---
license: cc-by-nc-sa-4.0
language:
- la
tags:
- latin
- medieval
- doc2vec
- embeddings
datasets:
- scta/scta-corpus
---

# SCTA Doc2Vec (2026.10)

Doc2Vec document vectors for the texts of the [Scholastic Commentaries and Texts Archive](https://scta.info),
**trained on [scta/scta-corpus 2026.10](https://huggingface.co/datasets/scta/scta-corpus/tree/2026.10)**, built with [scta-scikit](https://pypi.org/project/scta-scikit/).
Similar vectors mean similar content, even in different words. 2.1 GB.

| model | |
|---|---|
| `semantic` | trained on the paragraph texts: a vector for each unit of every text's structure, down to paragraphs |
| `semantic_lem` | the same, trained on the lemmatized texts |
| `semantic_wo_paras` | `semantic` without the paragraph vectors (levels and items only; smaller) |

Each is `models/<model>/vectors.npy` (300 dimensions, float32), `labels.json`
(the SCTA resource id of each row) and `info.json`; no pickles.
Training: PV-DM, window 10, min_count 5, negative 5, 40 epochs, seed 42.

## Use

```python
import scta_scikit as sc
sc.download(doc2vec=True)                 # these models and the corpus they were trained on
sc.download(doc2vec=["semantic_lem"])     # one of them
```
Then `scta sim <resource id> --semantic`.

## Versions

A release is tagged with the corpus version it was trained on (`2026.10`;
`2026.10-2` for a retraining on the same corpus), and `manifest.json` names
that corpus. Models are released less often than the corpus: a newer corpus
has texts these vectors don't cover.

## License

Derived from the SCTA corpus and published under the same license, Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0), https://creativecommons.org/licenses/by-nc-sa/4.0/.

Built 2026-10-24T12:44:01Z with scta-scikit 553c3a9; see `manifest.json`.