artm_determine_topics.py

def determine_optimal_artm_topics(data, 
                                  tokenized_documents, 
                                  batch_vectorizer, 
                                  min_topics, 
                                  max_topics):
    
    best_coherence = 0
    best_topics = 0
        
    for n_topics in range(min_topics, max_topics, 1):
        
        topic_names = ['topic_{}'.format(i) for i in range(n_topics)] # initialize topic names
        model_artm = get_artm_model(topic_names, batch_vectorizer.dictionary) # initalize ARTM model
        model_artm.fit_offline(batch_vectorizer=batch_vectorizer, num_collection_passes=10) # train model
        topics = get_artm_topics(model_artm) # get topic representations
        coherence = get_coherence(topics, tokenized_documents) # calculate coherence

        if coherence > best_coherence:
            best_coherence = coherence
            best_n_topics = n_topics
            
    return best_n_topics