NLP.SwiftLDA

Latent Dirichlet Allocation

Imperative implementation of a collapsed Gibbs sampler for LDA. This library uses the topic modeling terminology (documents, words, topics), even though it is generic. For example if used for word class induction, replace documents with word types, words with features and topics with word classes.

Samplers

pass :: Int -> LDA s -> Vector Doc -> ST s (Vector Doc)

pass batch runs one pass of Gibbs sampling on documents in batch

passOne :: Int -> LDA s -> Doc -> ST s Doc

Run a pass on a single doc

Datatypes

data LDA s

Abstract type holding the settings and the state of the sampler

type Doc = (D, Vector (W, Maybe Z))

type D = Int

type W = Int

type Z = Int

type Table2D = IntMap Table1D

type Table1D = IntMap Double

Access model information

data Finalized

Constructors

Finalized

Fields

docTopics :: !Table2D

Document topic counts

wordTopics :: !Table2D

Word topic counts

topics :: !Table1D

Topics counts

topicDocs :: !Table2D

Inverse document-topic counts

topicWords :: !Table2D

Inverse word-topic counts

alphasum :: !Double

alpha * K Dirichlet parameter (topic sparseness)

beta :: !Double

beta Dirichlet parameter (word sparseness)

topicNum :: !Int

Number of topics K

wSize :: !Int

Number of unique words

exponent :: !(Maybe Double)

Learning rate exponent

Instances

Generic Finalized

Initialization and finalization

initial :: Vector Word32 -> Int -> Double -> Double -> Maybe Double -> ST s (LDA s)

initial s k a b initializes model with k topics, a/k alpha hyperparameter, b beta hyperparameter and random seed s

finalize :: LDA s -> ST s Finalized

Create transparent immutable object holding model information from opaque internal representation

Querying evolving model

docTopicWeights_ :: LDA s -> Doc -> ST s (Vector Double)

priorDocTopicWeights_ :: LDA s -> D -> ST s (Vector Double)

Querying finalized model

docTopicWeights :: Finalized -> Doc -> Vector Double

docTopicWeights m doc returns unnormalized topic probabilities for document doc given LDA model m

wordTopicWeights :: Finalized -> D -> W -> Vector Double

topicWeights m d w returns the unnormalized probabilities of topics for word w in document d given LDA model m.

docCounts :: Finalized -> Table1D

For each document sum the topic counts