Sign Coordinate Regressor

This project fine-tunes a sentence-transformer model to predict a latitude and longitude pair from text observed on signs at an intersection.

The raw dataset is expected at training_data_raw.csv. It may include the pandas-exported index column; the preprocessing script only uses:

• intersection
• text_on_sign_exact
• latitude
• longitude

Workflow

Prepare bootstrapped training rows:

source .venv/bin/activate
python prepare_training_data.py --seed 1992 --bag-size 5 --samples-per-intersection 50

This writes training.csv with rows shaped like:

intersection,sample_id,text,latitude,longitude,unique_sign_count,raw_sign_count

Each row is a deterministic bootstrap sample of sign texts from one intersection, joined into a single text field. This trains on "some signs seen at this coordinate" instead of one sign or every sign at that coordinate.

Train the model:

python train.py

Evaluate and write predictions:

python eval.py

This writes predictions.csv and a map-style diagnostic plot:

It also writes a coordinate calibration plot:

Or run the full pipeline:

make

Useful Options

prepare_training_data.py:

• --seed: deterministic bootstrap seed.
• --bag-size: number of sign texts per sampled bag.
• --samples-per-intersection: number of bags generated per intersection.

train.py:

• --model-name: sentence-transformers base model.
• --epochs: training epochs.
• --batch-size: batch size.
• --device: device override such as cpu, cuda, or mps. Defaults to cuda.

Outputs

• training.csv: prepared bootstrapped dataset.
• output/: saved sentence-transformer encoder, coordinate head, and coordinate normalization metadata.
• predictions.csv: evaluation rows with predicted coordinates and error_km.
• plots/prediction_map.png: actual vs predicted coordinates with line segments showing the prediction error.
• plots/predicted_vs_actual.png: predicted vs actual latitude and longitude scatter plots.