Migrating from 0.x
ThematOS 1.0 runs on HyFI 2.0 and no longer imports HyFI itself. To stay on the old API, pin thematos<1 (which needs hyfi<2, lexikanon<1, and Python below 3.13).
| 0.x | 1.0 |
|---|---|
from thematos import HyFI, thematos console command |
import hyfi only where you compose configs; the hyfi CLI |
TopicModel(BatchTask), Corpus(BatchTask) with task_name, batch_name, batch, path |
plain Pydantic models; LdaModel(output_dir=...) |
files under workspace/<task_name>/<batch_name>/{models,outputs} |
everything under output_dir (${project.workspace}/topic/model in the config) |
model id LDA_model(0)_k(5) from the batch number |
LDA_<name>_k(5); the sweep names models model(0), model(1), ... |
batch.seed (random when -1), batch.num_workers |
seed (random when null), num_workers |
save_config() / load(batch_num=...) |
the run manifest records the config; load() reads the files for the same output_dir, name, and k |
train() with autosave: true wrote every artifact |
train(corpus) writes nothing; chain .save().summarize().plot(), or use the workflow steps |
Corpus(data_load=...) (a HyFI load_dataframe run) |
Corpus(data_file=...) or Corpus(data=dataframe) |
corpus doc ids saved to corpus_doc_ids.parquet |
ids are in <id>-doc_topic_dists.parquet |
TopicRunner with calls: [train, infer], LdaRunConfig, InferConfig |
workflows thematos_lda_sweep and thematos_lda_infer (thematos.steps.sweep, thematos.steps.infer) |
runner/<name>_summaries.json |
model-summary.jsonl, one line per model |
groups model/config, model/train, model/summary, model/plot, ngrams, runner, about |
model_args, train_args, train_summary_args, and ngrams are inline in model=lda and dataset=topic_corpus; model/plot is plot=wordcloud |
/stopwords@stopwords: __init__ (lexikanon 0.x) |
/stopwords@stopwords: default (lexikanon 1.x) |
LdaConfig.k: null |
k: 10 |
generate_config() and other HyFI model helpers |
removed |
word cloud font picked by HyFI (lang) |
wc.font_path; wordcloud's default font when unset |
Results did not change: with the same corpus, word priors, seed, and one worker, the tests reproduce the perplexity, coherence, and top words recorded with thematos 0.14.5 on hyfi 1.37.0 (tests/data/metrics_1x.json).
Fixed along the way:
- tomotopy 0.14 renamed
train(iter=)andinfer(iter=)toiterations=; 1.0 requires tomotopy 0.14 or newer. WordPrior.save()works whendata_filedoes not exist yet.