sympytorch
|
3 |
+
# Copied here to allow PySR-specific tweaks
|
4 |
+
#####
|
5 |
+
|
6 |
+
import collections as co
|
7 |
+
import functools as ft
|
8 |
+
import sympy
|
9 |
+
import torch
|
10 |
+
|
11 |
+
|
12 |
+
def _reduce(fn):
|
13 |
+
def fn_(*args):
|
14 |
+
return ft.reduce(fn, args)
|
15 |
+
return fn_
|
16 |
+
|
17 |
+
|
18 |
+
_global_func_lookup = {
|
19 |
+
sympy.Mul: _reduce(torch.mul),
|
20 |
+
sympy.Add: _reduce(torch.add),
|
21 |
+
sympy.div: torch.div,
|
22 |
+
sympy.Abs: torch.abs,
|
23 |
+
sympy.sign: torch.sign,
|
24 |
+
# Note: May raise error for ints.
|
25 |
+
sympy.ceiling: torch.ceil,
|
26 |
+
sympy.floor: torch.floor,
|
27 |
+
sympy.log: torch.log,
|
28 |
+
sympy.exp: torch.exp,
|
29 |
+
sympy.sqrt: torch.sqrt,
|
30 |
+
sympy.cos: torch.cos,
|
31 |
+
sympy.acos: torch.acos,
|
32 |
+
sympy.sin: torch.sin,
|
33 |
+
sympy.asin: torch.asin,
|
34 |
+
sympy.tan: torch.tan,
|
35 |
+
sympy.atan: torch.atan,
|
36 |
+
sympy.atan2: torch.atan2,
|
37 |
+
# Note: May give NaN for complex results.
|
38 |
+
sympy.cosh: torch.cosh,
|
39 |
+
sympy.acosh: torch.acosh,
|
40 |
+
sympy.sinh: torch.sinh,
|
41 |
+
sympy.asinh: torch.asinh,
|
42 |
+
sympy.tanh: torch.tanh,
|
43 |
+
sympy.atanh: torch.atanh,
|
44 |
+
sympy.Pow: torch.pow,
|
45 |
+
sympy.re: torch.real,
|
46 |
+
sympy.im: torch.imag,
|
47 |
+
sympy.arg: torch.angle,
|
48 |
+
# Note: May raise error for ints and complexes
|
49 |
+
sympy.erf: torch.erf,
|
50 |
+
sympy.loggamma: torch.lgamma,
|
51 |
+
sympy.Eq: torch.eq,
|
52 |
+
sympy.Ne: torch.ne,
|
53 |
+
sympy.StrictGreaterThan: torch.gt,
|
54 |
+
sympy.StrictLessThan: torch.lt,
|
55 |
+
sympy.LessThan: torch.le,
|
56 |
+
sympy.GreaterThan: torch.ge,
|
57 |
+
sympy.And: torch.logical_and,
|
58 |
+
sympy.Or: torch.logical_or,
|
59 |
+
sympy.Not: torch.logical_not,
|
60 |
+
sympy.Max: torch.max,
|
61 |
+
sympy.Min: torch.min,
|
62 |
+
# Matrices
|
63 |
+
sympy.MatAdd: torch.add,
|
64 |
+
sympy.HadamardProduct: torch.mul,
|
65 |
+
sympy.Trace: torch.trace,
|
66 |
+
# Note: May raise error for integer matrices.
|
67 |
+
sympy.Determinant: torch.det,
|
68 |
+
}
|
69 |
+
|
70 |
+
|
71 |
+
class _Node(torch.nn.Module):
|
72 |
+
def __init__(self, *, expr, _memodict, _func_lookup, **kwargs):
|
73 |
+
super().__init__(**kwargs)
|
74 |
+
|
75 |
+
self._sympy_func = expr.func
|
76 |
+
|
77 |
+
if issubclass(expr.func, sympy.Float):
|
78 |
+
self._value = torch.nn.Parameter(torch.tensor(float(expr)))
|
79 |
+
self._torch_func = lambda: self._value
|
80 |
+
self._args = ()
|
81 |
+
elif issubclass(expr.func, sympy.UnevaluatedExpr):
|
82 |
+
if len(expr.args) != 1 or not issubclass(expr.args[0].func, sympy.Float):
|
83 |
+
raise ValueError("UnevaluatedExpr should only be used to wrap floats.")
|
84 |
+
self.register_buffer('_value', torch.tensor(float(expr.args[0])))
|
85 |
+
self._torch_func = lambda: self._value
|
86 |
+
self._args = ()
|
87 |
+
elif issubclass(expr.func, sympy.Integer):
|
88 |
+
# Can get here if expr is one of the Integer special cases,
|
89 |
+
# e.g. NegativeOne
|
90 |
+
self._value = int(expr)
|
91 |
+
self._torch_func = lambda: self._value
|
92 |
+
self._args = ()
|
93 |
+
elif issubclass(expr.func, sympy.Symbol):
|
94 |
+
self._name = expr.name
|
95 |
+
self._torch_func = lambda value: value
|
96 |
+
self._args = ((lambda memodict: memodict[expr.name]),)
|
97 |
+
else:
|
98 |
+
self._torch_func = _func_lookup[expr.func]
|
99 |
+
args = []
|
100 |
+
for arg in expr.args:
|
101 |
+
try:
|
102 |
+
arg_ = _memodict[arg]
|
103 |
+
except KeyError:
|
104 |
+
arg_ = type(self)(expr=arg, _memodict=_memodict, _func_lookup=_func_lookup, **kwargs)
|
105 |
+
_memodict[arg] = arg_
|
106 |
+
args.append(arg_)
|
107 |
+
self._args = torch.nn.ModuleList(args)
|
108 |
+
|
109 |
+
def sympy(self, _memodict):
|
110 |
+
if issubclass(self._sympy_func, sympy.Float):
|
111 |
+
return self._sympy_func(self._value.item())
|
112 |
+
elif issubclass(self._sympy_func, sympy.UnevaluatedExpr):
|
113 |
+
return self._sympy_func(self._value.item())
|
114 |
+
elif issubclass(self._sympy_func, sympy.Integer):
|
115 |
+
return self._sympy_func(self._value)
|
116 |
+
elif issubclass(self._sympy_func, sympy.Symbol):
|
117 |
+
return self._sympy_func(self._name)
|
118 |
+
else:
|
119 |
+
if issubclass(self._sympy_func, (sympy.Min, sympy.Max)):
|
120 |
+
evaluate = False
|
121 |
+
else:
|
122 |
+
evaluate = True
|
123 |
+
args = []
|
124 |
+
for arg in self._args:
|
125 |
+
try:
|
126 |
+
arg_ = _memodict[arg]
|
127 |
+
except KeyError:
|
128 |
+
arg_ = arg.sympy(_memodict)
|
129 |
+
_memodict[arg] = arg_
|
130 |
+
args.append(arg_)
|
131 |
+
return self._sympy_func(*args, evaluate=evaluate)
|
132 |
+
|
133 |
+
def forward(self, memodict):
|
134 |
+
args = []
|
135 |
+
for arg in self._args:
|
136 |
+
try:
|
137 |
+
arg_ = memodict[arg]
|
138 |
+
except KeyError:
|
139 |
+
arg_ = arg(memodict)
|
140 |
+
memodict[arg] = arg_
|
141 |
+
args.append(arg_)
|
142 |
+
return self._torch_func(*args)
|
143 |
+
|
144 |
+
|
145 |
+
class SingleSymPyModule(torch.nn.Module):
|
146 |
+
def __init__(self, expression, symbols_in,
|
147 |
+
extra_funcs=None, **kwargs):
|
148 |
+
super().__init__(**kwargs)
|
149 |
+
|
150 |
+
if extra_funcs is None:
|
151 |
+
extra_funcs = {}
|
152 |
+
_func_lookup = co.ChainMap(_global_func_lookup, extra_funcs)
|
153 |
+
|
154 |
+
_memodict = {}
|
155 |
+
self._node = _Node(expr=expression, _memodict=_memodict, _func_lookup=_func_lookup)
|
156 |
+
self._expression_string = str(expression)
|
157 |
+
self.symbols_in = [str(symbol) for symbol in symbols_in]
|
158 |
+
|
159 |
+
def __repr__(self):
|
160 |
+
return f"{type(self).__name__}(expression={self._expression_string})"
|
161 |
+
|
162 |
+
def sympy(self):
|
163 |
+
_memodict = {}
|
164 |
+
return self._node.sympy(_memodict)
|
165 |
+
|
166 |
+
def forward(self, X):
|
167 |
+
symbols = {symbol: X[:, i]
|
168 |
+
for i, symbol in enumerate(self.symbols_in)}
|
169 |
+
return self._node(symbols)
|
170 |
+
|
171 |
+
def sympy2torch(expression, symbols_in):
|
172 |
+
return SingleSymPyModule(expression, symbols_in)
|
pysr/sr.py
CHANGED
@@ -14,6 +14,7 @@ from pathlib import Path
|
|
14 |
from datetime import datetime
|
15 |
import warnings
|
16 |
from .export_jax import sympy2jax
|
|
|
17 |
|
18 |
global_equation_file = 'hall_of_fame.csv'
|
19 |
global_n_features = None
|
|
|
14 |
from datetime import datetime
|
15 |
import warnings
|
16 |
from .export_jax import sympy2jax
|
17 |
+
from .export_torch import sympy2torch
|
18 |
|
19 |
global_equation_file = 'hall_of_fame.csv'
|
20 |
global_n_features = None
|