Add coherence and Diff logging for LDA #1381
Conversation
|
Please update coherence tutorial notebook with this new feature. |
|
I'm worried that there are a lot of arguments in the LDA constructor and they become even bigger. We need to think about this problem. |
parulsethi
changed the title
gensim/models/ldamodel.py
Outdated
| """ | ||
| cm = gensim.models.CoherenceModel(model=model, corpus=chunk, dictionary=self.id2word, coherence='u_mass') | ||
| corpus_words = sum(cnt for document in chunk for _, cnt in document) | ||
| logger.info("%.3f coherence estimate based on a held-out corpus of %i documents with %i words", cm.get_coherence(), len(chunk), corpus_words) |
There was a problem hiding this comment.
Calculate coherence only once and save it to variable
gensim/models/ldamodel.py
Outdated
| @@ -527,6 +528,15 @@ def log_perplexity(self, chunk, total_docs=None): | |||
| (perwordbound, np.exp2(-perwordbound), len(chunk), corpus_words)) | |||
| return perwordbound | |||
|
|
|||
| def log_coherence(self, model, chunk): | |||
There was a problem hiding this comment.
No needed to pass model at this method, you already have a model as self
parulsethi
changed the title
|
Updated to add diff logging (which was previously in #1399). |
| @@ -195,7 +197,9 @@ def __init__(self, corpus=None, num_topics=100, id2word=None, | |||
| alpha='symmetric', eta=None, decay=0.5, offset=1.0, | |||
| eval_every=10, iterations=50, gamma_threshold=0.001, | |||
| minimum_probability=0.01, random_state=None, ns_conf={}, | |||
| minimum_phi_value=0.01, per_word_topics=False): | |||
| minimum_phi_value=0.01, per_word_topics=False, coherence='u_mass', | |||
| coherence_texts=None, coherence_window_size=None, coherence_topn=10, | |||
There was a problem hiding this comment.
use metrics from another PR
| def update(self, corpus, chunksize=None, decay=None, offset=None, | ||
| passes=None, update_every=None, eval_every=None, iterations=None, | ||
| gamma_threshold=None, chunks_as_numpy=False): | ||
| gamma_threshold=None, chunks_as_numpy=False, coherence=None, |
There was a problem hiding this comment.
use metrics from another PR
|
Continued in #1399 |
Added a method to log coherence during training.