app.py

import pandas as pd
import streamlit as st
from utils import df_to_html, render_svg, combine_json_files, render_metadata, color_mapping

data = combine_json_files('./languages')

## Filter data from religous texts and bad confidences
filtered_data = {}
for isocode, details in data.items():
    filtered_sites = [
        site for site in details.get("Sites", [])
        if site["Confidence"] != "🟥" and "religious" not in site["Category"].lower()
    ]
    
    filtered_data[isocode] = details.copy()
    filtered_data[isocode]["Sites"] = filtered_sites
data = filtered_data


@st.cache_data
def render_home_table():
    """Renders home table."""
    # Compute number of unique domains/urls
    for key in data.keys():
        data[key]['Number of Sites'] = len(data[key].get('Sites', []))
        data[key]["Number of Links"] = sum(len(url_data["Links"]) for url_data in data[key].get('Sites', []))

    # Convert dict to df
    df_data = pd.DataFrame(data).transpose()
    df_data['ISO Code'] = df_data.index

    # Filter 0 sites
    df_data = df_data[df_data["Number of Sites"]!= 0]

    df_data['Number of Sites'] = df_data['Number of Sites'].astype(str)  # Convert to string
    df_data['ISO Code'] = df_data['ISO Code'].astype(str)  # Convert to string
    df_data['Number of Sites'] = df_data.apply(lambda row: '<a href="/?isocode={}&site=True" target="_self">{}</a>'.format(row['ISO Code'], row['Number of Sites']), axis=1)
    df_data['Number of Links'] = df_data.apply(lambda row: '<a href="/?isocode={}&links=True" target="_self">{}</a>'.format(row['ISO Code'], row['Number of Links']), axis=1)
    df_data["Support by MADLAD400, FLORES200, GLOT500"] = df_data.apply(lambda row: color_mapping([row["Supported by allenai/MADLAD-400"] + row["Supported by facebook/flores"] +  row["Supported by cis-lmu/Glot500"]]), axis =1)
    df_data['Color_Order'] = pd.Categorical(df_data['Support by MADLAD400, FLORES200, GLOT500'], categories=['🟥', '🟧', '🟨', '🟩'], ordered=True)
    # Sort by Color_Order then ISO Code
    df_data = df_data.sort_values(by=['Color_Order', 'ISO Code'])

    # Filter 🟩
    df_data = df_data[df_data["Support by MADLAD400, FLORES200, GLOT500"]!= '🟩']

    # Display the table
    df_data = df_data[['ISO Code', 'Language Name', 'Family', 'Number of Sites', 'Number of Links', 'Number of Speakers', 'Support by MADLAD400, FLORES200, GLOT500']]
    st.write(df_to_html(df_data), unsafe_allow_html=True)

@st.cache_data
def render_site_table(isocode):

    # back 
    back_text = '<a href="/?home=True" target="_self">[Back]</a>'
    st.markdown(back_text, unsafe_allow_html=True)

    # site
    urls = data[isocode].get('Sites', [])
    df_urls = pd.DataFrame(urls)
    df_urls['Number of Links'] = df_urls['Links'].apply(len)
    df_urls = df_urls.sort_values(by='Number of Links', ascending=False)
    df_urls = df_urls.reset_index(drop=True)
    df_urls['Number of Links'] = df_urls.apply(lambda row: '<a href="/?isocode={}&siteurl={}" target="_self">{}</a>'.format(isocode, row['Site URL'], row['Number of Links']), axis=1)
    df_urls['Site URL'] = df_urls['Site URL'].apply(lambda url: f'<a href="{url}">{url}</a>' if url != 'misc' else url)
    df_urls['Language Name'] = data[isocode]['Language Name']
    df_urls['ISO Code'] = isocode

    # Display the table
    df_urls = df_urls[['ISO Code', 'Site URL', 'Category', 'Number of Links', 'Possible Parallel Languages', 'Confidence']]
    st.write(df_to_html(df_urls), unsafe_allow_html=True)


@st.cache_data
def render_siteurl_table(isocode, url):

    # back
    back_text = '<a href="/?isocode={}&site=True" target="_self">[Back]</a>'.format(isocode)
    st.markdown(back_text, unsafe_allow_html=True)

    # Find selected domain
    urls = data[isocode].get('Sites', [])
    selected_domain = next((d for d in urls if 'Site URL' in d and d['Site URL'] == url), None)
    
    if selected_domain:
        st.write({'Language Name': data[isocode]['Language Name'], 'ISO Code': isocode, 'Site URL': url, 'Links': selected_domain['Links']})



@st.cache_data
def render_links_table(isocode):

    # back 
    back_text = '<a href="/?home=True" target="_self">[Back]</a>'
    st.markdown(back_text, unsafe_allow_html=True)

    # output
    urls = data[isocode].get('Sites', [])
    lang_name = data[isocode]['Language Name']
    all_urls = [{'Site URL': du['Site URL'], 'Links': du['Links']} for du in urls] 
        
    st.write({'Language Name': lang_name, 'ISO Code': isocode, 'URLs': all_urls})



# show logo
render_svg(open("assets/glotweb_logo.svg").read())

def main():
    params = st.query_params

    if 'isocode' in params:
        if 'siteurl' in params:
            render_siteurl_table(params['isocode'], params['siteurl'])
        if 'site' in params:
            render_site_table(params['isocode'])
        if 'links' in params:
            render_links_table(params['isocode'])

    else:
        # show home
        render_metadata()
        st.markdown("**GlotWeb** is an indexing service for low-resource languages. It indexes (almost) **non-religous** sites and links written in each language. This list can be used to create raw text or mining parallel corpora and to study low-resource languages on the web.\n")
        render_home_table()
        st.markdown("\n\n<font color='gray'>We compare the level of support for these languages in the three big datasets ([MADLAD400](https://huggingface.co/datasets/allenai/MADLAD-400), [FLORES200](https://huggingface.co/datasets/facebook/flores), [GLOT500](https://huggingface.co/datasets/cis-lmu/Glot500)) of low-resource languages (🟥 0/3 < 🟧 1/3 < 🟨 2/3 < 🟩 3/3). Although the support in these datasets for some of these languages could be just the religious texts.</font>", unsafe_allow_html=True)

main()