app.py

import gradio as gr
from transformers import pipeline
import geopy
import plotly.graph_objects as go
import torch

device = torch.device("mps" if torch.backends.mps.is_available() else ("cuda" if torch.cuda.is_available() else "cpu"))

binary_classifier = pipeline("text-classification", model="GEODE/bert-base-multilingual-cased-binary-classifier-edda-coords", truncation=True, device=device)
ner_pipeline = pipeline("token-classification", model="GEODE/camembert-base-edda-span-classification", aggregation_strategy="simple", device=device)
generator = pipeline("text2text-generation", model="GEODE/mt5-small-coords-norm", truncation=True, device=device)


def create_map(lat, long):
    fig = go.Figure(go.Scattermapbox(
            #customdata=text_list,
            lat=[lat],
            lon=[long],
            mode='markers',
            marker=go.scattermapbox.Marker(
                size=10
            ),
            #hoverinfo="text",
            #hovertemplate='<b>Name</b>: %{customdata[0]}<br><b>Price</b>: $%{customdata[1]}'
        ))

    fig.update_layout(
        mapbox_style="open-street-map",
        #hovermode='closest',
        mapbox=dict(
            bearing=0,
            center=go.layout.mapbox.Center(
                lat=lat,
                lon=long
            ),
            pitch=0,
            zoom=3
        ),
        margin=dict(l=0, r=0, t=0, b=0)  # <-- removes margins
    )

    return fig

def dms_to_dd(dms):
    try:
        point = geopy.Point(dms)
        return [point[0], point[1]-17.66]
    except:
        return None

def detect_coordinates(text):
    # Run binary classification
    result = binary_classifier(text)
    
    if result[0]['label'] == 'Positive':
        return "Coordinates found"
    else:
        return "No coordinates found"

def extract_coordinates(text):
    if detect_coordinates(text) == "No coordinates found":
        return "No coordinates found"
    # Run NER
    entities = ner_pipeline(text)
    
    # Collect coordinate entities
    coords_text = [ent['word'] for ent in entities if ent['entity_group'] == 'Latlong']
    
    if coords_text:
        return "\n".join(coords_text)
    else:
        return "No coordinates found"

# bert-base-multilingual-cased-binary-classifier-edda-coords

def norm_coordinates(text):
    result_text = ""
    if detect_coordinates(text) == "No coordinates found":
        result_text = "No coordinates found"
    # Example input text
    input_text = "extract_coordinates: " + text

    # Generate prediction using the pipeline
    predicted_coordinates_from_pipeline = generator(input_text, max_length=128)

    result_text = predicted_coordinates_from_pipeline[0]['generated_text']
    
    coords = dms_to_dd(result_text)
    
    return result_text, create_map(coords[0], coords[1]) if coords else None
    

examples = [
    "* AACH ou ACH, s. f. petite ville d'Allemagne dans le cercle de Souabe, près de la source de l'Aach. Long. 26. 57. lat. 47. 55.",
    "* ARCALU (Principauté d') petit état des Tartares-Monguls, sur la riviere d'Hoamko, où commence la grande muraille de la Chine, sous le 122e degré de longitude & le 42e de latitude septentrionale.",
    "* ARÉQUIPE, ou ARIQUIPA, (Géog.) ville de l'Amérique méridion. dans le Pérou, sur une riviere, dans un terrein fertile. Long. 308. lat. mérid. 16. 40.",
    "* AUTAN-KELURAN, (Géog.) ville du Turquestan. Long. 110d. & lat. 46. 45. selon Uluhbeg ; & long. 116. & lat. 45. selon Nassiredden.",
    "Boston ; c'est le nom qu'on a donné à la ville capitale de la nouvelle Angleterre, dans l'Amérique septentrionale ; elle est grande & a un très-bon port. Lat. 42 degrés, 20 minutes ; long. 306 degrés, 50 & quelques minutes.",
    "CABEÇA-DE-VIDE, (Géog.) petite ville avec château, en Portugal, dans l'Alentéjo, à cinq lieues de Port-Alegre. Longitude 10. 48. latitude 39."
]


description = """
# 🌍 Geographic Coordinate Extractor

This Space demonstrates the performance of AI models trained to automate the extraction of geographical coordinates from the 18th-century **Encyclopédie of Diderot and d'Alembert**.

### How it Works:
1. **Classification:** A BERT-like model identifies whether a given encyclopedia entry contains geographic coordinates.
2. **Extraction & Normalization:** An mT5-based model retrieves raw coordinate strings from the text and normalizes them into a standard DMS format.

You can test the models by entering a custom text or choosing from the examples provided.

- **Authors:** Ludovic Moncla, Pierre Nugues, Thierry Joliveau, and Katherine McDonough.  
- **Project:** [GEODE](https://geode-project.github.io)

### Reference:
> Moncla, L., Nugues, P., Joliveau, T., & McDonough, K. (2026). **EDDA-Coordinata: An Annotated Dataset of Historical Geographic Coordinates**. *arXiv preprint [arXiv:2602.23941](https://arxiv.org/abs/2602.23941). (accepted at LREC 2026)*
---
"""

with gr.Blocks() as demo:
    gr.Markdown(description)
    
    with gr.Row():
        with gr.Column():
            inp = gr.Textbox(
                label="Enter text", 
                placeholder="e.g. * AACH ou ACH, s. f. petite ville d'Allemagne dans le cercle de Souabe, près de la source de l'Aach. Long. 26. 57. lat. 47. 55.",
                lines=3
            )

            run_btn = gr.Button("Extract & Show")
            
            gr.Examples(
                examples=examples,
                inputs=inp,
                label="Examples"
            )
    
        with gr.Column():
            out_text = gr.Textbox(label="Detect coordinates (fine-tuned BERT binary classifier)")
            run_btn.click(fn=detect_coordinates, inputs=inp, outputs=out_text)

            out_text = gr.Textbox(label="Extract coordinates (fine-tuned CamemBERT NER)")
            run_btn.click(fn=extract_coordinates, inputs=inp, outputs=out_text)

            out_text = gr.Textbox(label="Extract and normalize DMS coordinates (fine-tuned mT5)")
            map = gr.Plot(label="Coordinates on Map")
            run_btn.click(fn=norm_coordinates, inputs=inp, outputs=[out_text, map])
        
# Launch
if __name__ == "__main__":
    demo.launch()