Auto n_components for multiple topic models

[x-tabdeveloping]

Added the following features:

Automatic detection of n_components

KeyNMF and GMM can now automatically detect the number of topics using the Bayesian Information Criterion.
The update also contains methods for effectively optimizing this quantity instead of using grid search.

from turftopic import KeyNMF, GMM

model = KeyNMF("auto")
model = GMM("auto")

[BETA] Contextualized Chunk Embeddings

You can now extract contextualized chunks' embeddings from documents with sentence-transformers using encode_chunks. This can sometimes enhance the performance of clustering topic models since they get access to smaller chunks of documents. More functionality coming soon.

from sentence_transformers import SentenceTransformer
from turftopic.encoders.utils import encode_chunks

encoder = SentenceTransformer("all-MiniLM-L6-v2")
chunk_embeddings, chunks = encode_chunks(encoder, sentences, return_chunks=True)

[BETA] Topeax

Added a new topic model, which detects clusters based on density peaks in document embedding space.
More details coming soon.

from turftopic import Topeax

model = Topeax()
model.fit(corpus)