Files
rdkit/Python/ML/Descriptors/Parser.py
Greg Landrum 75a79b6327 initial import
2006-05-06 22:20:08 +00:00

436 lines
12 KiB
Python
Executable File

#
# Copyright (C) 2001-2004 greg Landrum and Rational Discovery LLC
# All Rights Reserved
#
""" The "parser" for compound descriptors.
I almost hesitate to document this, because it's not the prettiest
thing the world has ever seen... but it does work (for at least some
definitions of the word).
Rather than getting into the whole mess of writing a parser for the
compound descriptor expressions, I'm just using string substitutions
and python's wonderful ability to *eval* code.
It would probably be a good idea at some point to replace this with a
real parser, if only for the flexibility and intelligent error
messages that would become possible.
The general idea is that we're going to deal with expressions where
atomic descriptors have some kind of method applied to them which
reduces them to a single number for the entire composition. Compound
descriptors (those applicable to the compound as a whole) are not
operated on by anything in particular (except for standard math stuff).
Here's the general flow of things:
1) Composition descriptor references ($a, $b, etc.) are replaced with the
corresponding descriptor names using string subsitution.
(*_SubForCompoundDescriptors*)
2) Atomic descriptor references ($1, $2, etc) are replaced with lookups
into the atomic dict with "DEADBEEF" in place of the atom name.
(*_SubForAtomicVars*)
3) Calls to Calculator Functions are augmented with a reference to
the composition and atomic dictionary
(*_SubMethodArgs*)
**NOTE:**
anytime we don't know the answer for a descriptor, rather than
throwing a (completely incomprehensible) exception, we just return
-666. So bad descriptor values should stand out like sore thumbs.
"""
__DEBUG=0
import RDConfig
import string
# we do this to allow the use of stuff in the math module
from math import *
#----------------------
# atomic descriptor section
#----------------------
# these are the methods which can be applied to ATOMIC descriptors.
knownMethods = ['SUM','MIN','MAX','MEAN','AVG','DEV','HAS']
def HAS(strArg,composList,atomDict):
""" *Calculator Method*
does a string search
**Arguments**
- strArg: the arguments in string form
- composList: the composition vector
- atomDict: the atomic dictionary
**Returns**
1 or 0
"""
splitArgs = string.split(strArg,',')
if len(splitArgs)>1:
for atom,num in composList:
tStr = string.replace(splitArgs[0],'DEADBEEF',atom)
where = eval(tStr)
what = eval(splitArgs[1])
if string.find(where,what)!= -1:
return 1
return 0
else:
return -666
def SUM(strArg,composList,atomDict):
""" *Calculator Method*
calculates the sum of a descriptor across a composition
**Arguments**
- strArg: the arguments in string form
- compos: the composition vector
- atomDict: the atomic dictionary
**Returns**
a float
"""
accum = 0.0
for atom,num in composList:
tStr = string.replace(strArg,'DEADBEEF',atom)
accum = accum + eval(tStr)*num
return accum
def MEAN(strArg,composList,atomDict):
""" *Calculator Method*
calculates the average of a descriptor across a composition
**Arguments**
- strArg: the arguments in string form
- compos: the composition vector
- atomDict: the atomic dictionary
**Returns**
a float
"""
accum = 0.0
nSoFar = 0
for atom,num in composList:
tStr = string.replace(strArg,'DEADBEEF',atom)
accum = accum + eval(tStr)*num
nSoFar = nSoFar + num
return accum/nSoFar
AVG = MEAN
def DEV(strArg,composList,atomDict):
""" *Calculator Method*
calculates the average deviation of a descriptor across a composition
**Arguments**
- strArg: the arguments in string form
- compos: the composition vector
- atomDict: the atomic dictionary
**Returns**
a float
"""
avg = MEAN(strArg,composList,atomDict)
accum = 0.0
nSoFar = 0.0
for atom,num in composList:
tStr = string.replace(strArg,'DEADBEEF',atom)
accum = accum + abs(eval(tStr)-avg)*num
nSoFar = nSoFar + num
return accum/nSoFar
def MIN(strArg,composList,atomDict):
""" *Calculator Method*
calculates the minimum value of a descriptor across a composition
**Arguments**
- strArg: the arguments in string form
- compos: the composition vector
- atomDict: the atomic dictionary
**Returns**
a float
"""
accum = []
for atom,num in composList:
tStr = string.replace(strArg,'DEADBEEF',atom)
accum.append(eval(tStr))
return min(accum)
def MAX(strArg,composList,atomDict):
""" *Calculator Method*
calculates the maximum value of a descriptor across a composition
**Arguments**
- strArg: the arguments in string form
- compos: the composition vector
- atomDict: the atomic dictionary
**Returns**
a float
"""
accum = []
for atom,num in composList:
tStr = string.replace(strArg,'DEADBEEF',atom)
accum.append(eval(tStr))
return max(accum)
#------------------
# string replacement routines
# these are not intended to be called by clients
#------------------
def _SubForAtomicVars(cExpr,varList,dictName):
""" replace atomic variables with the appropriate dictionary lookup
*Not intended for client use*
"""
for i in xrange(len(varList)):
cExpr = string.replace(cExpr,'$%d'%(i+1),
'%s["DEADBEEF"]["%s"]'%(dictName,varList[i]))
return cExpr
def _SubForCompoundDescriptors(cExpr,varList,dictName):
""" replace compound variables with the appropriate list index
*Not intended for client use*
"""
for i in xrange(len(varList)):
cExpr = string.replace(cExpr,'$%s'%chr(ord('a')+i),
'%s["%s"]'%(dictName,varList[i]))
return cExpr
def _SubMethodArgs(cExpr,knownMethods):
""" alters the arguments of calls to calculator methods
*Not intended for client use*
This is kind of putrid (and the code ain't so pretty either)
The general idea is that the various special methods for atomic
descriptors need two extra arguments (the composition and the atomic
dict). Rather than make the user type those in, we just find
invocations of these methods and fill out the function calls using
string replacements.
"""
res = cExpr
for method in knownMethods:
p = 0
while p != -1 and p < len(res):
p = string.find(res,method,p)
if p != -1:
p = p + len(method) + 1
start = p
parenCount = 1
while parenCount and p < len(res):
if res[p] == ')':
parenCount = parenCount - 1
elif res[p] == '(':
parenCount = parenCount + 1
p = p + 1
if p <= len(res):
res = res[0:start]+"'%s',compos,atomDict"%(res[start:p-1])+res[p-1:]
return res
def CalcSingleCompoundDescriptor(compos,argVect,atomDict,propDict):
""" calculates the value of the descriptor for a single compound
**ARGUMENTS:**
- compos: a vector/tuple containing the composition
information... in the form:
'[("Fe",1.),("Pt",2.),("Rh",0.02)]'
- argVect: a vector/tuple with three elements:
1) AtomicDescriptorNames: a list/tuple of the names of the
atomic descriptors being used. These determine the
meaning of $1, $2, etc. in the expression
2) CompoundDescriptorNames: a list/tuple of the names of the
compound descriptors being used. These determine the
meaning of $a, $b, etc. in the expression
3) Expr: a string containing the expression to be used to
evaluate the final result.
- atomDict:
a dictionary of atomic descriptors. Each atomic entry is
another dictionary containing the individual descriptors
and their values
- propVect:
a list of descriptors for the composition.
**RETURNS:**
the value of the descriptor, -666 if a problem was encountered
**NOTE:**
- because it takes rather a lot of work to get everything set
up to calculate a descriptor, if you are calculating the
same descriptor for multiple compounds, you probably want to
be calling _CalcMultipleCompoundsDescriptor()_.
"""
try:
atomVarNames = argVect[0]
compositionVarNames = argVect[1]
formula = argVect[2]
formula = _SubForCompoundDescriptors(formula,compositionVarNames,'propDict')
formula = _SubForAtomicVars(formula,atomVarNames,'atomDict')
evalTarget = _SubMethodArgs(formula,knownMethods)
except:
if __DEBUG:
import sys,traceback
print 'Sub Failure!'
traceback.print_exc()
print evalTarget
print propDict
raise RuntimeError,'Failure 1'
else:
return -666
try:
v = eval(evalTarget)
except:
if __DEBUG:
import sys,traceback
outF = open(RDConfig.RDCodeDir+'/ml/descriptors/log.txt','a+')
outF.write('#------------------------------\n')
outF.write('formula: %s\n'%repr(formula))
outF.write('target: %s\n'%repr(evalTarget))
outF.write('propDict: %s\n'%(repr(propDict)))
try:
outF.write('keys: %s\n'%(repr(atomDict.keys())))
except:
outF.write('no atomDict\n')
outF.close()
print 'ick!'
print 'formula:',formula
print 'target:',evalTarget
print 'propDict:',propDict
print 'keys:',atomDict.keys()
traceback.print_exc()
raise RuntimeError,'Failure 2'
else:
v = -666
return v
def CalcMultipleCompoundsDescriptor(composVect,argVect,atomDict,propDictList):
""" calculates the value of the descriptor for a list of compounds
**ARGUMENTS:**
- composVect: a vector of vector/tuple containing the composition
information.
See _CalcSingleCompoundDescriptor()_ for an explanation of the elements.
- argVect: a vector/tuple with three elements:
1) AtomicDescriptorNames: a list/tuple of the names of the
atomic descriptors being used. These determine the
meaning of $1, $2, etc. in the expression
2) CompoundDsscriptorNames: a list/tuple of the names of the
compound descriptors being used. These determine the
meaning of $a, $b, etc. in the expression
3) Expr: a string containing the expression to be used to
evaluate the final result.
- atomDict:
a dictionary of atomic descriptors. Each atomic entry is
another dictionary containing the individual descriptors
and their values
- propVectList:
a vector of vectors of descriptors for the composition.
**RETURNS:**
a vector containing the values of the descriptor for each
compound. Any given entry will be -666 if problems were
encountered
"""
res = [-666]*len(composVect)
try:
atomVarNames = argVect[0]
compositionVarNames = argVect[1]
formula = argVect[2]
formula = _SubForCompoundDescriptors(formula,compositionVarNames,'propDict')
formula = _SubForAtomicVars(formula,atomVarNames,'atomDict')
evalTarget = _SubMethodArgs(formula,knownMethods)
except:
return res
for i in xrange(len(composVect)):
propDict = propDictList[i]
compos = composVect[i]
try:
v = eval(evalTarget)
except:
v = -666
res[i] = v
return res
#------------
# Demo/testing code
#------------
if __name__ == '__main__':
piece1 = [['d1','d2'],['d1','d2']]
aDict = {'Fe':{'d1':1.,'d2':2.},'Pt':{'d1':10.,'d2':20.}}
pDict = {'d1':100.,'d2':200.}
compos = [('Fe',1),('Pt',1)]
cExprs = ["SUM($1)","SUM($1)+SUM($2)","SUM($1)+SUM($1)","MEAN($1)","DEV($2)","MAX($1)","MIN($1)/MAX($1)",
"MIN($2)","SUM($1)/$a","sqrt($a+$b)","SUM((3.*$1)/($2))","foo"]
for cExpr in cExprs:
argVect = piece1 + [cExpr]
print cExpr
print CalcSingleCompoundDescriptor(compos,argVect,aDict,pDict)
print CalcMultipleCompoundsDescriptor([compos,compos],argVect,aDict,[pDict,pDict])