Eddy1919/openetruscan-classifier

OpenEtruscan Neural Inscription Classifier

Model Description

Character-level CNN and Micro-Transformer models for automated classification of Etruscan inscriptions into 7 categories:

Category	Description
funerary	Tomb inscriptions, epitaphs, death/life formulae
votive	Offerings to deities, dedication verbs
boundary	Territorial markers, civic designations
ownership	Object ownership marks ("mi" = I am)
legal	Administrative texts, magistrate titles
commercial	Trade records, numerals, vessel terms
dedicatory	Temple dedications, deity names

Architecture

• CharCNN (recommended): 27,943 parameters, F1=0.72
• MicroTransformer: 273,159 parameters, F1=0.64

Both models operate at the character level — no tokenizer needed. Input is a raw Etruscan inscription string.

Training Data

1,050 weakly-labeled Etruscan inscriptions bootstrapped via keyword-based weak supervision from the OpenEtruscan corpus (4,728 inscriptions from the Larth dataset + Burman concordance enrichment).

Usage

Python (PyTorch)

from openetruscan.neural import NeuralClassifier

clf = NeuralClassifier()
clf.load("path/to/models", model_type="cnn")
print(clf.predict("mi araθia velθurus"))  # → "ownership"

ONNX Runtime (lightweight inference)

import onnxruntime as ort
import numpy as np
import json

with open("cnn.json") as f:
    meta = json.load(f)

session = ort.InferenceSession("cnn.onnx")
# Tokenize: map each character to vocab index
text = "mi araθia velθurus"
ids = [meta["vocab"]["char_to_idx"].get(c, 1) for c in text.lower()]
ids = ids[:128] + [0] * max(0, 128 - len(ids))  # pad to 128

logits = session.run(None, {"input": np.array([ids], dtype=np.int64)})[0]
label = meta["labels"][logits.argmax()]
print(label)  # → "ownership"

In-Browser (onnxruntime-web)

The ONNX model runs directly in the browser via WebAssembly. See the OpenEtruscan web app Classifier tab for a live demo.

Files

File	Description
cnn.onnx	CNN model in ONNX format (production)
cnn.json	Vocabulary + label metadata for CNN
cnn.pt	CNN PyTorch weights
transformer.onnx	Transformer model in ONNX format
transformer.json	Vocabulary + label metadata for Transformer
transformer.pt	Transformer PyTorch weights
metrics.json	Training metrics and per-class F1 scores

Citation

If you use these models in your research, please cite:

@software{openetruscan_neural,
  title = {OpenEtruscan Neural Inscription Classifier},
  author = {Panichi, Edoardo},
  year = {2024},
  url = {https://github.com/Eddy1919/openEtruscan},
  license = {MIT}
}

License

MIT