rdkit/Python/ML/CompositeRun.py

# $Id$
#
#  Copyright (C) 2002-2006  greg Landrum and Rational Discovery LLC
#
#   @@ All Rights Reserved  @@
#
""" contains a class to store parameters for and results from
Composite building

"""
import RDConfig
from Dbase.DbConnection import DbConnect
import DataStructs
from Dbase import DbModule
##from ML.SVM import SVMClassificationModel as SVM

def SetDefaults(runDetails):
  """  initializes a details object with default values

      **Arguments**

        - details:  (optional) a _CompositeRun.CompositeRun_ object.
          If this is not provided, the global _runDetails will be used.

      **Returns**

        the initialized _CompositeRun_ object.


  """
  runDetails.nRuns = 1
  runDetails.nModels = 10
  runDetails.outName = ''
  runDetails.badName = ''
  runDetails.splitRun=0
  runDetails.splitFrac=0.7
  runDetails.lockRandom = 0
  runDetails.randomActivities = 0
  runDetails.shuffleActivities = 0
  runDetails.replacementSelection = 0

  #
  # Tree Parameters
  #
  runDetails.useTrees = 1
  runDetails.pruneIt=0
  runDetails.lessGreedy=0 
  runDetails.limitDepth=-1
  runDetails.recycleVars=0
  runDetails.randomDescriptors=0  # toggles growing of random forests
  
  #
  # KNN Parameters
  #
  runDetails.useKNN = 0
  runDetails.knnDistFunc = ''
  runDetails.knnNeighs = 0

  #
  # SigTree Parameters
  #
  runDetails.useSigTrees =0
  runDetails.useCMIM=0
  runDetails.allowCollections=False

  #
  # Naive Bayes Classifier Parameters
  #
  runDetails.useNaiveBayes = 0
  runDetails.mEstimateVal=-1.0
  runDetails.useSigBayes =0
  
##   #
##   # SVM Parameters
##   #
##   runDetails.useSVM = 0
##   runDetails.svmKernel = SVM.radialKernel
##   runDetails.svmType = SVM.cSVCType
##   runDetails.svmGamma = None
##   runDetails.svmCost = None
##   runDetails.svmWeights = None
##   runDetails.svmDataType = 'float'
##   runDetails.svmDegree = 3
##   runDetails.svmCoeff = 0.0
##   runDetails.svmEps = 0.001
##   runDetails.svmNu = 0.5
##   runDetails.svmCache = 40
##   runDetails.svmShrink = 1
##   runDetails.svmDataType='float'

  
  runDetails.bayesModel = 0
  runDetails.dbName = ''
  runDetails.dbUser = RDConfig.defaultDBUser
  runDetails.dbPassword = RDConfig.defaultDBPassword
  runDetails.dbWhat='*'
  runDetails.dbWhere=''
  runDetails.dbJoin=''
  runDetails.qTableName = ''
  runDetails.qBounds = []
  runDetails.qBoundCount = ''
  runDetails.activityBounds = []
  runDetails.activityBoundsVals = ''
  runDetails.detailedRes = 0
  runDetails.noScreen = 0
  runDetails.threshold = 0.0
  runDetails.filterFrac = 0.0
  runDetails.filterVal = 0.0
  runDetails.modelFilterVal = 0.0
  runDetails.modelFilterFrac = 0.0
  runDetails.internalHoldoutFrac = 0.3
  runDetails.pickleDataFileName=''
  runDetails.startAt=None
  runDetails.persistTblName=''
  runDetails.randomSeed=(23,42)
  runDetails.note=''

  
  return runDetails


class CompositeRun:
  """ class to store parameters for and results from Composite building

   This class has a default set of fields which are added to the database.

   By default these fields are stored in a tuple, so they are immutable.  This
     is probably what you want.
   

  """
  fields = (\
    ("rundate","varchar(32)"),
    ("dbName","varchar(200)"),
    ("dbWhat","varchar(200)"),
    ("dbWhere","varchar(200)"),
    ("dbJoin","varchar(200)"),
    ("tableName","varchar(80)"),
    ("note","varchar(120)"),
    ("shuffled","smallint"),
    ("randomized","smallint"),
    ("overall_error","float"),
    ("holdout_error","float"),
    ("overall_fraction_dropped","float"),
    ("holdout_fraction_dropped","float"),
    ("overall_correct_conf","float"),
    ("overall_incorrect_conf","float"),
    ("holdout_correct_conf","float"),
    ("holdout_incorrect_conf","float"),
    ("overall_result_matrix","varchar(256)"),
    ("holdout_result_matrix","varchar(256)"),
    ("threshold","float"),
    ("splitFrac","float"),
    ("filterFrac","float"),
    ("filterVal","float"),
    ("modelFilterVal", "float"),
    ("modelFilterFrac", "float"),
    ("nModels","int"),
    ("limitDepth","int"),
    ("bayesModels","int"),
    ("qBoundCount","varchar(3000)"),
    ("activityBoundsVals","varchar(200)"),
    ("cmd","varchar(500)"),
    ("model",DbModule.binaryTypeName),
    )

  def _CreateTable(self,cn,tblName):
    """ *Internal Use only*

    """
    names = map(lambda x:x.strip().upper(),cn.GetTableNames())
    if tblName.upper() not in names:
      curs = cn.GetCursor()
      fmt = []
      for name,value in self.fields:
        fmt.append('%s %s'%(name,value))
      fmtStr = ','.join(fmt)  
      curs.execute('create table %s (%s)'%(tblName,fmtStr))
      cn.Commit()
    else:
      heads = [x.upper() for x in cn.GetColumnNames()]
      curs = cn.GetCursor()
      for name,value in self.fields:
        if name.upper() not in heads:
          curs.execute('alter table %s add %s %s'%(tblName,name,value))
      cn.Commit()    
  def Store(self,db='models.gdb',table='results',
            user='sysdba',password='masterkey'):
    """ adds the result to a database

      **Arguments**

        - db: name of the database to use

        - table: name of the table to use

        - user&password: connection information

    """
    cn = DbConnect(db,table,user,password)
    curs = cn.GetCursor()
    self._CreateTable(cn,table)

    cols = []
    vals = []
    for name,typ in self.fields:
      try:
        v = getattr(self,name)
      except AttributeError:
        pass
      else:
        cols.append('%s'%name)
        vals.append(v)

    nToDo = len(vals)
    qs = ','.join([DbModule.placeHolder]*nToDo)
    vals = tuple(vals)
    
    cmd = 'insert into %s (%s) values (%s)'%(table,
                                             ','.join(cols),
                                             qs)
    curs.execute(cmd,vals)
    cn.Commit()

  def GetDataSet(self,**kwargs):
    """ Returns a MLDataSet pulled from a database using our stored
    values.

    """
    from ML.Data import DataUtils
    data = DataUtils.DBToData(self.dbName,self.tableName,
                              user=self.dbUser,password=self.dbPassword,
                              what=self.dbWhat,where=self.dbWhere,
                              join=self.dbJoin,**kwargs)

    return data
    
    
  def GetDataSetInfo(self,**kwargs):
    """ Returns a MLDataSet pulled from a database using our stored
    values.

    """
    from Dbase.DbConnection import DbConnect
    conn = DbConnect(self.dbName,self.tableName)
    res = conn.GetColumnNamesAndTypes(join=self.dbJoin,what=self.dbWhat,where=self.dbWhere)
    return res