rdkit.ML.InfoTheory.rdInfoTheory module

Module containing bunch of functions for information metrics and a ranker to rank bits

class rdkit.ML.InfoTheory.rdInfoTheory.BitCorrMatGenerator(self)

Bases: object

A class to generate a pairwise correlation matrix between a list of bits The mode of operation for this class is something like this

>>> cmg = BitCorrMatGenerator()
>>> cmg.SetBitList(blist)
>>> for fp in fpList:
>>>    cmg.CollectVotes(fp)
>>> corrMat = cmg.GetCorrMatrix()

The resulting correlation matrix is a one dimensional nummeric array containing the lower triangle elements

CollectVotes(self, bitVect: object) → None

For each pair of on bits (bi, bj) in fp increase the correlation count for the pair by 1

Parameters:

fp (-) – a bit vector to collect the fingerprints from

GetCorrMatrix(self) → numpy.ndarray[dtype=float64, shape=(*)]

Get the correlation matrix following the collection of votes from a bunch of fingerprints

SetBitList(self, bitList: collections.abc.Iterable) → None

Set the list of bits that need to be correllated

This may for example be their top ranking ensemble bits

Parameters:

bitList (-) – an integer list of bit IDs

class rdkit.ML.InfoTheory.rdInfoTheory.InfoBitRanker(self, nBits: int, nClasses: int)
class rdkit.ML.InfoTheory.rdInfoTheory.InfoBitRanker(self, nBits: int, nClasses: int, infoType: rdkit.ML.InfoTheory.rdInfoTheory.InfoType)

Bases: object

A class to rank the bits from a series of labelled fingerprints A simple demonstration may help clarify what this class does. Here’s a small set of vectors:

>>> for i,bv in enumerate(bvs): print(bv.ToBitString(),acts[i])
...
0001 0
0101 0
0010 1
1110 1

Default ranker, using infogain:

>>> ranker = InfoBitRanker(4,2)
>>> for i,bv in enumerate(bvs): ranker.AccumulateVotes(bv,acts[i])
...
>>> for bit,gain,n0,n1 in ranker.GetTopN(3): print(int(bit),'%.3f'%gain,int(n0),int(n1))
...
3 1.000 2 0
2 1.000 0 2
0 0.311 0 1

Using the biased infogain:

>>> ranker = InfoBitRanker(4,2,InfoTheory.InfoType.BIASENTROPY)
>>> ranker.SetBiasList((1,))
>>> for i,bv in enumerate(bvs): ranker.AccumulateVotes(bv,acts[i])
...
>>> for bit,gain,n0,n1 in ranker.GetTopN(3): print(int(bit),'%.3f'%gain,int(n0),int(n1))
...
2 1.000 0 2
0 0.311 0 1
1 0.000 1 1

A chi squared ranker is also available:

>>> ranker = InfoBitRanker(4,2,InfoTheory.InfoType.CHISQUARE)
>>> for i,bv in enumerate(bvs): ranker.AccumulateVotes(bv,acts[i])
...
>>> for bit,gain,n0,n1 in ranker.GetTopN(3): print(int(bit),'%.3f'%gain,int(n0),int(n1))
...
3 4.000 2 0
2 4.000 0 2
0 1.333 0 1

As is a biased chi squared:

>>> ranker = InfoBitRanker(4,2,InfoTheory.InfoType.BIASCHISQUARE)
>>> ranker.SetBiasList((1,))
>>> for i,bv in enumerate(bvs): ranker.AccumulateVotes(bv,acts[i])
...
>>> for bit,gain,n0,n1 in ranker.GetTopN(3): print(int(bit),'%.3f'%gain,int(n0),int(n1))
...
2 4.000 0 2
0 1.333 0 1
1 0.000 1 1
AccumulateVotes(self, bitVect: object, label: int) → None

Accumulate the votes for all the bits turned on in a bit vector

Parameters:

  • bv (-) – bit vector either ExplicitBitVect or SparseBitVect operator

  • label (-) – the class label for the bit vector. It is assumed that 0 <= class < nClasses

GetTopN(self, num: int) → numpy.ndarray[dtype=float64, shape=(*, *)]

Returns the top n bits ranked by the information metric This is actually the function where most of the work of ranking is happening

Parameters:

num (-) – the number of top ranked bits that are required

SetBiasList(self, classList: collections.abc.Iterable) → None

Set the classes to which the entropy calculation should be biased

This list contains a set of class ids used when in the BIASENTROPY mode of ranking bits. In this mode, a bit must be correlated higher with one of the biased classes than all the other classes. For example, in a two class problem with actives and inactives, the fraction of actives that hit the bit has to be greater than the fraction of inactives that hit the bit

Parameters:

classList (-) – list of class ids that we want a bias towards

SetMaskBits(self, maskBits: collections.abc.Iterable) → None

Set the mask bits for the calculation

Parameters:

maskBits (-) – list of mask bits to use

Tester(self, bitVect: object) → None
WriteTopBitsToFile(self, fileName: str) → None

Write the bits that have been ranked to a file

class rdkit.ML.InfoTheory.rdInfoTheory.InfoType(*values)

Bases: Enum

BIASCHISQUARE = 4
BIASENTROPY = 2
CHISQUARE = 3
ENTROPY = 1