def test2XValClass(self):
fName = os.path.join(RDConfig.RDCodeDir, 'ML', 'KNN', 'test_data',
'random_pts.csv')
data = DataUtils.TextFileToData(fName)
examples = data.GetNamedData()
npvals = data.GetNPossibleVals()
nvars = data.GetNVars()
attrs = list(range(1, nvars + 1))
numNeigh = 11
mod, err = CrossValidate.CrossValidationDriver(examples,
attrs,
npvals,
numNeigh,
silent=1)
self.assertAlmostEqual(err, 0.01075, 4)
neighborList = []
res = mod.ClassifyExample(examples[0], neighborList=neighborList)
self.assertEqual(res, 1)
self.assertEqual(neighborList[0][1], examples[0])
self.assertEqual(mod.GetName(), '')
mod.SetName('name')
self.assertEqual(mod.GetName(), 'name')
self.assertEqual(mod.type(), 'Classification Model')
mod.NameModel('this argument is ignored')
self.assertEqual(mod.GetName(), 'Classification Model')
def test2NaiveBayes(self):
fName = os.path.join(RDConfig.RDCodeDir, 'ML', 'NaiveBayes',
'test_data', 'stddata.csv')
data = DataUtils.TextFileToData(fName)
examples = data.GetNamedData()
nvars = data.GetNVars()
attrs = list(range(1, nvars + 1))
npvals = [0] + [3] * nvars + [2]
qBounds = [0] + [2] * nvars + [0]
mod, err = CrossValidate.CrossValidationDriver(examples,
attrs,
npvals,
qBounds,
mEstimateVal=20.0,
silent=True)
self.assertTrue(isinstance(mod, NaiveBayesClassifier))
self.assertAlmostEqual(err, 0.1818, 4)
self.assertEqual(mod.GetName(), '')
mod.SetName('modelName')
self.assertEqual(mod.GetName(), 'modelName')
mod.NameModel(None)
self.assertEqual(mod.GetName(), 'NaiveBayesClassifier')
self.assertGreater(len(mod.GetExamples()), 0)
self.assertGreater(len(mod.GetTrainingExamples()), 0)
self.assertEqual(sorted(mod.GetTrainingExamples() + mod.GetExamples()),
sorted(examples))
def test1NaiveBayes(self):
fName = os.path.join(RDConfig.RDCodeDir, 'ML', 'NaiveBayes',
'test_data', 'stddata.csv')
data = DataUtils.TextFileToData(fName)
examples = data.GetNamedData()
nvars = data.GetNVars()
attrs = range(1, nvars + 1)
npvals = [0] + [3] * nvars + [2]
qBounds = [0] + [2] * nvars + [0]
mod, err = CrossValidate.CrossValidationDriver(examples,
attrs,
npvals,
qBounds,
silent=True)
self.assertAlmostEqual(mod._classProbs[0], 0.5000, 4)
self.assertAlmostEqual(mod._classProbs[1], 0.5000, 4)
self.assertAlmostEqual(mod._QBoundVals[1][0], -0.0360, 4)
self.assertAlmostEqual(mod._QBoundVals[1][1], 0.114)
self.assertAlmostEqual(mod._QBoundVals[2][0], -0.7022, 4)
self.assertAlmostEqual(mod._QBoundVals[2][1], -0.16635, 4)
self.assertAlmostEqual(mod._QBoundVals[3][0], -0.3659, 4)
self.assertAlmostEqual(mod._QBoundVals[3][1], 0.4305, 4)
self.assertAlmostEqual(err, 0.2121, 4)
def test1NaiveBayes(self):
fName = os.path.join(RDConfig.RDCodeDir, 'ML', 'NaiveBayes',
'test_data', 'stddata.csv')
data = DataUtils.TextFileToData(fName)
examples = data.GetNamedData()
nvars = data.GetNVars()
attrs = list(range(1, nvars + 1))
npvals = [0] + [3] * nvars + [2]
qBounds = [0] + [2] * nvars + [0]
mod, err = CrossValidate.CrossValidationDriver(examples,
attrs,
npvals,
qBounds,
silent=True)
self.assertAlmostEqual(mod._classProbs[0], 0.5000, 4)
self.assertAlmostEqual(mod._classProbs[1], 0.5000, 4)
self.assertAlmostEqual(mod._QBoundVals[1][0], -0.0360, 4)
self.assertAlmostEqual(mod._QBoundVals[1][1], 0.114)
self.assertAlmostEqual(mod._QBoundVals[2][0], -0.7022, 4)
self.assertAlmostEqual(mod._QBoundVals[2][1], -0.16635, 4)
self.assertAlmostEqual(mod._QBoundVals[3][0], -0.3659, 4)
self.assertAlmostEqual(mod._QBoundVals[3][1], 0.4305, 4)
self.assertAlmostEqual(err, 0.2121, 4)
mod, err = CrossValidate.CrossValidationDriver(examples,
attrs,
npvals,
qBounds,
silent=True,
calcTotalError=True)
self.assertAlmostEqual(mod._classProbs[0], 0.515151, 4)
self.assertAlmostEqual(mod._classProbs[1], 0.484848, 4)
self.assertAlmostEqual(mod._QBoundVals[1][0], -0.40315, 4)
self.assertAlmostEqual(mod._QBoundVals[1][1], 0.114)
self.assertAlmostEqual(mod._QBoundVals[2][0], -0.62185, 4)
self.assertAlmostEqual(mod._QBoundVals[2][1], -0.19965, 4)
self.assertAlmostEqual(mod._QBoundVals[3][0], 0.4305, 4)
self.assertAlmostEqual(mod._QBoundVals[3][1], 0.80305, 4)
self.assertAlmostEqual(err, 0.14563, 4)
mod, err = CrossValidate.CrossValidationDriver(
examples,
attrs,
npvals,
qBounds,
silent=True,
replacementSelection=True)
self.assertAlmostEqual(mod._classProbs[0], 0.5131578, 4)
self.assertAlmostEqual(mod._classProbs[1], 0.4868421, 4)
self.assertAlmostEqual(mod._QBoundVals[1][0], -0.036, 4)
self.assertAlmostEqual(mod._QBoundVals[1][1], 0.93465, 4)
self.assertAlmostEqual(mod._QBoundVals[2][0], -0.6696, 4)
self.assertAlmostEqual(mod._QBoundVals[2][1], -0.19965, 4)
self.assertAlmostEqual(mod._QBoundVals[3][0], -1.06785, 4)
self.assertAlmostEqual(mod._QBoundVals[3][1], 0.4305, 4)
self.assertAlmostEqual(err, 0.3, 4)
def RunIt(details, progressCallback=None, saveIt=1, setDescNames=0):
""" does the actual work of building a composite model
**Arguments**
- details: a _CompositeRun.CompositeRun_ object containing details
(options, parameters, etc.) about the run
- progressCallback: (optional) a function which is called with a single
argument (the number of models built so far) after each model is built.
- saveIt: (optional) if this is nonzero, the resulting model will be pickled
and dumped to the filename specified in _details.outName_
- setDescNames: (optional) if nonzero, the composite's _SetInputOrder()_ method
will be called using the results of the data set's _GetVarNames()_ method;
it is assumed that the details object has a _descNames attribute which
is passed to the composites _SetDescriptorNames()_ method. Otherwise
(the default), _SetDescriptorNames()_ gets the results of _GetVarNames()_.
**Returns**
the composite model constructed
"""
details.rundate = time.asctime()
fName = details.tableName.strip()
if details.outName == '':
details.outName = fName + '.pkl'
if not details.dbName:
if details.qBounds != []:
data = DataUtils.TextFileToData(fName)
else:
data = DataUtils.BuildQuantDataSet(fName)
elif details.useSigTrees or details.useSigBayes:
details.tableName = fName
data = details.GetDataSet(pickleCol=0,
pickleClass=DataStructs.ExplicitBitVect)
elif details.qBounds != [] or not details.useTrees:
details.tableName = fName
data = details.GetDataSet()
else:
data = DataUtils.DBToQuantData(
details.dbName, # Function no longer defined
fName,
quantName=details.qTableName,
user=details.dbUser,
password=details.dbPassword)
composite = RunOnData(details,
data,
progressCallback=progressCallback,
saveIt=saveIt,
setDescNames=setDescNames)
return composite
def test2XValClass(self): fName = os.path.join(RDConfig.RDCodeDir,'ML','KNN','test_data','random_pts.csv') data = DataUtils.TextFileToData(fName) examples = data.GetNamedData() npvals = data.GetNPossibleVals() nvars = data.GetNVars() attrs = range(1,nvars+1) numNeigh = 11 mod, err = CrossValidate.CrossValidationDriver(examples, attrs, npvals, numNeigh,silent=1) self.assertAlmostEqual(err,0.01075,4)
def test4XValRegress(self):
fName = os.path.join(RDConfig.RDCodeDir,'ML','KNN','test_data','random_pts.csv')
data = DataUtils.TextFileToData(fName)
examples = data.GetNamedData()
npvals = data.GetNPossibleVals()
nvars = data.GetNVars()
attrs = range(1,nvars+1)
numNeigh = 11
mod, err = CrossValidate.CrossValidationDriver(examples, attrs, npvals, numNeigh,silent=1,
modelBuilder=CrossValidate.makeRegressionModel)
# NOTE: this number hasn't been extensively checked
self.assertAlmostEqual(err,0.0777,4)
def test2NaiveBayes(self):
fName = os.path.join(RDConfig.RDCodeDir, 'ML', 'NaiveBayes',
'test_data', 'stddata.csv')
data = DataUtils.TextFileToData(fName)
examples = data.GetNamedData()
nvars = data.GetNVars()
attrs = range(1, nvars + 1)
npvals = [0] + [3] * nvars + [2]
qBounds = [0] + [2] * nvars + [0]
mod, err = CrossValidate.CrossValidationDriver(examples,
attrs,
npvals,
qBounds,
mEstimateVal=20.0)
assert feq(err, 0.19354)
def test1Neighbors(self):
fName = os.path.join(RDConfig.RDCodeDir,'ML','KNN','test_data','random_pts.csv')
data = DataUtils.TextFileToData(fName)
examples = data.GetNamedData()
npvals = data.GetNPossibleVals()
nvars = data.GetNVars()
attrs = range(1,nvars+1)
numNeigh = 11
metric = DistFunctions.EuclideanDist
mdl = KNNModel.KNNModel(numNeigh,attrs,metric)
pt = examples.pop(0)
tgt = [(metric(pt,ex,attrs),ex) for ex in examples]
tgt.sort()
mdl.SetTrainingExamples(examples)
neighbors = mdl.GetNeighbors(pt)
for i in range(numNeigh):
assert feq(-tgt[i][0],neighbors[i][0])
assert tgt[i][1][0]==neighbors[i][1][0]
def test3Regress(self):
# """ a carefully laid out regression data set where the results are clear: """
fName = os.path.join(RDConfig.RDCodeDir, 'ML', 'KNN', 'test_data',
'sample_pts.csv')
data = DataUtils.TextFileToData(fName)
examples = data.GetNamedData()
nvars = data.GetNVars()
attrs = list(range(1, nvars + 1))
numNeigh = 4
metric = DistFunctions.EuclideanDist
mdl = KNNRegressionModel.KNNRegressionModel(numNeigh, attrs, metric)
mdl.SetTrainingExamples(examples)
res = mdl.PredictExample([4, -3.5, 2.5, 0])
assert feq(res, 1.25)
res = mdl.PredictExample([4, 3, 2, 0])
assert feq(res, 1.5)
res = mdl.PredictExample([4, 3, -2.5, 0])
assert feq(res, -1.5)
# Use a distance dependent weight for the neighbours
res = mdl.PredictExample([4, 3, -2.5, 0], weightedAverage=True)
self.assertAlmostEqual(res, -1.6)
# Check the case that the example is identical to one of the neighbours (distance = 0)
neighborList = []
res = mdl.PredictExample(examples[0],
weightedAverage=True,
neighborList=neighborList)
self.assertAlmostEqual(res, 1.5857864)
self.assertEqual(neighborList[0][1], examples[0])
self.assertEqual(mdl.GetBadExamples(), [])
self.assertEqual(mdl.GetName(), '')
mdl.SetName('name')
self.assertEqual(mdl.GetName(), 'name')
self.assertEqual(mdl.type(), 'Regression Model')
mdl.NameModel('this argument is ignored')
self.assertEqual(mdl.GetName(), 'Regression Model')
self.assertEqual(
sorted(mdl.GetTrainingExamples() + mdl.GetTestExamples()),
sorted(examples))
def test3Regress(self):
""" a carefully laid out regression data set where the results are clear:
"""
fName = os.path.join(RDConfig.RDCodeDir,'ML','KNN','test_data','sample_pts.csv')
data = DataUtils.TextFileToData(fName)
examples = data.GetNamedData()
npvals = data.GetNPossibleVals()
nvars = data.GetNVars()
attrs = range(1,nvars+1)
numNeigh = 4
metric = DistFunctions.EuclideanDist
mdl = KNNRegressionModel.KNNRegressionModel(numNeigh,attrs,metric)
mdl.SetTrainingExamples(examples)
res = mdl.PredictExample([4,-3.5,2.5,0])
assert feq(res,1.25)
res = mdl.PredictExample([4,3,2,0])
assert feq(res,1.5)
res = mdl.PredictExample([4,3,-2.5,0])
assert feq(res,-1.5)
def test1NaiveBayes(self):
fName = os.path.join(RDConfig.RDCodeDir, 'ML', 'NaiveBayes',
'test_data', 'stddata.csv')
data = DataUtils.TextFileToData(fName)
examples = data.GetNamedData()
nvars = data.GetNVars()
attrs = range(1, nvars + 1)
npvals = [0] + [3] * nvars + [2]
qBounds = [0] + [2] * nvars + [0]
mod, err = CrossValidate.CrossValidationDriver(examples, attrs, npvals,
qBounds)
assert feq(mod._classProbs[0], 0.54167)
assert feq(mod._classProbs[1], 0.45833)
assert feq(mod._QBoundVals[1][0], -0.56995)
assert feq(mod._QBoundVals[1][1], 0.114)
assert feq(mod._QBoundVals[2][0], -0.7022)
assert feq(mod._QBoundVals[2][1], -0.2347)
assert feq(mod._QBoundVals[3][0], -0.3659)
assert feq(mod._QBoundVals[3][1], 1.17275)
assert feq(err, 0.16129)
def FingerprintsFromDetails(details, reportFreq=10):
data = None
if details.dbName and details.tableName:
from rdkit.Dbase.DbConnection import DbConnect
from rdkit.Dbase import DbInfo
from rdkit.ML.Data import DataUtils
try:
conn = DbConnect(details.dbName, details.tableName)
except Exception:
import traceback
error('Problems establishing connection to database: %s|%s\n' %
(details.dbName, details.tableName))
traceback.print_exc()
if not details.idName:
details.idName = DbInfo.GetColumnNames(details.dbName,
details.tableName)[0]
dataSet = DataUtils.DBToData(details.dbName,
details.tableName,
what='%s,%s' %
(details.idName, details.smilesName))
idCol = 0
smiCol = 1
elif details.inFileName and details.useSmiles:
from rdkit.ML.Data import DataUtils
conn = None
if not details.idName:
details.idName = 'ID'
try:
dataSet = DataUtils.TextFileToData(
details.inFileName,
onlyCols=[details.idName, details.smilesName])
except IOError:
import traceback
error('Problems reading from file %s\n' % (details.inFileName))
traceback.print_exc()
idCol = 0
smiCol = 1
elif details.inFileName and details.useSD:
conn = None
dataset = None
if not details.idName:
details.idName = 'ID'
dataSet = []
try:
s = Chem.SDMolSupplier(details.inFileName)
except Exception:
import traceback
error('Problems reading from file %s\n' % (details.inFileName))
traceback.print_exc()
else:
while 1:
try:
m = s.next()
except StopIteration:
break
if m:
dataSet.append(m)
if reportFreq > 0 and not len(dataSet) % reportFreq:
message('Read %d molecules\n' % (len(dataSet)))
if details.maxMols > 0 and len(
dataSet) >= details.maxMols:
break
for i, mol in enumerate(dataSet):
if mol.HasProp(details.idName):
nm = mol.GetProp(details.idName)
else:
nm = mol.GetProp('_Name')
dataSet[i] = (nm, mol)
else:
dataSet = None
fps = None
if dataSet and not details.useSD:
data = dataSet.GetNamedData()
if not details.molPklName:
fps = FingerprintsFromSmiles(data, idCol, smiCol,
**details.__dict__)
else:
fps = FingerprintsFromPickles(data, idCol, smiCol,
**details.__dict__)
elif dataSet and details.useSD:
fps = FingerprintsFromMols(dataSet, **details.__dict__)
if fps:
if details.outFileName:
outF = open(details.outFileName, 'wb+')
for i in range(len(fps)):
pickle.dump(fps[i], outF)
outF.close()
dbName = details.outDbName or details.dbName
if details.outTableName and dbName:
from rdkit.Dbase.DbConnection import DbConnect
from rdkit.Dbase import DbUtils, DbModule
conn = DbConnect(dbName)
#
# We don't have a db open already, so we'll need to figure out
# the types of our columns...
#
colTypes = DbUtils.TypeFinder(data, len(data), len(data[0]))
typeStrs = DbUtils.GetTypeStrings(
[details.idName, details.smilesName],
colTypes,
keyCol=details.idName)
cols = '%s, %s %s' % (typeStrs[0], details.fpColName,
DbModule.binaryTypeName)
# FIX: we should really check to see if the table
# is already there and, if so, add the appropriate
# column.
#
# create the new table
#
if details.replaceTable or \
details.outTableName.upper() not in [x.upper() for x in conn.GetTableNames()]:
conn.AddTable(details.outTableName, cols)
#
# And add the data
#
for ID, fp in fps:
tpl = ID, DbModule.binaryHolder(fp.ToBinary())
conn.InsertData(details.outTableName, tpl)
conn.Commit()
return fps
