Model implementation map

This page maps every shipped model to the files a change touches: its core algorithm, its PyO3 binding, the shared machinery it builds on, the Cargo feature it needs, and where its reference-parity tests live. Use it to find the surfaces for a model without searching the tree.

It is generated from the registry, not hand-maintained here. The source of truth is the IMPL map in python/topica/registry.py; scripts/gen_model_tables.py renders it into the block below, and tests/test_registry.py fails CI if the block is stale, if IMPL and the model registry cover different models, or if any path or Cargo feature it names does not exist. So this map cannot silently drift from the code.

To change a row, edit registry.py and run python scripts/gen_model_tables.py (the preflight hook and CI reject a stale block); do not edit the generated block directly.

Column meanings:

• Source — the core algorithm file. A Rust src/*.rs (or topica-core/) file, or a python/topica/*.py module for the pure-Python models.
• Binding — the PyO3 binding: src/python/mod.rs, an extracted src/python/<model>.rs module, or — for a pure-Python model.
• Core / family — the shared machinery the model builds on (the collapsed Gibbs core in model.rs/sampler.rs, the CTM/STM variational core in topica-core, the ProdLDA VAE, and so on).
• Feature — the Cargo feature required beyond the default build. default builds with a plain cargo build; embeddings is the clustering-model gate (cargo test --features embeddings,umap,tsne covers it).
• Validation — the reference-parity / gold artifacts for the model. Every registered model is additionally covered by the conformance suite (tests/test_conformance.py).

General-purpose

Model	Source	Binding	Core / family	Feature	Validation
LDA	src/model.rs	src/python/mod.rs	SparseLDA collapsed Gibbs (model.rs, sampler.rs)	default	parity/lda_gold.py, parity/mallet_parity.py
CTM	topica-core/src/ctm.rs	src/python/mod.rs	CTM/STM variational core (topica-core)	default	parity/ctm_gold.py, parity/ctm_r_compare.py
ProdLDA	src/prodlda.rs	src/python/neural.rs	hand-coded batched VAE (prodlda.rs)	default	parity/prodlda_gold.py, parity/prodlda_compare.py
HDP	src/hdp.rs	src/python/mod.rs	collapsed Gibbs (model.rs, sampler.rs)	default	parity/hdp_gold.py
NMF	src/nmf.rs	src/python/nmf_lsa.rs	multiplicative-update matrix factorization	default	parity/nmf_vs_sklearn.py
LSA	src/lsa.rs	src/python/nmf_lsa.rs	truncated SVD (linalg)	default	parity/lsa_vs_sklearn.py
AnchorLDA	python/topica/anchor.py	— (Python)	spectral anchor-word recovery (Python over Rust primitives)	default	tests/test_anchor.py
TensorLDA	src/tlda.rs	src/python/tlda.rs	method-of-moments cumulants (linalg, spectral)	default	parity/tlda_gold.py, parity/tlda_compare.py
PolylingualLDA	src/pltm.rs	src/python/pltm.rs	collapsed Gibbs (model.rs, sampler.rs)	default	parity/pltm_compare.py

Covariates & structure

Model	Source	Binding	Core / family	Feature	Validation
STM	src/sts.rs	src/python/mod.rs	CTM/STM variational core (topica-core)	default	parity/stm_gold.py, parity/stm_r_compare.py
STS	src/sts.rs	src/python/mod.rs	CTM/STM variational core (topica-core)	default	parity/sts_gold.py, parity/sts_r_compare.py
SAGE	src/sage.rs	src/python/mod.rs	collapsed Gibbs + SAGE deviations	default	parity/sage_gold.py
DMR	src/dmr.rs	src/python/mod.rs	collapsed Gibbs + DMR prior (optimize.rs)	default	parity/dmr_gold.py
GDMR	src/dmr.rs	src/python/mod.rs	collapsed Gibbs + DMR prior (optimize.rs)	default	parity/gdmr_gold.py, parity/test_gdmr_tomotopy.py
Scholar	src/scholar.rs	src/python/scholar.rs	ProdLDA VAE + covariate prior (prodlda.rs)	default	tests/test_scholar.py
NarrativeTM	python/topica/narrative.py	— (Python)	intra-document trajectory over Gibbs core (Python)	default	tests/test_content_trajectory.py
RTM	src/rtm.rs	src/python/rtm.rs	variational EM + link head (optimize.rs digamma)	default	parity/rtm_compare.py, parity/rtm_reference.py, tests/test_rtm.py

Guided & supervised

Model	Source	Binding	Core / family	Feature	Validation
KeyATM	src/keyatm.rs	src/python/mod.rs	collapsed Gibbs + keyword index	default	parity/keyatm_gold.py, parity/keyatm_r_compare.py
SeededLDA	src/seeded.rs	src/python/mod.rs	collapsed Gibbs (model.rs, sampler.rs)	default	parity/seededlda_gold.py
LabeledLDA	src/labeled.rs	src/python/mod.rs	collapsed Gibbs (model.rs, sampler.rs)	default	parity/labeledlda_gold.py
SupervisedLDA	src/slda.rs	src/python/mod.rs	variational EM + Gaussian response head	default	parity/supervisedlda_gold.py
DiscLDA	src/disclda.rs	src/python/disclda.rs	collapsed Gibbs + class transform	default	parity/disclda_20ng.py, tests/test_disclda.py

Short text

Model	Source	Binding	Core / family	Feature	Validation
GSDMM	src/gsdmm.rs	src/python/mod.rs	collapsed Gibbs mixture (one topic/doc)	default	parity/gsdmm_gold.py
PT	src/pt.rs	src/python/mod.rs	collapsed Gibbs over pseudo-documents	default	parity/pt_gold.py
BTM	src/btm.rs	src/python/btm.rs	collapsed Gibbs over biterms	default	parity/btm_compare.py, tests/test_btm.py

Dynamic & hierarchical

Model	Source	Binding	Core / family	Feature	Validation
DTM	src/dtm.rs	src/python/mod.rs	variational Kalman over time slices	default	parity/dtm_gold.py
DETM	src/detm.rs	src/python/neural.rs	embedding VAE + LSTM q(eta) (etm_vae.rs)	default	parity/detm_gold.py
HLDA	src/hlda.rs	src/python/hierarchical.rs	nested-CRP collapsed Gibbs	default	parity/hlda_gold.py
PA	src/pa.rs	src/python/hierarchical.rs	collapsed Gibbs over a topic DAG	default	parity/pa_gold.py

Embedding-based

Model	Source	Binding	Core / family	Feature	Validation
BERTopic	src/bertopic.rs	src/python/embedding_cluster.rs	embedding clustering (cluster.rs, reduce.rs, represent.rs)	embeddings	parity/bertopic_gold.py
Top2Vec	src/top2vec.rs	src/python/embedding_cluster.rs	embedding clustering (cluster.rs, reduce.rs)	embeddings	parity/top2vec_gold.py, parity/top2vec_compare.py
ETM	src/etm.rs	src/python/neural.rs	variational EM over word embeddings (ctm.rs)	default	parity/etm_gold.py
IdealPointTM	src/idealpoint.rs	src/python/idealpoint.rs	variational EM + ideal-point head	default	tests/test_idealpoint.py, tests/test_idealpoint_counts.py
IdealPointSentenceTM	src/sentence_ideal.rs	src/python/sentence_ideal.rs	Gaussian-cluster EM over embeddings	default	tests/test_sentence_ideal.py
FASTopic	src/fastopic.rs	src/python/mod.rs	reverse-mode Sinkhorn optimal transport	default	parity/fastopic_gold.py, parity/fastopic_compare.py
EmbeddingLDA	python/topica/embedding.py	— (Python)	seeded Gibbs + embedding NN expansion (Python)	default	parity/embeddinglda_gold.py, tests/test_embedding_lda.py
CombinedTM	src/prodlda.rs	src/python/neural.rs	contextualized ProdLDA VAE (prodlda.rs)	default	parity/combinedtm_gold.py, parity/combinedtm_compare.py
ZeroShotTM	src/prodlda.rs	src/python/neural.rs	contextualized ProdLDA VAE (prodlda.rs)	default	parity/zeroshot_gold.py, parity/zeroshot_compare.py
InfoCTM	src/infoctm.rs	src/python/neural.rs	two ProdLDA VAEs + TAMI alignment (prodlda.rs)	default	parity/infoctm_gold.py, parity/infoctm_compare.py

Ideal point

Model	Source	Binding	Core / family	Feature	Validation
Wordfish	src/wordfish.rs	src/python/wordfish.rs	Poisson-scaling EM	default	parity/wordfish_r_compare.py, tests/test_wordfish.py
TBIP	src/tbip.rs	src/python/tbip.rs	Poisson-factorization mean-field SVI	default	parity/tbip_parity.py, tests/test_tbip.py
PartyEmbeddings	src/party_embeddings.rs	src/python/party_embeddings.rs	PV-DM paragraph vectors (negative sampling)	default	parity/party_embeddings_compare.py, tests/test_party_embeddings.py

LLM-based

Model	Source	Binding	Core / family	Feature	Validation
TopicGPT	python/topica/llm.py	— (Python)	LLM prompting pipeline (Python)	default	tests/test_topicgpt.py