API reference
Corpus
Bases: Model
Documents from a DataFrame (data) or a file (data_file), as a tomotopy corpus.
load()
Tokenize the documents (and concatenate n-grams) into a tomotopy corpus.
TopicModel
Bases: Model
A topic model whose artifacts live under output_dir.
File names start with model_id (<MODEL_TYPE>_<name>_k(<k>)), so models that differ
in name or k can share one directory.
num_topics
property
Number of topics in the model
model_summary_file
property
One JSON line per trained model, shared by every model in output_dir.
train(corpus=None)
Train on corpus (or the one already set) and evaluate coherence.
Nothing is written; call :meth:save, :meth:summarize, and :meth:plot for artifacts.
save()
Write the model file, log-likelihood trace, distributions, and top words.
summarize()
Write the tomotopy summary and append this model's metrics to the summary jsonl.
plot()
Write the log-likelihood plot, topic word clouds, and LDAvis (if installed).
load()
Load the model file and saved distributions written by :meth:save for model_id.
infer(corpus, output_file=None, iterations=100, tolerance=-1, num_workers=0, together=False)
Infer topic distributions for corpus and save them (with log-likelihoods).
LdaConfig
Bases: Model
Keyword arguments for tomotopy.LDAModel.
TrainSummaryConfig
Bases: Model
Keyword arguments for tomotopy.LDAModel.summary.
WordPrior
Bases: Model
Seed words per topic, from prior_data and/or a YAML or JSON data_file.
add(id_, prior)
Add a prior to the word prior
load(missing_ok=False)
Load words from data_file and prior_data.
A missing data_file raises FileNotFoundError unless missing_ok (used by
:meth:save to create the file).
save()
Save words to data_file: JSON for a .json file, YAML otherwise.
WordCloud
Bases: Model
Keyword arguments for wordcloud.WordCloud; mask may be an image path.
save_image_pages(images, output_dir, output_file_format, num_images_per_page, num_cols, num_rows=None, titles=None, title_fontsize=10, title_color='black', figsize=None, width_multiple=4, height_multiple=2, dpi=100)
Lay images out in grids of num_images_per_page and save one file per page.
Raises ValueError when an explicit num_rows x num_cols grid cannot hold a page.
Workflow steps: small functions that chain a load, corpus, and model through HyFI steps:.
Each step returns what the next one takes as its first argument (HyFI pipe). They work the
same when called from plain Python.
build_corpus(data=None, *, corpus)
Tokenize data (or corpus.data_file when no data is piped in) into corpus.
train(corpus, *, model)
Train model on corpus and save the model file and its distributions.
sweep(corpus, *, model, k, alpha=None, eta=None)
Train, save, summarize, and plot one model per (k, alpha, eta) combination.
Models are named <name>(0), <name>(1), ... in sweep order, and every model's summary
is appended to <output_dir>/model-summary.jsonl. Returns the summaries.
infer(corpus, *, model, output_file=None, iterations=100, tolerance=-1, num_workers=0, together=False)
Load model (as saved under its output_dir) and infer topics for corpus.