How marjanhs/prnn works
This repository is an authorship-verification experiment: it compares two documents and predicts whether they share an author. The implemented workflow loads embeddings and paired-document data, prepares sequences, runs stratified cross-validation with a parallel recurrent model, and reports metrics while saving intermediate outputs. The README also describes a transformation-encoder approach, but the sampled runtime source implements only the recurrent model; that documented capability is shown separately. Dataset wiring and the transformation approach are documented or represented incompletely, so unsupported runtime links are omitted.
Main entry points: main.py Cross-validation · util.py Pair data loader · model.py Parallel recurrent model · data Verification datasets
Components 15
Inputs
- Verification datasets
data - GloVe embeddings
Data preparation
- Embedding loader
util.py - Pair data loader
util.py - Text preprocessing
util.py - Sequence encoding
util.py
Verification models
- Parallel recurrent model
model.py - Pairwise metrics
util.py - Similarity classifier
model.py - Transformation encoder
README.md
Training and evaluation
- Cross-validation
main.py - Fit and evaluate
main.py - Fold metrics
main.py - Intermediate outputs
main.py
Other components
- Experiment runner
Connections 14
- Experiment runner → Cross-validation: starts CV
main.py - Cross-validation → Embedding loader: loads embeddings
main.py - Embedding loader → GloVe embeddings: reads vectors
util.py - Cross-validation → Pair data loader: loads pairs
main.py - Pair data loader → Text preprocessing: extracts text
util.py - Pair data loader → Verification datasets: reads dataset files
- Pair data loader → Sequence encoding: encodes pairs
util.py - Cross-validation → Parallel recurrent model: constructs model
main.py - Parallel recurrent model → Pairwise metrics: uses pair metrics
model.py - Parallel recurrent model → Similarity classifier: feeds similarity
model.py - Cross-validation → Fit and evaluate: trains folds
main.py - Fit and evaluate → Parallel recurrent model: fits and predicts
main.py - Fit and evaluate → Fold metrics: reports metrics
main.py - Fit and evaluate → Intermediate outputs: writes activations
main.py