app.py

import re
import gradio as gr
from transformers import pipeline
from transformers import AutoTokenizer
from transformers import AutoModelForSequenceClassification

def process_tweet(tweet):
    # remove links
    tweet = re.sub('((www\.[\s]+)|(https?://[^\s]+))', '', tweet)
    # remove usernames
    tweet = re.sub('@[^\s]+', '', tweet)
    # remove additional white spaces
    tweet = re.sub('[\s]+', ' ', tweet)
    # replace hashtags with words
    tweet = re.sub(r'#([^\s]+)', r'\1', tweet)
    # trim
    tweet = tweet.strip('\'"')
    return tweet #if len(tweet) > 0 else ""

tokenizer = AutoTokenizer.from_pretrained(
    "azamat/geocoder_model_xlm_roberta_50"
)

relevancy_pipeline = pipeline("sentiment-analysis", model="azamat/geocoder_model")

coordinates_model = AutoModelForSequenceClassification.from_pretrained(
    "azamat/geocoder_model_xlm_roberta_50",
)

def predict_relevancy(text):
    outputs = relevancy_pipeline(text)
    return outputs[0]['label'], outputs[0]['score']

def predict_coordinates(text):
    encoding = tokenizer(text, padding="max_length", truncation=True, \
        max_length=128, return_tensors='pt')
    outputs = coordinates_model(**encoding)
    return round(outputs[0][0][0].item(), 3), round(outputs[0][0][1].item(), 3)

def predict(text):
    text = process_tweet(text)
    relevancy_label, relevancy_score = predict_relevancy(text)
    if relevancy_label == 'relevant':
        lat, lon = predict_coordinates(text)
        return f"Confident for {round(relevancy_score * 100, 2)}% that tweet has the geolocation relevant information.\n" + \
            f"Predicted coordinates are: lat: {lat} lon: {lon}"
    return f"Confident for {relevancy_score * 100}% that tweet does not have the geolocation relevant information."

iface = gr.Interface(fn=predict, inputs="text", outputs="text")
iface.launch()