Tomotopy coherence code is unnecessarily memory intensive

[hdanderson]

The current code in the tutorial for aggregating coherence at various numbers of topics is very memory intensive and can cause python to crash. This is because it aggregates all of the lda models in the lda_models array, even though lda_models is not used downstream. Removing this variable from the code should fix the issue.

[hdanderson]

Something like this should work for chunk 11

for k in range(len(ks)): 
    for r in range(reps):
        #perform topic modeling for a variety of topic numbers
        lda = tp.LDAModel(k=ks[k], rm_top=0)
        lda.burn_in = 50
        lda.add_corpus(corpus)
        #may want to change the parallel parameters
        lda.train(iter=150, parallel=4)
        coherence_cv[k,r] = Coherence(lda, coherence='c_v').get_score()