I am training a ldamallet model in python and saving it. I am also saving training dictionary that I can use to create corpus for unseen documents later. If I perform every action (i.e. train a model, save trained model, load saved model, infer unseen corpus) within same console, everything works fine. However, I want to use the trained model in different console / computer.
I passed prefix while training to look at the temp files created by the model. Following files are created when the model is trained:
'corpus.mallet'
'corpus.txt'
'doctopics'txt'
inferencer.mallet'
'state.mallet.gz'
'topickeys.txt'
Now when I load the saved model in a different console and infer unseen corpus created using the saved dictionary, I can see no other temp files being created and produces following error:
FileNotFounderror: No such file or directory : 'my_directory\\doctopics.txt.infer'
For some odd reason, if I load the saved model in same console (console it was trained on) and infer unseen corpus like above, 'corpus.txt' is updated and two new temp files are created:
'corpus.mallet.infer'
'doctopics.txt.infer'
Any idea why I might be having this issue?
I have tried using LdaModel instead of LdaMallet and LdaModel works fine irrespective of whether I perform whole task in same console or different console.
Below is the snippet of the code I am using.
def find_optimum_model(self):
lemmatized_words = self.lemmatization()
id2word = corpora.Dictionary(lemmatized_words)
all_corpus = [id2word.doc2bow(text) for text in lemmatized_words]
#For two lines below update with your path to new_mallet
os.environ['MALLET_HOME'] = r'C:\\users\\axk0er8\\Sentiment_Analysis_Working\\new_mallet\\mallet-2.0.8'
mallet_path = r'C:\\users\\axk0er8\\Sentiment_Analysis_Working\\new_mallet\\mallet-2.0.8\\bin\\mallet.bat'
prefix_path = r'C:\\users\\axk0er8\\Sentiment_Analysis_Working\\new_mallet\\mallet_temp\\'
def compute_coherence_values(dictionary, all_corpus, texts, limit, start=2, step=4):
coherence_values = []
model_list = []
num_topics_list = []
for num_topics in range(start, limit, step):
model = gensim.models.wrappers.LdaMallet(mallet_path, corpus=all_corpus, num_topics=num_topics, id2word=dictionary,
random_seed=42)
#model = gensim.models.ldamodel.LdaModel(corpus=all_corpus,num_topics=num_topics,id2word=dictionary,eval_every=1,
# alpha='auto',random_state=42)
model_list.append(model)
coherencemodel = CoherenceModel(model=model, texts=texts, dictionary=dictionary, coherence='c_v')
coherence_values.append(coherencemodel.get_coherence())
num_topics_list.append(num_topics)
return model_list, coherence_values, num_topics_list
model_list, coherence_values, num_topics_list = compute_coherence_values(dictionary=id2word,all_corpus=all_corpus,
texts=lemmatized_words,start=5,limit=40, step=6)
model_values_df = pd.DataFrame({'model_list':model_list,'coherence_values':coherence_values,'num_topics':num_topics_list})
optimal_num_topics = model_values_df.loc[model_values_df['coherence_values'].idxmax()]['num_topics']
optimal_model = gensim.models.wrappers.LdaMallet(mallet_path, corpus=all_corpus, num_topics=optimal_num_topics, id2word=id2word,
prefix=prefix_path, random_seed=42)
#joblib.dump(id2word,'id2word_dictionary_mallet.pkl')
#joblib.dump(optimal_model,'optimal_ldamallet_model.pkl')
id2word.save('id2word_dictionary.gensim')
optimal_model.save('optimal_lda_model.gensim')
def generate_dominant_topic(self):
lemmatized_words = self.lemmatization()
id2word = corpora.Dictionary.load('id2word_dictionary.gensim')
#id2word = joblib.load('id2word_dictionary_mallet.pkl')
new_corpus = [id2word.doc2bow(text) for text in lemmatized_words]
optimal_model = gensim.models.wrappers.LdaMallet.load('optimal_lda_model.gensim')
#optimal_model = joblib.load('optimal_ldamallet_model.pkl')
def format_topics_sentences(ldamodel, new_corpus):
sent_topics_df = pd.DataFrame()
for i, row in enumerate(ldamodel[new_corpus]):
row = sorted(row, key=lambda x: (x[1]), reverse=True)
for j, (topic_num, prop_topic) in enumerate(row):
if j == 0:
wp = ldamodel.show_topic(topic_num)
topic_keywords = ", ".join([word for word, prop in wp])
sent_topics_df = sent_topics_df.append(pd.Series([int(topic_num), round(prop_topic,4), topic_keywords]),
ignore_index=True)
else:
break
sent_topics_df.columns = ['Dominant_Topic', 'Perc_Contribution', 'Topic_Keywords']
return (sent_topics_df)
My expectation is use find_optimum_model function with the training data and save the optimum model and dictionary. Once saved, use generate_dominant_topic function to load saved model and dictionary, create corpus for unseen text and run the model to get desired topic modeling output.
