How do I get perplexity and log likelihood in Spar

Question

I'm trying to get perplexity and log likelihood of a Spark LDA model (with Spark 2.1). The code below does not work (methods logLikelihood and logPerplexity not found) although I can save the model.

from pyspark.mllib.clustering import LDA
from pyspark.mllib.linalg import Vectors

# construct corpus
# run LDA
ldaModel = LDA.train(corpus, k=10, maxIterations=10)
logll = ldaModel.logLikelihood(corpus)
perplexity = ldaModel.logPerplexity(corpus)

Notice that such methods do not come up with dir(LDA).

What would be a working example?

Answer 1

I can do train but not fit. 'LDA' object has no attribute 'fit'

That's because you are working with the old, RDD-based API (MLlib), i.e.

from pyspark.mllib.clustering import LDA # WRONG import

whose LDA class indeed does not include fit, logLikelihood, or logPerplexity methods.

In order to work with these methods, you should switch to the new, dataframe-based API (ML):

from pyspark.ml.clustering import LDA  # NOTE: different import

# Loads data.
dataset = (spark.read.format("libsvm")
    .load("data/mllib/sample_lda_libsvm_data.txt"))

# Trains a LDA model.
lda = LDA(k=10, maxIter=10)
model = lda.fit(dataset)

ll = model.logLikelihood(dataset)
lp = model.logPerplexity(dataset)