app.py

import gradio as gr
from pathlib import Path
import pandas as pd

from src.process_data import *


abs_path = Path(__file__).parent
CITATION_TEXT = open(f"pages/citation.bib", "r").read()

title=(
    """
    <center> 
    <h1> CAMEO Leaderboard</h1>
    </center>
    """
)

description=(
    """
    <center>
    <h3> The CAMEO Leaderboard ranks and evaluates the Speech Emotion Recognition (SER) models for the CAMEO dataset.</h3>
    </center>
    """
)


def app():
    with gr.Blocks() as demo:
        gr.HTML("""
                <link href="https://fonts.googleapis.com/css2?family=Inter&display=swap" rel="stylesheet">
            """)

        demo.css = """
            .tab-item {
                font-size: 14px;
                padding: 10px 20px;
                font-family: 'Inter', sans-serif;
            }
            """

        gr.HTML(title, elem_classes='tab-item')
        gr.HTML(description, elem_classes='tab-item')


        with gr.Tabs():
            with gr.Tab("🏆 Leaderboard", elem_classes='tab-item'):
                languages = ['All', 'Bengali', 'English', 'French', 'German', 'Italian', 'Polish', 'Russian', 'Spanish']
                datasets = ['All', 'CaFE', 'CREMA-D', 'EMNS', 'Emozionalmente', 'eNTERFACE', 'JL-Corpus', 'MESD', 'nEMO', 'Oreau', 'PAVOQUE', 'RAVDESS', 'RESD', 'SUBESCO']
                emotions = ['All', 'anger', 'anxiety',
               'apology', 'assertiveness', 'calm', 'concern', 'disgust',
               'encouragement', 'enthusiasm', 'excitement', 'fear', 'happiness',
               'neutral', 'poker', 'sadness', 'sarcasm', 'surprise']
                metric=["f1_macro", "accuracy", "weighted_f1"]

                # with gr.Tabs():
                with gr.Tab("Overall Results", elem_classes='tab-item'):
                    gr.Markdown('The results show the values of the `f1_macro`, `weighted_f1` and `accuracy` metrics between the expected and generated emotion labels by each model with the given `temperature`.', elem_classes='tab-item')
                    overall_table = gr.Dataframe()

                with gr.Tab("Results per Language", elem_classes='tab-item'):
                    gr.Markdown('The results show the values of the `f1_macro`, `weighted_f1` or `accuracy` metrics between the expected and generated emotion labels by each model with a given `temperature` for each `language` in the CAMEO dataset.', elem_classes='tab-item')
                    languages_filter = gr.CheckboxGroup(choices=languages, label="Select column", value=languages)
                    select_lang_metric = gr.Radio(metric, value='f1_macro', label="Select metric")
                    lang_table = gr.Dataframe()

                with gr.Tab("Results per Dataset", elem_classes='tab-item'):
                    gr.Markdown(
                        'The results show the values of the `f1_macro`, `weighted_f1` or `accuracy` metrics between the expected and generated emotion labels by each model with a given `temperature` for each `dataset` in the CAMEO dataset.',
                        elem_classes='tab-item')
                    dataset_filter = gr.CheckboxGroup(choices=datasets, label="Select column", value=datasets)
                    select_ds_metric = gr.Radio(metric, value='f1_macro', label="Select metric")
                    dataset_table = gr.Dataframe()

                with gr.Tab("Results per Emotion", elem_classes='tab-item'):
                    gr.Markdown(
                        'The results show the values of the `f1_macro` metrics between the expected and generated emotion labels by each model with a given `temperature` for each `emotion` in the CAMEO dataset.',
                        elem_classes='tab-item')
                    emo_filter = gr.CheckboxGroup(choices=emotions, label="Select column", value=emotions)
                    emotion_table = gr.Dataframe()

                df_state = gr.State()

                def update_leaderboards(languages=[], datasets=[], emotions=[], select_lang_metric="f1_macro", select_ds_metric="f1_macro"):
                    df = pd.read_json(str(abs_path / "results.jsonl"), lines=True)
                    lang_dict = build_lang_dict(df)
                    ds_dict = build_ds_dict(df)
                    emo_dict = build_emo_dict(df)
                    overall = overall_leaderboard(df)
                    by_lang = leaderboard_per_group(lang_dict, languages, metric=select_lang_metric)
                    by_dataset = leaderboard_per_group(ds_dict, datasets, metric=select_ds_metric)
                    by_emotion = leaderboard_per_group(emo_dict, emotions)
                    return overall, by_lang, by_dataset, by_emotion, "Loaded successfully."

                demo.load(
                    update_leaderboards,
                    inputs=[languages_filter, dataset_filter, emo_filter],
                    outputs=[overall_table, lang_table, dataset_table, emotion_table, df_state]
                )

                def on_change(selected_languages, selected_lang_metric, selected_datasets, selected_ds_metric, selected_emotions):
                    return update_leaderboards(languages=selected_languages, select_lang_metric=selected_lang_metric, datasets=selected_datasets, select_ds_metric=selected_ds_metric, emotions=selected_emotions)

                languages_filter.change(on_change, [languages_filter, select_lang_metric, dataset_filter, select_ds_metric, emo_filter],
                                       [overall_table, lang_table, dataset_table, emotion_table])

                select_lang_metric.change(on_change, [languages_filter, select_lang_metric, dataset_filter, select_ds_metric, emo_filter],
                                       [overall_table, lang_table, dataset_table, emotion_table])

                dataset_filter.change(on_change, [languages_filter, select_lang_metric, dataset_filter, select_ds_metric, emo_filter],
                                       [overall_table, lang_table, dataset_table, emotion_table])

                select_ds_metric.change(on_change, [languages_filter, select_lang_metric, dataset_filter, select_ds_metric, emo_filter],
                                          [overall_table, lang_table, dataset_table, emotion_table])

                emo_filter.change(on_change, [languages_filter, select_lang_metric, dataset_filter, select_ds_metric, emo_filter],
                                       [overall_table, lang_table, dataset_table, emotion_table])
            
            with gr.Tab("📝 About", elem_classes='tab-item'):
                gr.Markdown(open("pages/about.md", "r").read(), elem_classes='tab-item')

            with gr.Tab("🔢 Evaluate your model", elem_classes='tab-item'):
                gr.Markdown(open("pages/evaluate.md", "r").read(), elem_classes='tab-item')
            
            with gr.Tab("🚀 Submit here!", elem_classes='tab-item'):
                gr.Markdown(open("pages/submit.md", "r").read(), elem_classes='tab-item')
        
        with gr.Column():
            with gr.Accordion("📙 Citation", open=False, elem_classes='tab-item'):
                citation_button = gr.Textbox(
                    label="",
                    value=CITATION_TEXT,
                    lines=20,
                    elem_id="citation-button",
                    show_copy_button=True,
                )
    
    return demo

if __name__ == "__main__":
    demo = app()
    demo.launch()