def GenVarTable(examples, nPossibleVals, vars):
"""Generates a list of variable tables for the examples passed in.
The table for a given variable records the number of times each possible value
of that variable appears for each possible result of the function.
**Arguments**
- examples: a list (nInstances long) of lists of variable values + instance
values
- nPossibleVals: a list containing the number of possible values of
each variable + the number of values of the function.
- vars: a list of the variables to include in the var table
**Returns**
a list of variable result tables. Each table is a Numeric array
which is varValues x nResults
"""
nVars = len(vars)
res = [None] * nVars
nFuncVals = nPossibleVals[-1]
for i in xrange(nVars):
res[i] = numpy.zeros((nPossibleVals[vars[i]], nFuncVals), 'i')
for example in examples:
val = int(example[-1])
for i in xrange(nVars):
res[i][int(example[vars[i]]), val] += 1
return res
def ConstructNodes(self, nodeCounts, actFunc, actFuncParms):
""" build an unconnected network and set node counts
**Arguments**
- nodeCounts: a list containing the number of nodes to be in each layer.
the ordering is:
(nInput,nHidden1,nHidden2, ... , nHiddenN, nOutput)
"""
self.nodeCounts = nodeCounts
self.numInputNodes = nodeCounts[0]
self.numOutputNodes = nodeCounts[-1]
self.numHiddenLayers = len(nodeCounts) - 2
self.numInHidden = [None] * self.numHiddenLayers
for i in xrange(self.numHiddenLayers):
self.numInHidden[i] = nodeCounts[i + 1]
numNodes = sum(self.nodeCounts)
self.nodeList = [None] * (numNodes)
for i in xrange(numNodes):
self.nodeList[i] = NetNode.NetNode(i,
self.nodeList,
actFunc=actFunc,
actFuncParms=actFuncParms)
self.layerIndices = [None] * len(nodeCounts)
start = 0
for i in xrange(len(nodeCounts)):
end = start + nodeCounts[i]
self.layerIndices[i] = range(start, end)
start = end
def ConstructNodes(self,nodeCounts,actFunc,actFuncParms):
""" build an unconnected network and set node counts
**Arguments**
- nodeCounts: a list containing the number of nodes to be in each layer.
the ordering is:
(nInput,nHidden1,nHidden2, ... , nHiddenN, nOutput)
"""
self.nodeCounts = nodeCounts
self.numInputNodes = nodeCounts[0]
self.numOutputNodes = nodeCounts[-1]
self.numHiddenLayers = len(nodeCounts)-2
self.numInHidden = [None]*self.numHiddenLayers
for i in xrange(self.numHiddenLayers):
self.numInHidden[i] = nodeCounts[i+1]
numNodes = sum(self.nodeCounts)
self.nodeList = [None]*(numNodes)
for i in xrange(numNodes):
self.nodeList[i] = NetNode.NetNode(i,self.nodeList,
actFunc=actFunc,
actFuncParms=actFuncParms)
self.layerIndices = [None]*len(nodeCounts)
start = 0
for i in xrange(len(nodeCounts)):
end = start + nodeCounts[i]
self.layerIndices[i] = range(start,end)
start = end
def GenVarTable(examples,nPossibleVals,vars):
"""Generates a list of variable tables for the examples passed in.
The table for a given variable records the number of times each possible value
of that variable appears for each possible result of the function.
**Arguments**
- examples: a list (nInstances long) of lists of variable values + instance
values
- nPossibleVals: a list containing the number of possible values of
each variable + the number of values of the function.
- vars: a list of the variables to include in the var table
**Returns**
a list of variable result tables. Each table is a Numeric array
which is varValues x nResults
"""
nVars = len(vars)
res = [None]*nVars
nFuncVals = nPossibleVals[-1]
for i in xrange(nVars):
res[i] = numpy.zeros((nPossibleVals[vars[i]],nFuncVals),'i')
for example in examples:
val = int(example[-1])
for i in xrange(nVars):
res[i][int(example[vars[i]]),val] += 1
return res
def ClassifyExample(self,example,appendExamples=0):
""" classifies a given example and returns the results of the output layer.
**Arguments**
- example: the example to be classified
**NOTE:**
if the output layer is only one element long,
a scalar (not a list) will be returned. This is why a lot of the other
network code claims to only support single valued outputs.
"""
if len(example) > self.numInputNodes:
if len(example)-self.numInputNodes > self.numOutputNodes:
example = example[1:-self.numOutputNodes]
else:
example = example[:-self.numOutputNodes]
assert len(example) == self.numInputNodes
totNumNodes = sum(self.nodeCounts)
results = numpy.zeros(totNumNodes,numpy.float64)
for i in xrange(self.numInputNodes):
results[i] = example[i]
for i in xrange(self.numInputNodes,totNumNodes):
self.nodeList[i].Eval(results)
self.lastResults = results[:]
if self.numOutputNodes == 1:
return results[-1]
else:
return results
def ClassifyExample(self, example, appendExamples=0):
""" classifies a given example and returns the results of the output layer.
**Arguments**
- example: the example to be classified
**NOTE:**
if the output layer is only one element long,
a scalar (not a list) will be returned. This is why a lot of the other
network code claims to only support single valued outputs.
"""
if len(example) > self.numInputNodes:
if len(example) - self.numInputNodes > self.numOutputNodes:
example = example[1:-self.numOutputNodes]
else:
example = example[:-self.numOutputNodes]
assert len(example) == self.numInputNodes
totNumNodes = sum(self.nodeCounts)
results = numpy.zeros(totNumNodes, numpy.float64)
for i in xrange(self.numInputNodes):
results[i] = example[i]
for i in xrange(self.numInputNodes, totNumNodes):
self.nodeList[i].Eval(results)
self.lastResults = results[:]
if self.numOutputNodes == 1:
return results[-1]
else:
return results
def CalcNPossibleUsingMap(data,order,qBounds,nQBounds=None):
""" calculates the number of possible values for each variable in a data set
**Arguments**
- data: a list of examples
- order: the ordering map between the variables in _data_ and _qBounds_
- qBounds: the quantization bounds for the variables
**Returns**
a list with the number of possible values each variable takes on in the data set
**Notes**
- variables present in _qBounds_ will have their _nPossible_ number read
from _qbounds
- _nPossible_ for other numeric variables will be calculated
"""
numericTypes = [int, float]
if six.PY2:
numericTypes.append(long)
print('order:',order, len(order))
print('qB:',qBounds)
#print('nQB:',nQBounds, len(nQBounds))
assert (qBounds and len(order)==len(qBounds)) or (nQBounds and len(order)==len(nQBounds)),\
'order/qBounds mismatch'
nVars = len(order)
nPossible = [-1]*nVars
cols = range(nVars)
for i in xrange(nVars):
if nQBounds and nQBounds[i] != 0:
nPossible[i] = -1
cols.remove(i)
elif len(qBounds[i])>0:
nPossible[i] = len(qBounds[i])
cols.remove(i)
nPts = len(data)
for i in xrange(nPts):
for col in cols[:]:
d = data[i][order[col]]
if type(d) in numericTypes:
if int(d) == d:
nPossible[col] = max(int(d),nPossible[col])
else:
nPossible[col] = -1
cols.remove(col)
else:
print('bye bye col %d: %s'%(col,repr(d)))
nPossible[col] = -1
cols.remove(col)
return list(map(lambda x:int(x)+1,nPossible))
def CalcNPossibleUsingMap(data, order, qBounds, nQBounds=None):
""" calculates the number of possible values for each variable in a data set
**Arguments**
- data: a list of examples
- order: the ordering map between the variables in _data_ and _qBounds_
- qBounds: the quantization bounds for the variables
**Returns**
a list with the number of possible values each variable takes on in the data set
**Notes**
- variables present in _qBounds_ will have their _nPossible_ number read
from _qbounds
- _nPossible_ for other numeric variables will be calculated
"""
numericTypes = [int, float]
if six.PY2:
numericTypes.append(long)
print('order:', order, len(order))
print('qB:', qBounds)
#print('nQB:',nQBounds, len(nQBounds))
assert (qBounds and len(order)==len(qBounds)) or (nQBounds and len(order)==len(nQBounds)),\
'order/qBounds mismatch'
nVars = len(order)
nPossible = [-1] * nVars
cols = range(nVars)
for i in xrange(nVars):
if nQBounds and nQBounds[i] != 0:
nPossible[i] = -1
cols.remove(i)
elif len(qBounds[i]) > 0:
nPossible[i] = len(qBounds[i])
cols.remove(i)
nPts = len(data)
for i in xrange(nPts):
for col in cols[:]:
d = data[i][order[col]]
if type(d) in numericTypes:
if int(d) == d:
nPossible[col] = max(int(d), nPossible[col])
else:
nPossible[col] = -1
cols.remove(col)
else:
print('bye bye col %d: %s' % (col, repr(d)))
nPossible[col] = -1
cols.remove(col)
return list(map(lambda x: int(x) + 1, nPossible))
def test3Classify(self):
" testing classification "
self._setupTree1()
self._setupTree2()
for i in xrange(len(self.examples1)):
assert self.t1.ClassifyExample(self.examples1[i])==self.examples1[i][-1],\
'examples1[%d] misclassified'%i
for i in xrange(len(self.examples2)):
assert self.t2.ClassifyExample(self.examples2[i])==self.examples2[i][-1],\
'examples2[%d] misclassified'%i
def test3Classify(self):
" testing classification "
self._setupTree1()
self._setupTree2()
for i in xrange(len(self.examples1)):
assert self.t1.ClassifyExample(self.examples1[i])==self.examples1[i][-1],\
'examples1[%d] misclassified'%i
for i in xrange(len(self.examples2)):
assert self.t2.ClassifyExample(self.examples2[i])==self.examples2[i][-1],\
'examples2[%d] misclassified'%i
def TypeFinder(data, nRows, nCols, nullMarker=None):
"""
finds the types of the columns in _data_
if nullMarker is not None, elements of the data table which are
equal to nullMarker will not count towards setting the type of
their columns.
"""
priorities = {float: 3, int: 2, str: 1, -1: -1}
res = [None] * nCols
for col in xrange(nCols):
typeHere = [-1, 1]
for row in xrange(nRows):
d = data[row][col]
if d is None:
continue
locType = type(d)
if locType != float and locType != int:
locType = str
try:
d = str(d)
except UnicodeError as msg:
print(
'cannot convert text from row %d col %d to a string' %
(row + 2, col))
print('\t>%s' % (repr(d)))
raise UnicodeError(msg)
else:
typeHere[1] = max(typeHere[1], len(str(d)))
if isinstance(d, string_types):
if nullMarker is None or d != nullMarker:
l = max(len(d), typeHere[1])
typeHere = [str, l]
else:
try:
fD = float(int(d))
except OverflowError:
locType = float
else:
if fD == d:
locType = int
if not isinstance(typeHere[0], string_types) and \
priorities[locType] > priorities[typeHere[0]]:
typeHere[0] = locType
res[col] = typeHere
return res
def testSingleCalcs(self):
" testing calculation of a single descriptor "
for i in xrange(len(self.cExprs)):
cExpr= self.cExprs[i]
argVect = self.piece1 + [cExpr]
res = Parser.CalcSingleCompoundDescriptor(self.compos,argVect,self.aDict,self.pDict)
self.assertAlmostEqual(res,self.results[i],2)
def CrossValidate(tree, testExamples, appendExamples=0):
""" Determines the classification error for the testExamples
**Arguments**
- tree: a decision tree (or anything supporting a _ClassifyExample()_ method)
- testExamples: a list of examples to be used for testing
- appendExamples: a toggle which is passed along to the tree as it does
the classification. The trees can use this to store the examples they
classify locally.
**Returns**
a 2-tuple consisting of:
1) the percent error of the tree
2) a list of misclassified examples
"""
nTest = len(testExamples)
nBad = 0
badExamples = []
for i in xrange(nTest):
testEx = testExamples[i]
trueRes = testEx[-1]
res = tree.ClassifyExample(testEx, appendExamples)
if (trueRes != res).any():
badExamples.append(testEx)
nBad += 1
return float(nBad) / nTest, badExamples
def CharacteristicPolynomial(mol,mat=None):
""" calculates the characteristic polynomial for a molecular graph
if mat is not passed in, the molecule's Weighted Adjacency Matrix will
be used.
The approach used is the Le Verrier-Faddeev-Frame method described
in _Chemical Graph Theory, 2nd Edition_ by Nenad Trinajstic (CRC Press,
1992), pg 76.
"""
nAtoms = mol.GetNumAtoms()
if mat is None:
# FIX: complete this:
#A = mol.GetWeightedAdjacencyMatrix()
pass
else:
A = mat
I = 1.*numpy.identity(nAtoms)
An = A
res = numpy.zeros(nAtoms+1,numpy.float)
res[0] = 1.0
for n in xrange(1,nAtoms+1):
res[n] = 1./n*numpy.trace(An)
Bn = An - res[n]*I
An = numpy.dot(A,Bn)
res[1:] *= -1
return res
def PyInfoGain(varMat):
""" calculates the information gain for a variable
**Arguments**
varMat is a Numeric array with the number of possible occurances
of each result for reach possible value of the given variable.
So, for a variable which adopts 4 possible values and a result which
has 3 possible values, varMat would be 4x3
**Returns**
The expected information gain
"""
variableRes = numpy.sum(varMat,1) # indexed by variable, Sv in Mitchell's notation
overallRes = numpy.sum(varMat,0) # indexed by result, S in Mitchell's notation
term2 = 0
for i in xrange(len(variableRes)):
term2 = term2 + variableRes[i] * InfoEntropy(varMat[i])
tSum = sum(overallRes)
if tSum != 0.0:
term2 = 1./tSum * term2
gain = InfoEntropy(overallRes) - term2
else:
gain = 0
return gain
def ProcessSimpleList(self):
""" Handles the list of simple descriptors
This constructs the list of _nonZeroDescriptors_ and _requiredDescriptors_.
There's some other magic going on that I can't decipher at the moment.
"""
global countOptions
self.nonZeroDescriptors = []
lCopy = self.simpleList[:]
tList = map(lambda x: x[0], countOptions)
for i in xrange(len(lCopy)):
entry = lCopy[i]
if 'NONZERO' in entry[1]:
if entry[0] not in tList:
self.nonZeroDescriptors.append('%s != 0' % entry[0])
if len(entry[1]) == 1:
self.simpleList.remove(entry)
else:
self.simpleList[self.simpleList.index(entry)][1].remove(
'NONZERO')
self.requiredDescriptors = map(lambda x: x[0], self.simpleList)
for entry in tList:
if entry in self.requiredDescriptors:
self.requiredDescriptors.remove(entry)
def ReadVars(inFile):
""" reads the variables and quantization bounds from a .qdat or .dat file
**Arguments**
- inFile: a file object
**Returns**
a 2-tuple containing:
1) varNames: a list of the variable names
2) qbounds: the list of quantization bounds for each variable
"""
varNames = []
qBounds = []
fileutils.MoveToMatchingLine(inFile,'Variable Table')
inLine = inFile.readline()
while inLine.find('# ----') == -1:
splitLine = inLine[2:].split('[')
varNames.append(splitLine[0].strip())
qBounds.append(splitLine[1][:-2])
inLine = inFile.readline()
for i in xrange(len(qBounds)):
if qBounds[i] != '':
l = qBounds[i].split(',')
qBounds[i] = []
for item in l:
qBounds[i].append(float(item))
else:
qBounds[i] = []
return varNames,qBounds
def testTreeGrow(self):
" testing tree-based composite "
with open(
RDConfig.RDCodeDir +
'/ML/Composite/test_data/composite_base.pkl', 'rb') as pklF:
self.refCompos = cPickle.load(pklF)
composite = Composite.Composite()
composite._varNames = self.varNames
composite.SetQuantBounds(self.qBounds, self.nPoss)
from rdkit.ML.DecTree import CrossValidate
driver = CrossValidate.CrossValidationDriver
pruner = None
composite.Grow(self.examples,
self.attrs, [],
buildDriver=driver,
pruner=pruner,
nTries=100,
silent=1)
composite.AverageErrors()
composite.SortModels()
#with open(RDConfig.RDCodeDir+'/ML/Composite/test_data/composite_base.pkl','wb') as pklF:
# cPickle.dump(composite,pklF)
self.treeComposite = composite
self.assertEqual(len(composite), len(self.refCompos))
for i in xrange(len(composite)):
t1, c1, e1 = composite[i]
t2, c2, e2 = self.refCompos[i]
self.assertEqual(e1, e2)
def ChooseOptimalRoot(examples, trainExamples, testExamples, attrs, nPossibleVals, treeBuilder,
nQuantBounds=[], **kwargs):
""" loops through all possible tree roots and chooses the one which produces the best tree
**Arguments**
- examples: the full set of examples
- trainExamples: the training examples
- testExamples: the testing examples
- attrs: a list of attributes to consider in the tree building
- nPossibleVals: a list of the number of possible values each variable can adopt
- treeBuilder: the function to be used to actually build the tree
- nQuantBounds: an optional list. If present, it's assumed that the builder
algorithm takes this argument as well (for building QuantTrees)
**Returns**
The best tree found
**Notes**
1) Trees are built using _trainExamples_
2) Testing of each tree (to determine which is best) is done using _CrossValidate_ and
the entire set of data (i.e. all of _examples_)
3) _trainExamples_ is not used at all, which immediately raises the question of
why it's even being passed in
"""
attrs = attrs[:]
if nQuantBounds:
for i in range(len(nQuantBounds)):
if nQuantBounds[i] == -1 and i in attrs:
attrs.remove(i)
nAttrs = len(attrs)
trees = [None] * nAttrs
errs = [0] * nAttrs
errs[0] = 1e6
for i in xrange(1, nAttrs):
argD = {'initialVar': attrs[i]}
argD.update(kwargs)
if nQuantBounds is None or nQuantBounds == []:
trees[i] = treeBuilder(trainExamples, attrs, nPossibleVals, **argd)
else:
trees[i] = treeBuilder(trainExamples, attrs, nPossibleVals, nQuantBounds, **argD)
if trees[i]:
errs[i], foo = CrossValidate(trees[i], examples, appendExamples=0)
else:
errs[i] = 1e6
best = numpy.argmin(errs)
# FIX: this used to say 'trees[i]', could that possibly have been right?
return trees[best]
def CalcCompoundDescriptorsForComposition(self, compos='', composList=None, propDict={}):
""" calculates all simple descriptors for a given composition
**Arguments**
- compos: a string representation of the composition
- composList: a *composVect*
- propDict: a dictionary containing the properties of the composition
as a whole (e.g. structural variables, etc.)
The client must provide either _compos_ or _composList_. If both are
provided, _composList_ takes priority.
**Returns**
the list of descriptor values
**Notes**
- when _compos_ is provided, this uses _chemutils.SplitComposition_
to split the composition into its individual pieces
"""
if composList is None:
composList = chemutils.SplitComposition(compos)
res = []
for i in xrange(len(self.compoundList)):
val = Parser.CalcSingleCompoundDescriptor(composList, self.compoundList[i][1:], self.atomDict,
propDict)
res.append(val)
return res
def ProcessSimpleList(self):
""" Handles the list of simple descriptors
This constructs the list of _nonZeroDescriptors_ and _requiredDescriptors_.
There's some other magic going on that I can't decipher at the moment.
"""
global countOptions
self.nonZeroDescriptors = []
lCopy = self.simpleList[:]
tList = map(lambda x: x[0], countOptions)
for i in xrange(len(lCopy)):
entry = lCopy[i]
if 'NONZERO' in entry[1]:
if entry[0] not in tList:
self.nonZeroDescriptors.append('%s != 0' % entry[0])
if len(entry[1]) == 1:
self.simpleList.remove(entry)
else:
self.simpleList[self.simpleList.index(entry)][1].remove('NONZERO')
self.requiredDescriptors = map(lambda x: x[0], self.simpleList)
for entry in tList:
if entry in self.requiredDescriptors:
self.requiredDescriptors.remove(entry)
def testTreeGrow(self):
" testing tree-based composite "
with open(RDConfig.RDCodeDir + "/ML/Composite/test_data/composite_base.pkl", "r") as pklTF:
buf = pklTF.read().replace("\r\n", "\n").encode("utf-8")
pklTF.close()
with io.BytesIO(buf) as pklF:
self.refCompos = cPickle.load(pklF)
composite = Composite.Composite()
composite._varNames = self.varNames
composite.SetQuantBounds(self.qBounds, self.nPoss)
from rdkit.ML.DecTree import CrossValidate
driver = CrossValidate.CrossValidationDriver
pruner = None
composite.Grow(self.examples, self.attrs, [], buildDriver=driver, pruner=pruner, nTries=100, silent=1)
composite.AverageErrors()
composite.SortModels()
# with open(RDConfig.RDCodeDir+'/ML/Composite/test_data/composite_base.pkl','wb') as pklF:
# cPickle.dump(composite,pklF)
self.treeComposite = composite
self.assertEqual(len(composite), len(self.refCompos))
for i in xrange(len(composite)):
t1, c1, e1 = composite[i]
t2, c2, e2 = self.refCompos[i]
self.assertEqual(e1, e2)
def PyInfoGain(varMat):
""" calculates the information gain for a variable
**Arguments**
varMat is a Numeric array with the number of possible occurances
of each result for reach possible value of the given variable.
So, for a variable which adopts 4 possible values and a result which
has 3 possible values, varMat would be 4x3
**Returns**
The expected information gain
"""
variableRes = numpy.sum(varMat, 1) # indexed by variable, Sv in Mitchell's notation
overallRes = numpy.sum(varMat, 0) # indexed by result, S in Mitchell's notation
term2 = 0
for i in xrange(len(variableRes)):
term2 = term2 + variableRes[i] * InfoEntropy(varMat[i])
tSum = sum(overallRes)
if tSum != 0.0:
term2 = 1. / tSum * term2
gain = InfoEntropy(overallRes) - term2
else:
gain = 0
return gain
def testSingleCalcs(self):
" testing calculation of a single descriptor "
for i in xrange(len(self.cExprs)):
cExpr = self.cExprs[i]
argVect = self.piece1 + [cExpr]
res = Parser.CalcSingleCompoundDescriptor(self.compos, argVect, self.aDict, self.pDict)
self.assertAlmostEqual(res, self.results[i], 2)
def _CalcNPossible(self, data):
"""calculates the number of possible values of each variable (where possible)
**Arguments**
-data: a list of examples to be used
**Returns**
a list of nPossible values for each variable
"""
nVars = self.GetNVars() + self.nResults
nPossible = [-1] * nVars
cols = list(xrange(nVars))
for i, bounds in enumerate(self.qBounds):
if len(bounds) > 0:
nPossible[i] = len(bounds)
cols.remove(i)
nPts = self.GetNPts()
for i, pt in enumerate(self.data):
for col in cols[:]:
d = pt[col]
if type(d) in numericTypes:
if math.floor(d) == d:
nPossible[col] = max(math.floor(d), nPossible[col])
else:
nPossible[col] = -1
cols.remove(col)
else:
nPossible[col] = -1
cols.remove(col)
return [int(x) + 1 for x in nPossible]
def ReadVars(inFile):
""" reads the variables and quantization bounds from a .qdat or .dat file
**Arguments**
- inFile: a file object
**Returns**
a 2-tuple containing:
1) varNames: a list of the variable names
2) qbounds: the list of quantization bounds for each variable
"""
varNames = []
qBounds = []
fileutils.MoveToMatchingLine(inFile, 'Variable Table')
inLine = inFile.readline()
while inLine.find('# ----') == -1:
splitLine = inLine[2:].split('[')
varNames.append(splitLine[0].strip())
qBounds.append(splitLine[1][:-2])
inLine = inFile.readline()
for i in xrange(len(qBounds)):
if qBounds[i] != '':
l = qBounds[i].split(',')
qBounds[i] = []
for item in l:
qBounds[i].append(float(item))
else:
qBounds[i] = []
return varNames, qBounds
def CrossValidate(tree, testExamples, appendExamples=0):
""" Determines the classification error for the testExamples
**Arguments**
- tree: a decision tree (or anything supporting a _ClassifyExample()_ method)
- testExamples: a list of examples to be used for testing
- appendExamples: a toggle which is passed along to the tree as it does
the classification. The trees can use this to store the examples they
classify locally.
**Returns**
a 2-tuple consisting of:
1) the percent error of the tree
2) a list of misclassified examples
"""
nTest = len(testExamples)
nBad = 0
badExamples = []
for i in xrange(nTest):
testEx = testExamples[i]
trueRes = testEx[-1]
res = tree.ClassifyExample(testEx, appendExamples)
if (trueRes != res).any():
badExamples.append(testEx)
nBad += 1
return float(nBad) / nTest, badExamples
def CharacteristicPolynomial(mol, mat=None):
""" calculates the characteristic polynomial for a molecular graph
if mat is not passed in, the molecule's Weighted Adjacency Matrix will
be used.
The approach used is the Le Verrier-Faddeev-Frame method described
in _Chemical Graph Theory, 2nd Edition_ by Nenad Trinajstic (CRC Press,
1992), pg 76.
"""
nAtoms = mol.GetNumAtoms()
if mat is None:
# FIX: complete this:
#A = mol.GetWeightedAdjacencyMatrix()
pass
else:
A = mat
I = 1. * numpy.identity(nAtoms)
An = A
res = numpy.zeros(nAtoms + 1, numpy.float)
res[0] = 1.0
for n in xrange(1, nAtoms + 1):
res[n] = 1. / n * numpy.trace(An)
Bn = An - res[n] * I
An = numpy.dot(A, Bn)
res[1:] *= -1
return res
def TypeFinder(data, nRows, nCols, nullMarker=None):
"""
finds the types of the columns in _data_
if nullMarker is not None, elements of the data table which are
equal to nullMarker will not count towards setting the type of
their columns.
"""
priorities = {float: 3, int: 2, str: 1, -1: -1}
res = [None] * nCols
for col in xrange(nCols):
typeHere = [-1, 1]
for row in xrange(nRows):
d = data[row][col]
if d is None:
continue
locType = type(d)
if locType != float and locType != int:
locType = str
try:
d = str(d)
except UnicodeError as msg: # pragma: nocover
print('cannot convert text from row %d col %d to a string' % (row + 2, col))
print('\t>%s' % (repr(d)))
raise UnicodeError(msg)
else:
typeHere[1] = max(typeHere[1], len(str(d)))
if isinstance(d, string_types):
if nullMarker is None or d != nullMarker:
l = max(len(d), typeHere[1])
typeHere = [str, l]
else:
try:
fD = float(int(d))
except OverflowError: # pragma: nocover
locType = float
else:
if fD == d:
locType = int
if not isinstance(typeHere[0], string_types) and \
priorities[locType] > priorities[typeHere[0]]:
typeHere[0] = locType
res[col] = typeHere
return res
def GetAtomicData(atomDict, descriptorsDesired, dBase=_atomDbName, table='atomic_data', where='',
user='******', password='******', includeElCounts=0):
""" pulls atomic data from a database
**Arguments**
- atomDict: the dictionary to populate
- descriptorsDesired: the descriptors to pull for each atom
- dBase: the DB to use
- table: the DB table to use
- where: the SQL where clause
- user: the user name to use with the DB
- password: the password to use with the DB
- includeElCounts: if nonzero, valence electron count fields are added to
the _atomDict_
"""
extraFields = ['NVAL', 'NVAL_NO_FULL_F', 'NVAL_NO_FULL_D', 'NVAL_NO_FULL']
from rdkit.Dbase import DbModule
cn = DbModule.connect(dBase, user, password)
c = cn.cursor()
descriptorsDesired = [s.upper() for s in descriptorsDesired]
if 'NAME' not in descriptorsDesired:
descriptorsDesired.append('NAME')
if includeElCounts and 'CONFIG' not in descriptorsDesired:
descriptorsDesired.append('CONFIG')
for field in extraFields:
if field in descriptorsDesired:
descriptorsDesired.remove(field)
toPull = ','.join(descriptorsDesired)
command = 'select %s from atomic_data %s' % (toPull, where)
try:
c.execute(command)
except Exception:
print('Problems executing command:', command)
return
res = c.fetchall()
for atom in res:
tDict = {}
for i in xrange(len(descriptorsDesired)):
desc = descriptorsDesired[i]
val = atom[i]
tDict[desc] = val
name = tDict['NAME']
atomDict[name] = tDict
if includeElCounts:
config = atomDict[name]['CONFIG']
atomDict[name]['NVAL'] = ConfigToNumElectrons(config)
atomDict[name]['NVAL_NO_FULL_F'] = ConfigToNumElectrons(config, ignoreFullF=1)
atomDict[name]['NVAL_NO_FULL_D'] = ConfigToNumElectrons(config, ignoreFullD=1)
atomDict[name]['NVAL_NO_FULL'] = ConfigToNumElectrons(config, ignoreFullF=1, ignoreFullD=1)
def test4UnusedVars(self):
" testing unused variables "
self._setupTree1a()
with open(self.qTree1Name,'rb') as inFile:
t2 = cPickle.load(inFile)
assert self.t1 == t2, 'Incorrect tree generated.'
for i in xrange(len(self.examples1)):
assert self.t1.ClassifyExample(self.examples1[i])==self.examples1[i][-1],\
'examples1[%d] misclassified'%i
def test4UnusedVars(self):
" testing unused variables "
self._setupTree1a()
with open(self.qTree1Name, 'rb') as inFile:
t2 = cPickle.load(inFile)
assert self.t1 == t2, 'Incorrect tree generated.'
for i in xrange(len(self.examples1)):
assert self.t1.ClassifyExample(self.examples1[i])==self.examples1[i][-1],\
'examples1[%d] misclassified'%i
def testMultipleCalcs(self):
" testing calculation of multiple descriptors "
for i in xrange(len(self.cExprs)):
cExpr = self.cExprs[i]
argVect = self.piece1 + [cExpr]
res = Parser.CalcMultipleCompoundsDescriptor([self.compos, self.compos], argVect, self.aDict,
[self.pDict, self.pDict])
self.assertAlmostEqual(res[0], self.results[i], 2)
self.assertAlmostEqual(res[1], self.results[i], 2)
def testMultipleCalcs(self):
" testing calculation of multiple descriptors "
for i in xrange(len(self.cExprs)):
cExpr= self.cExprs[i]
argVect = self.piece1 + [cExpr]
res = Parser.CalcMultipleCompoundsDescriptor([self.compos,self.compos],argVect,
self.aDict,[self.pDict,self.pDict])
self.assertAlmostEqual(res[0],self.results[i],2)
self.assertAlmostEqual(res[1],self.results[i],2)
def __str__(self):
""" provides a string representation of the network """
outStr = 'Network:\n'
for i in xrange(len(self.nodeList)):
outStr = outStr + '\tnode(% 3d):\n'%i
outStr = outStr + '\t\tinputs: %s\n'%(str(self.nodeList[i].GetInputs()))
outStr = outStr + '\t\tweights: %s\n'%(str(self.nodeList[i].GetWeights()))
outStr = outStr + 'Total Number of Connections: % 4d'%self.nConnections
return outStr
def testQuantize(self):
" testing data quantization "
qBounds = [[], [1, 2, 3]]
examples = [["foo", 0], ["foo", 1.5], ["foo", 5.5], ["foo", 2.5]]
answers = [["foo", 0], ["foo", 1], ["foo", 3], ["foo", 2]]
nPoss = [0, 4]
composite = Composite.Composite()
composite.SetQuantBounds(qBounds, nPoss)
for i in xrange(len(examples)):
qEx = composite.QuantizeExample(examples[i])
self.assertEqual(qEx, answers[i])
def testQuantize(self):
" testing data quantization "
qBounds = [[], [1, 2, 3]]
examples = [['foo', 0], ['foo', 1.5], ['foo', 5.5], ['foo', 2.5]]
answers = [['foo', 0], ['foo', 1], ['foo', 3], ['foo', 2]]
nPoss = [0, 4]
composite = Composite.Composite()
composite.SetQuantBounds(qBounds, nPoss)
for i in xrange(len(examples)):
qEx = composite.QuantizeExample(examples[i])
self.assertEqual(qEx, answers[i])
def _AddDataToDb(dBase, table, user, password, colDefs, colTypes, data, nullMarker=None,
blockSize=100, cn=None):
""" *For Internal Use*
(drops and) creates a table and then inserts the values
"""
if not cn:
cn = DbModule.connect(dBase, user, password)
c = cn.cursor()
try:
c.execute('drop table %s' % (table))
except Exception:
print('cannot drop table %s' % (table))
try:
sqlStr = 'create table %s (%s)' % (table, colDefs)
c.execute(sqlStr)
except Exception:
print('create table failed: ', sqlStr)
print('here is the exception:')
import traceback
traceback.print_exc()
return
cn.commit()
c = None
block = []
entryTxt = [DbModule.placeHolder] * len(data[0])
dStr = ','.join(entryTxt)
sqlStr = 'insert into %s values (%s)' % (table, dStr)
nDone = 0
for row in data:
entries = [None] * len(row)
for col in xrange(len(row)):
if row[col] is not None and \
(nullMarker is None or row[col] != nullMarker):
if colTypes[col][0] == float:
entries[col] = float(row[col])
elif colTypes[col][0] == int:
entries[col] = int(row[col])
else:
entries[col] = str(row[col])
else:
entries[col] = None
block.append(tuple(entries))
if len(block) >= blockSize:
nDone += _insertBlock(cn, sqlStr, block)
if not hasattr(cn, 'autocommit') or not cn.autocommit:
cn.commit()
block = []
if len(block):
nDone += _insertBlock(cn, sqlStr, block)
if not hasattr(cn, 'autocommit') or not cn.autocommit:
cn.commit()
def _AddDataToDb(dBase,table,user,password,colDefs,colTypes,data,
nullMarker=None,blockSize=100,cn=None):
""" *For Internal Use*
(drops and) creates a table and then inserts the values
"""
if not cn:
cn = DbModule.connect(dBase,user,password)
c = cn.cursor()
try:
c.execute('drop table %s'%(table))
except:
print('cannot drop table %s'%(table))
try:
sqlStr = 'create table %s (%s)'%(table,colDefs)
c.execute(sqlStr)
except:
print('create table failed: ', sqlStr)
print('here is the exception:')
import traceback
traceback.print_exc()
return
cn.commit()
c = None
block = []
entryTxt = [DbModule.placeHolder]*len(data[0])
dStr = ','.join(entryTxt)
sqlStr = 'insert into %s values (%s)'%(table,dStr)
nDone = 0
for row in data:
entries = [None]*len(row)
for col in xrange(len(row)):
if row[col] is not None and \
(nullMarker is None or row[col] != nullMarker):
if colTypes[col][0] == float:
entries[col] = float(row[col])
elif colTypes[col][0] == int:
entries[col] = int(row[col])
else:
entries[col] = str(row[col])
else:
entries[col] = None
block.append(tuple(entries))
if len(block)>=blockSize:
nDone += _insertBlock(cn,sqlStr,block)
if not hasattr(cn,'autocommit') or not cn.autocommit:
cn.commit()
block = []
if len(block):
nDone += _insertBlock(cn,sqlStr,block)
if not hasattr(cn,'autocommit') or not cn.autocommit:
cn.commit()
def testSimpleDescriptorCalc(self):
" testing simple descriptor calculation "
composList = ['Nb','Nb3','NbPt','Nb2Pt']
compare = [[2.32224798203, 0.0, 1.34000003338, 1.34000003338],
[2.32224798203, 0.0, 1.34000003338, 1.34000003338],
[1.51555249095, 0.806695491076, 1.34000003338, 1.29999995232],
[1.78445098797, 0.717062658734, 1.34000003338, 1.29999995232]]
for i in xrange(len(composList)):
assert max(map(lambda x,y:abs(x-y),compare[i],
self.desc.CalcSimpleDescriptorsForComposition(composList[i]))) < self.tol,\
'Descriptor calculation failed'
def test4UnusedVars(self):
" testing unused variables "
self._setupTree1a()
with open(self.qTree1Name, 'r') as inTFile:
buf = inTFile.read().replace('\r\n', '\n').encode('utf-8')
inTFile.close()
with io.BytesIO(buf) as inFile:
t2 = cPickle.load(inFile)
assert self.t1 == t2, 'Incorrect tree generated.'
for i in xrange(len(self.examples1)):
assert self.t1.ClassifyExample(self.examples1[i])==self.examples1[i][-1],\
'examples1[%d] misclassified'%i
def test4UnusedVars(self):
" testing unused variables "
self._setupTree1a()
with open(self.qTree1Name, 'r') as inTFile:
buf = inTFile.read().replace('\r\n', '\n').encode('utf-8')
inTFile.close()
with io.BytesIO(buf) as inFile:
t2 = cPickle.load(inFile)
assert self.t1 == t2, 'Incorrect tree generated.'
for i in xrange(len(self.examples1)):
assert self.t1.ClassifyExample(self.examples1[i])==self.examples1[i][-1],\
'examples1[%d] misclassified'%i
def GetNamedData(self):
""" returns a list of named examples
**Note**
a named example is the result of prepending the example
name to the data list
"""
res = [None] * self.nPts
for i in xrange(self.nPts):
res[i] = [self.ptNames[i]] + self.data[i].tolist()
return res
def ID3Boot(examples,
attrs,
nPossibleVals,
initialVar=None,
depth=0,
maxDepth=-1,
**kwargs):
""" Bootstrapping code for the ID3 algorithm
see ID3 for descriptions of the arguments
If _initialVar_ is not set, the algorithm will automatically
choose the first variable in the tree (the standard greedy
approach). Otherwise, _initialVar_ will be used as the first
split.
"""
totEntropy = CalcTotalEntropy(examples, nPossibleVals)
varTable = GenVarTable(examples, nPossibleVals, attrs)
tree = DecTree.DecTreeNode(None, 'node')
#tree.SetExamples(examples)
tree._nResultCodes = nPossibleVals[-1]
# <perl>you've got to love any language which will let you
# do this much work in a single line :-)</perl>
if initialVar is None:
best = attrs[numpy.argmax([entropy.InfoGain(x) for x in varTable])]
else:
best = initialVar
tree.SetName('Var: %d' % best)
tree.SetData(totEntropy)
tree.SetLabel(best)
tree.SetTerminal(0)
nextAttrs = list(attrs)
if not kwargs.get('recycleVars', 0):
nextAttrs.remove(best)
for val in xrange(nPossibleVals[best]):
nextExamples = []
for example in examples:
if example[best] == val:
nextExamples.append(example)
tree.AddChildNode(
ID3(nextExamples, best, nextAttrs, nPossibleVals, depth, maxDepth,
**kwargs))
return tree
def GetTypeStrings(colHeadings,colTypes,keyCol=None):
""" returns a list of SQL type strings
"""
typeStrs=[]
for i in xrange(len(colTypes)):
typ = colTypes[i]
if typ[0] == float:
typeStrs.append('%s double precision'%colHeadings[i])
elif typ[0] == int:
typeStrs.append('%s integer'%colHeadings[i])
else:
typeStrs.append('%s varchar(%d)'%(colHeadings[i],typ[1]))
if colHeadings[i] == keyCol:
typeStrs[-1] = '%s not null primary key'%(typeStrs[-1])
return typeStrs
def GetTypeStrings(colHeadings, colTypes, keyCol=None):
""" returns a list of SQL type strings
"""
typeStrs = []
for i in xrange(len(colTypes)):
typ = colTypes[i]
if typ[0] == float:
typeStrs.append('%s double precision' % colHeadings[i])
elif typ[0] == int:
typeStrs.append('%s integer' % colHeadings[i])
else:
typeStrs.append('%s varchar(%d)' % (colHeadings[i], typ[1]))
if colHeadings[i] == keyCol:
typeStrs[-1] = '%s not null primary key' % (typeStrs[-1])
return typeStrs
def GenRandomExamples(nVars=10,randScale=0.3,bitProb=0.5,nExamples=500,seed=(0,0),
addResults=1):
random.seed(seed[0])
varWeights = numpy.array([random.random() for x in range(nVars)])*randScale
examples = [None]*nExamples
for i in xrange(nExamples):
varVals=[random.random()>bitProb for x in range(nVars)]
temp = numpy.array(varVals) * varWeights
res = sum(temp)
if addResults:
varVals.append(res>=1.)
examples[i] = varVals
nPossibleVals = [2]*(nExamples+1)
attrs = list(range(nVars))
return (examples,attrs,nPossibleVals)
def ReadGeneralExamples(inFile):
""" reads the examples from a .dat file
**Arguments**
- inFile: a file object
**Returns**
a 2-tuple containing:
1) the names of the examples
2) a list of lists containing the examples themselves
**Note**
- this attempts to convert variable values to ints, then floats.
if those both fail, they are left as strings
"""
expr1 = re.compile(r'^#')
expr2 = re.compile(r'[\ ]*|[\t]*')
examples = []
names = []
inLine = inFile.readline()
while inLine:
if expr1.search(inLine) is None:
resArr = expr2.split(inLine)[:-1]
if len(resArr) > 1:
for i in xrange(1, len(resArr)):
d = resArr[i]
try:
resArr[i] = int(d)
except ValueError:
try:
resArr[i] = float(d)
except ValueError:
pass
examples.append(resArr[1:])
names.append(resArr[0])
inLine = inFile.readline()
return names, examples
def ID3Boot(examples,attrs,nPossibleVals,initialVar=None,depth=0,maxDepth=-1,
**kwargs):
""" Bootstrapping code for the ID3 algorithm
see ID3 for descriptions of the arguments
If _initialVar_ is not set, the algorithm will automatically
choose the first variable in the tree (the standard greedy
approach). Otherwise, _initialVar_ will be used as the first
split.
"""
totEntropy = CalcTotalEntropy(examples,nPossibleVals)
varTable = GenVarTable(examples,nPossibleVals,attrs)
tree=DecTree.DecTreeNode(None,'node')
#tree.SetExamples(examples)
tree._nResultCodes = nPossibleVals[-1]
# <perl>you've got to love any language which will let you
# do this much work in a single line :-)</perl>
if initialVar is None:
best = attrs[numpy.argmax([entropy.InfoGain(x) for x in varTable])]
else:
best = initialVar
tree.SetName('Var: %d'%best)
tree.SetData(totEntropy)
tree.SetLabel(best)
tree.SetTerminal(0)
nextAttrs = list(attrs)
if not kwargs.get('recycleVars',0):
nextAttrs.remove(best)
for val in xrange(nPossibleVals[best]):
nextExamples = []
for example in examples:
if example[best] == val:
nextExamples.append(example)
tree.AddChildNode(ID3(nextExamples,best,nextAttrs,nPossibleVals,depth,maxDepth,
**kwargs))
return tree
def ReadGeneralExamples(inFile):
""" reads the examples from a .dat file
**Arguments**
- inFile: a file object
**Returns**
a 2-tuple containing:
1) the names of the examples
2) a list of lists containing the examples themselves
**Note**
- this attempts to convert variable values to ints, then floats.
if those both fail, they are left as strings
"""
expr1 = re.compile(r'^#')
expr2 = re.compile(r'[\ ]*|[\t]*')
examples = []
names = []
inLine = inFile.readline()
while inLine:
if expr1.search(inLine) is None:
resArr = expr2.split(inLine)[:-1]
if len(resArr)>1:
for i in xrange(1,len(resArr)):
d = resArr[i]
try:
resArr[i] = int(d)
except ValueError:
try:
resArr[i] = float(d)
except ValueError:
pass
examples.append(resArr[1:])
names.append(resArr[0])
inLine = inFile.readline()
return names,examples
def FullyConnectNodes(self):
""" Fully connects each layer in the network to the one above it
**Note**
this sets the connections, but does not assign weights
"""
nodeList = range(self.numInputNodes)
nConnections = 0
for layer in xrange(self.numHiddenLayers):
for i in self.layerIndices[layer+1]:
self.nodeList[i].SetInputs(nodeList)
nConnections = nConnections + len(nodeList)
nodeList = self.layerIndices[layer+1]
for i in self.layerIndices[-1]:
self.nodeList[i].SetInputs(nodeList)
nConnections = nConnections + len(nodeList)
self.nConnections = nConnections
def FullyConnectNodes(self):
""" Fully connects each layer in the network to the one above it
**Note**
this sets the connections, but does not assign weights
"""
nodeList = range(self.numInputNodes)
nConnections = 0
for layer in xrange(self.numHiddenLayers):
for i in self.layerIndices[layer + 1]:
self.nodeList[i].SetInputs(nodeList)
nConnections = nConnections + len(nodeList)
nodeList = self.layerIndices[layer + 1]
for i in self.layerIndices[-1]:
self.nodeList[i].SetInputs(nodeList)
nConnections = nConnections + len(nodeList)
self.nConnections = nConnections
def CalcSimpleDescriptorsForComposition(self, compos='', composList=None):
""" calculates all simple descriptors for a given composition
**Arguments**
- compos: a string representation of the composition
- composList: a *composVect*
The client must provide either _compos_ or _composList_. If both are
provided, _composList_ takes priority.
**Returns**
the list of descriptor values
**Notes**
- when _compos_ is provided, this uses _chemutils.SplitComposition_
to split the composition into its individual pieces
- if problems are encountered because of either an unknown descriptor or
atom type, a _KeyError_ will be raised.
"""
if composList is None:
composList = chemutils.SplitComposition(compos)
try:
res = []
for i in xrange(len(self.simpleList)):
descName, targets = self.simpleList[i]
for target in targets:
try:
method = getattr(self, target)
except AttributeError:
print('Method %s does not exist' % (target))
else:
res.append(method(descName, composList))
except KeyError as msg:
print('composition %s caused problems' % composList)
raise KeyError(msg)
return res
def _AdjustColHeadings(colHeadings, maxColLabelLen):
""" *For Internal Use*
removes illegal characters from column headings
and truncates those which are too long.
"""
for i in xrange(len(colHeadings)):
# replace unallowed characters and strip extra white space
colHeadings[i] = colHeadings[i].strip()
colHeadings[i] = colHeadings[i].replace(' ', '_')
colHeadings[i] = colHeadings[i].replace('-', '_')
colHeadings[i] = colHeadings[i].replace('.', '_')
if len(colHeadings[i]) > maxColLabelLen:
# interbase (at least) has a limit on the maximum length of a column name
newHead = colHeadings[i].replace('_', '')
newHead = newHead[:maxColLabelLen]
print('\tHeading %s too long, changed to %s' % (colHeadings[i], newHead))
colHeadings[i] = newHead
return colHeadings
def GetDescriptorNames(self):
""" returns a list of the names of the descriptors this calculator generates
"""
if self.descriptorNames is not None:
return self.descriptorNames
else:
res = []
for i in xrange(len(self.simpleList)):
descName, targets = self.simpleList[i]
for target in targets:
try:
method = getattr(self, target)
except AttributeError:
print('Method %s does not exist' % (target))
else:
res.append('%s_%s' % (target, descName))
for entry in self.compoundList:
res.append(entry[0])
self.descriptorNames = res[:]
return tuple(res)
