def test_filter_symbols():
s = numbered_symbols()
filtered = filter_symbols(s, symbols("x0 x2 x3"))
assert take(filtered, 3) == list(symbols("x1 x4 x5"))
def cse(exprs,
symbols=None,
optimizations=None,
postprocess=None,
order='canonical',
ignore=()):
""" Perform common subexpression elimination on an expression.
Parameters
==========
exprs : list of sympy expressions, or a single sympy expression
The expressions to reduce.
symbols : infinite iterator yielding unique Symbols
The symbols used to label the common subexpressions which are pulled
out. The ``numbered_symbols`` generator is useful. The default is a
stream of symbols of the form "x0", "x1", etc. This must be an
infinite iterator.
optimizations : list of (callable, callable) pairs
The (preprocessor, postprocessor) pairs of external optimization
functions. Optionally 'basic' can be passed for a set of predefined
basic optimizations. Such 'basic' optimizations were used by default
in old implementation, however they can be really slow on larger
expressions. Now, no pre or post optimizations are made by default.
postprocess : a function which accepts the two return values of cse and
returns the desired form of output from cse, e.g. if you want the
replacements reversed the function might be the following lambda:
lambda r, e: return reversed(r), e
order : string, 'none' or 'canonical'
The order by which Mul and Add arguments are processed. If set to
'canonical', arguments will be canonically ordered. If set to 'none',
ordering will be faster but dependent on expressions hashes, thus
machine dependent and variable. For large expressions where speed is a
concern, use the setting order='none'.
ignore : iterable of Symbols
Substitutions containing any Symbol from ``ignore`` will be ignored.
Returns
=======
replacements : list of (Symbol, expression) pairs
All of the common subexpressions that were replaced. Subexpressions
earlier in this list might show up in subexpressions later in this
list.
reduced_exprs : list of sympy expressions
The reduced expressions with all of the replacements above.
Examples
========
>>> from sympy import cse, SparseMatrix
>>> from sympy.abc import x, y, z, w
>>> cse(((w + x + y + z)*(w + y + z))/(w + x)**3)
([(x0, w + y + z)], [x0*(x + x0)/(w + x)**3])
Note that currently, y + z will not get substituted if -y - z is used.
>>> cse(((w + x + y + z)*(w - y - z))/(w + x)**3)
([(x0, w + x)], [(w - y - z)*(x0 + y + z)/x0**3])
List of expressions with recursive substitutions:
>>> m = SparseMatrix([x + y, x + y + z])
>>> cse([(x+y)**2, x + y + z, y + z, x + z + y, m])
([(x0, x + y), (x1, x0 + z)], [x0**2, x1, y + z, x1, Matrix([
[x0],
[x1]])])
Note: the type and mutability of input matrices is retained.
>>> isinstance(_[1][-1], SparseMatrix)
True
The user may disallow substitutions containing certain symbols:
>>> cse([y**2*(x + 1), 3*y**2*(x + 1)], ignore=(y,))
([(x0, x + 1)], [x0*y**2, 3*x0*y**2])
"""
from sympy.matrices import (MatrixBase, Matrix, ImmutableMatrix,
SparseMatrix, ImmutableSparseMatrix)
# Handle the case if just one expression was passed.
if isinstance(exprs, (Basic, MatrixBase)):
exprs = [exprs]
copy = exprs
temp = []
for e in exprs:
if isinstance(e, (Matrix, ImmutableMatrix)):
temp.append(Tuple(*e._mat))
elif isinstance(e, (SparseMatrix, ImmutableSparseMatrix)):
temp.append(Tuple(*e._smat.items()))
else:
temp.append(e)
exprs = temp
del temp
if optimizations is None:
optimizations = list()
elif optimizations == 'basic':
optimizations = basic_optimizations
# Preprocess the expressions to give us better optimization opportunities.
reduced_exprs = [preprocess_for_cse(e, optimizations) for e in exprs]
excluded_symbols = set().union(
*[expr.atoms(Symbol) for expr in reduced_exprs])
if symbols is None:
symbols = numbered_symbols()
else:
# In case we get passed an iterable with an __iter__ method instead of
# an actual iterator.
symbols = iter(symbols)
symbols = filter_symbols(symbols, excluded_symbols)
# Find other optimization opportunities.
opt_subs = opt_cse(reduced_exprs, order)
# Main CSE algorithm.
replacements, reduced_exprs = tree_cse(reduced_exprs, symbols, opt_subs,
order, ignore)
# Postprocess the expressions to return the expressions to canonical form.
exprs = copy
for i, (sym, subtree) in enumerate(replacements):
subtree = postprocess_for_cse(subtree, optimizations)
replacements[i] = (sym, subtree)
reduced_exprs = [
postprocess_for_cse(e, optimizations) for e in reduced_exprs
]
# Get the matrices back
for i, e in enumerate(exprs):
if isinstance(e, (Matrix, ImmutableMatrix)):
reduced_exprs[i] = Matrix(e.rows, e.cols, reduced_exprs[i])
if isinstance(e, ImmutableMatrix):
reduced_exprs[i] = reduced_exprs[i].as_immutable()
elif isinstance(e, (SparseMatrix, ImmutableSparseMatrix)):
m = SparseMatrix(e.rows, e.cols, {})
for k, v in reduced_exprs[i]:
m[k] = v
if isinstance(e, ImmutableSparseMatrix):
m = m.as_immutable()
reduced_exprs[i] = m
if postprocess is None:
return replacements, reduced_exprs
return postprocess(replacements, reduced_exprs)
def cse(exprs,
symbols=None,
optimizations=None,
postprocess=None,
order='canonical'):
""" Perform common subexpression elimination on an expression.
Parameters
==========
exprs : list of sympy expressions, or a single sympy expression
The expressions to reduce.
symbols : infinite iterator yielding unique Symbols
The symbols used to label the common subexpressions which are pulled
out. The ``numbered_symbols`` generator is useful. The default is a
stream of symbols of the form "x0", "x1", etc. This must be an
infinite iterator.
optimizations : list of (callable, callable) pairs
The (preprocessor, postprocessor) pairs of external optimization
functions. Optionally 'basic' can be passed for a set of predefined
basic optimizations. Such 'basic' optimizations were used by default
in old implementation, however they can be really slow on larger
expressions. Now, no pre or post optimizations are made by default.
postprocess : a function which accepts the two return values of cse and
returns the desired form of output from cse, e.g. if you want the
replacements reversed the function might be the following lambda:
lambda r, e: return reversed(r), e
order : string, 'none' or 'canonical'
The order by which Mul and Add arguments are processed. If set to
'canonical', arguments will be canonically ordered. If set to 'none',
ordering will be faster but dependent on expressions hashes, thus
machine dependent and variable. For large expressions where speed is a
concern, use the setting order='none'.
Returns
=======
replacements : list of (Symbol, expression) pairs
All of the common subexpressions that were replaced. Subexpressions
earlier in this list might show up in subexpressions later in this
list.
reduced_exprs : list of sympy expressions
The reduced expressions with all of the replacements above.
"""
from sympy.matrices import Matrix
# Handle the case if just one expression was passed.
if isinstance(exprs, Basic):
exprs = [exprs]
if optimizations is None:
optimizations = list()
elif optimizations == 'basic':
optimizations = basic_optimizations
# Preprocess the expressions to give us better optimization opportunities.
reduced_exprs = [preprocess_for_cse(e, optimizations) for e in exprs]
excluded_symbols = set.union(
*[expr.atoms(Symbol) for expr in reduced_exprs])
if symbols is None:
symbols = numbered_symbols()
else:
# In case we get passed an iterable with an __iter__ method instead of
# an actual iterator.
symbols = iter(symbols)
symbols = filter_symbols(symbols, excluded_symbols)
# Find other optimization opportunities.
opt_subs = opt_cse(reduced_exprs, order)
# Main CSE algorithm.
replacements, reduced_exprs = tree_cse(reduced_exprs, symbols, opt_subs,
order)
# Postprocess the expressions to return the expressions to canonical form.
for i, (sym, subtree) in enumerate(replacements):
subtree = postprocess_for_cse(subtree, optimizations)
replacements[i] = (sym, subtree)
reduced_exprs = [
postprocess_for_cse(e, optimizations) for e in reduced_exprs
]
if isinstance(exprs, Matrix):
reduced_exprs = [Matrix(exprs.rows, exprs.cols, reduced_exprs)]
if postprocess is None:
return replacements, reduced_exprs
return postprocess(replacements, reduced_exprs)
def test_filter_symbols():
s = numbered_symbols()
filtered = filter_symbols(s, symbols("x0 x2 x3"))
assert take(filtered, 3) == list(symbols("x1 x4 x5"))
def cse(exprs, symbols=None, optimizations=None, postprocess=None,
order='canonical'):
""" Perform common subexpression elimination on an expression.
Parameters
==========
exprs : list of sympy expressions, or a single sympy expression
The expressions to reduce.
symbols : infinite iterator yielding unique Symbols
The symbols used to label the common subexpressions which are pulled
out. The ``numbered_symbols`` generator is useful. The default is a
stream of symbols of the form "x0", "x1", etc. This must be an
infinite iterator.
optimizations : list of (callable, callable) pairs
The (preprocessor, postprocessor) pairs of external optimization
functions. Optionally 'basic' can be passed for a set of predefined
basic optimizations. Such 'basic' optimizations were used by default
in old implementation, however they can be really slow on larger
expressions. Now, no pre or post optimizations are made by default.
postprocess : a function which accepts the two return values of cse and
returns the desired form of output from cse, e.g. if you want the
replacements reversed the function might be the following lambda:
lambda r, e: return reversed(r), e
order : string, 'none' or 'canonical'
The order by which Mul and Add arguments are processed. If set to
'canonical', arguments will be canonically ordered. If set to 'none',
ordering will be faster but dependent on expressions hashes, thus
machine dependent and variable. For large expressions where speed is a
concern, use the setting order='none'.
Returns
=======
replacements : list of (Symbol, expression) pairs
All of the common subexpressions that were replaced. Subexpressions
earlier in this list might show up in subexpressions later in this
list.
reduced_exprs : list of sympy expressions
The reduced expressions with all of the replacements above.
"""
from sympy.matrices import Matrix
# Handle the case if just one expression was passed.
if isinstance(exprs, Basic):
exprs = [exprs]
if optimizations is None:
optimizations = list()
elif optimizations == 'basic':
optimizations = basic_optimizations
# Preprocess the expressions to give us better optimization opportunities.
reduced_exprs = [preprocess_for_cse(e, optimizations) for e in exprs]
excluded_symbols = set.union(*[expr.atoms(Symbol)
for expr in reduced_exprs])
if symbols is None:
symbols = numbered_symbols()
else:
# In case we get passed an iterable with an __iter__ method instead of
# an actual iterator.
symbols = iter(symbols)
symbols = filter_symbols(symbols, excluded_symbols)
# Find other optimization opportunities.
opt_subs = opt_cse(reduced_exprs, order)
# Main CSE algorithm.
replacements, reduced_exprs = tree_cse(reduced_exprs, symbols, opt_subs,
order)
# Postprocess the expressions to return the expressions to canonical form.
for i, (sym, subtree) in enumerate(replacements):
subtree = postprocess_for_cse(subtree, optimizations)
replacements[i] = (sym, subtree)
reduced_exprs = [postprocess_for_cse(e, optimizations)
for e in reduced_exprs]
if isinstance(exprs, Matrix):
reduced_exprs = [Matrix(exprs.rows, exprs.cols, reduced_exprs)]
if postprocess is None:
return replacements, reduced_exprs
return postprocess(replacements, reduced_exprs)
def cse(exprs, symbols=None, optimizations=None, postprocess=None,
order='canonical', ignore=()):
""" Perform common subexpression elimination on an expression.
Parameters
==========
exprs : list of sympy expressions, or a single sympy expression
The expressions to reduce.
symbols : infinite iterator yielding unique Symbols
The symbols used to label the common subexpressions which are pulled
out. The ``numbered_symbols`` generator is useful. The default is a
stream of symbols of the form "x0", "x1", etc. This must be an
infinite iterator.
optimizations : list of (callable, callable) pairs
The (preprocessor, postprocessor) pairs of external optimization
functions. Optionally 'basic' can be passed for a set of predefined
basic optimizations. Such 'basic' optimizations were used by default
in old implementation, however they can be really slow on larger
expressions. Now, no pre or post optimizations are made by default.
postprocess : a function which accepts the two return values of cse and
returns the desired form of output from cse, e.g. if you want the
replacements reversed the function might be the following lambda:
lambda r, e: return reversed(r), e
order : string, 'none' or 'canonical'
The order by which Mul and Add arguments are processed. If set to
'canonical', arguments will be canonically ordered. If set to 'none',
ordering will be faster but dependent on expressions hashes, thus
machine dependent and variable. For large expressions where speed is a
concern, use the setting order='none'.
ignore : iterable of Symbols
Substitutions containing any Symbol from ``ignore`` will be ignored.
Returns
=======
replacements : list of (Symbol, expression) pairs
All of the common subexpressions that were replaced. Subexpressions
earlier in this list might show up in subexpressions later in this
list.
reduced_exprs : list of sympy expressions
The reduced expressions with all of the replacements above.
Examples
========
>>> from sympy import cse, SparseMatrix
>>> from sympy.abc import x, y, z, w
>>> cse(((w + x + y + z)*(w + y + z))/(w + x)**3)
([(x0, w + y + z)], [x0*(x + x0)/(w + x)**3])
Note that currently, y + z will not get substituted if -y - z is used.
>>> cse(((w + x + y + z)*(w - y - z))/(w + x)**3)
([(x0, w + x)], [(w - y - z)*(x0 + y + z)/x0**3])
List of expressions with recursive substitutions:
>>> m = SparseMatrix([x + y, x + y + z])
>>> cse([(x+y)**2, x + y + z, y + z, x + z + y, m])
([(x0, x + y), (x1, x0 + z)], [x0**2, x1, y + z, x1, Matrix([
[x0],
[x1]])])
Note: the type and mutability of input matrices is retained.
>>> isinstance(_[1][-1], SparseMatrix)
True
The user may disallow substitutions containing certain symbols:
>>> cse([y**2*(x + 1), 3*y**2*(x + 1)], ignore=(y,))
([(x0, x + 1)], [x0*y**2, 3*x0*y**2])
"""
from sympy.matrices import (MatrixBase, Matrix, ImmutableMatrix,
SparseMatrix, ImmutableSparseMatrix)
# Handle the case if just one expression was passed.
if isinstance(exprs, (Basic, MatrixBase)):
exprs = [exprs]
copy = exprs
temp = []
for e in exprs:
if isinstance(e, (Matrix, ImmutableMatrix)):
temp.append(Tuple(*e._mat))
elif isinstance(e, (SparseMatrix, ImmutableSparseMatrix)):
temp.append(Tuple(*e._smat.items()))
else:
temp.append(e)
exprs = temp
del temp
if optimizations is None:
optimizations = list()
elif optimizations == 'basic':
optimizations = basic_optimizations
# Preprocess the expressions to give us better optimization opportunities.
reduced_exprs = [preprocess_for_cse(e, optimizations) for e in exprs]
excluded_symbols = set().union(*[expr.atoms(Symbol)
for expr in reduced_exprs])
if symbols is None:
symbols = numbered_symbols()
else:
# In case we get passed an iterable with an __iter__ method instead of
# an actual iterator.
symbols = iter(symbols)
symbols = filter_symbols(symbols, excluded_symbols)
# Find other optimization opportunities.
opt_subs = opt_cse(reduced_exprs, order)
# Main CSE algorithm.
replacements, reduced_exprs = tree_cse(reduced_exprs, symbols, opt_subs,
order, ignore)
# Postprocess the expressions to return the expressions to canonical form.
exprs = copy
for i, (sym, subtree) in enumerate(replacements):
subtree = postprocess_for_cse(subtree, optimizations)
replacements[i] = (sym, subtree)
reduced_exprs = [postprocess_for_cse(e, optimizations)
for e in reduced_exprs]
# Get the matrices back
for i, e in enumerate(exprs):
if isinstance(e, (Matrix, ImmutableMatrix)):
reduced_exprs[i] = Matrix(e.rows, e.cols, reduced_exprs[i])
if isinstance(e, ImmutableMatrix):
reduced_exprs[i] = reduced_exprs[i].as_immutable()
elif isinstance(e, (SparseMatrix, ImmutableSparseMatrix)):
m = SparseMatrix(e.rows, e.cols, {})
for k, v in reduced_exprs[i]:
m[k] = v
if isinstance(e, ImmutableSparseMatrix):
m = m.as_immutable()
reduced_exprs[i] = m
if postprocess is None:
return replacements, reduced_exprs
return postprocess(replacements, reduced_exprs)
