src/process_leaderboard_data.py

import pandas as pd
import gradio as gr


class Leaderboard():
    def __init__(self, data_path='results.csv'):
        self.data_path = data_path
        self.models_performance_dataframe = self.initialize_models_performance_dataframe()
        self.link_dict = self.initialize_link_dict()
        self.dataset_list = sorted(list(set(self.models_performance_dataframe['dataset'])))
        self.emotional_dataset_list = sorted(self.initialize_emotional_datasets())
        self.emotion_list = sorted(list(set(self.models_performance_dataframe['emotion'])))

    def get_dataset_list(self) -> list:
        return self.dataset_list
    
    def get_emotional_dataset_list(self) -> list:
        return self.emotional_dataset_list
    
    def get_emotion_list(self) -> list:
        return self.emotion_list

    def get_models_performance_dataframe_column(self) -> list:
        return self.models_performance_dataframe.columns

    # Load the models performance dataframe from a CSV file
    def initialize_models_performance_dataframe(self) -> pd.DataFrame:
        model_performance_dataframe = pd.read_csv(self.data_path)
        initial_columns = ['model', 'dataset', 'emotion', 'wavlm']
        columns = initial_columns + [kol for kol in model_performance_dataframe.columns if kol not in initial_columns]
        return model_performance_dataframe[columns]
    
    # Initialize a list of emotional datasets    
    def initialize_emotional_datasets(self) -> list:
        emotional_dataset_list = set()
        for _, row in self.models_performance_dataframe.iterrows():
            if row['emotion'] != 'All':
                emotional_dataset_list.add(row['dataset'])
        return list(emotional_dataset_list)
    
    # Dictionary mapping model names to their GitHub repository links
    def initialize_link_dict(self) -> dict:
        return {
            'WhisperSpeech/WhisperSpeech': 'https://huggingface.co/WhisperSpeech/WhisperSpeech',
            'microsoft/speecht5_vc': 'https://huggingface.co/microsoft/speecht5_vc',
            'Plachtaa/VALL-E-X': 'https://github.com/Plachtaa/VALL-E-X',
            'coqui/XTTS-v2': 'https://huggingface.co/coqui/XTTS-v2',
            'OuteAI/OuteTTS-0.2-500M': 'https://huggingface.co/OuteAI/OuteTTS-0.2-500M'
        }
        

    def create_leaderboard_data(self, selected_emotion_or_dataset, selected_feature, emotion_or_dataset) -> pd.DataFrame:
        '''
        Create leaderboard data.

        Args:
            selected_emotion_or_dataset (str): The selected emotion or dataset to filter the leaderboard data.
            selected_feature (str): The selected feature to display in the leaderboard.
            emotion_or_dataset (str): Indicates whether the selection is based on 'emotion' or 'dataset'.

        Returns:
            pd.DataFrame: A dataframe containing the leaderboard data. 
        '''
        # Create a copy of the dataframe.
        models_performance = self.models_performance_dataframe 

        # Determine the opposite selection emotion or dataset
        opposite_emotion_or_dataset = 'dataset' if emotion_or_dataset == 'emotion' else 'emotion'

        # Filter the dataframe based on the selected emotion or dataset.
        models_performance = models_performance[models_performance[emotion_or_dataset] == selected_emotion_or_dataset]

        # Creating a dictionary that aggregates information from the dataframe for each model.
        leaderboard_data = {}
        for _, row in models_performance.iterrows():
            if row['model'] not in leaderboard_data.keys():
                leaderboard_data[row['model']] ={}
            if row[opposite_emotion_or_dataset] == 'All':
                leaderboard_data[row['model']]['Average'] = row[selected_feature]
            else:
                leaderboard_data[row['model']][row[opposite_emotion_or_dataset]] = row[selected_feature]
            
        # Creating a dataframe based on leaderboard_data dictionary
        data = []
        for model, performance in leaderboard_data.items():
            row = {'Model': model}
            row.update(performance)
            data.append(row)
        leaderboard_dataframe = pd.DataFrame(data)
        
        # Ensure specific columns appear first in the dataframe.
        if 'LS Clean' in leaderboard_dataframe.keys():
            initial_columns = ['Model', 'Average', 'LS Clean']
        else:
            initial_columns = ['Model', 'Average']

        # Add other columns in sorted order.
        sorted_columns = initial_columns + sorted([col for col in leaderboard_dataframe.columns if col not in initial_columns])
        leaderboard_dataframe = leaderboard_dataframe[sorted_columns]

        # Round results
        for column_name in leaderboard_dataframe.columns:
            if column_name != "Model":
                leaderboard_dataframe[column_name] = leaderboard_dataframe[column_name].round(4)

        # Map model names to hyperlinks using the link_dict.
        leaderboard_dataframe['Model'] = leaderboard_dataframe['Model'].map(lambda model: f"[{model}]({self.link_dict.get(model, '')})")
        leaderboard_dataframe.sort_values(by='Average', ascending=False, inplace=True)
        leaderboard_dataframe.insert(0, "Rank", range(1, len(leaderboard_dataframe) + 1))
        return leaderboard_dataframe

    
    def update_leaderboard_data_in_emotion_section(self, selected_emotion_or_dataset, emotion_or_dataset, leaderboard_table) -> pd.DataFrame:    
        '''
        Update leaderboard data based on selected emotion or dataset in the emotion section

        Args:
            selected_emotion_or_dataset (str): The selected emotion or dataset to filter the leaderboard data.
            emotion_or_dataset (str): Indicates whether the selection is based on 'emotion' or 'dataset'.
            leaderboard_table (pd.Dataframe): previous leaderboard data.

        Returns:
            pd.DataFrame: A dataframe containing the leaderboard data.
        '''
        if selected_emotion_or_dataset != None:
            Leaderboard_dataframe = self.create_leaderboard_data(selected_emotion_or_dataset, 'wavlm', emotion_or_dataset)

            # Drop models without emotion-based division
            for dataset in self.dataset_list:
                if dataset in Leaderboard_dataframe.columns and dataset not in self.emotional_dataset_list:
                    Leaderboard_dataframe.drop(columns=[dataset], inplace=True)

            return gr.update(value=None), Leaderboard_dataframe
        else:
            return gr.update(), leaderboard_table 


    def update_leaderboard_data_in_feature_section(self, selected_emotion_or_dataset, selected_feature, emotion_or_dataset, leaderboard_table) -> pd.DataFrame:
        '''
        Update leaderboard data based on selected emotion or dataset in the feature section

        Args:
            selected_emotion_or_dataset (str): The selected emotion or dataset to filter the leaderboard data.
            selected_feature (str): The selected feature to display in the leaderboard.
            emotion_or_dataset (str): Indicates whether the selection is based on 'emotion' or 'dataset'.
            leaderboard_table (pd.Dataframe): previous leaderboard data.

        Returns:
            pd.DataFrame: A dataframe containing the leaderboard data.
        '''
        if selected_emotion_or_dataset != None:
            return gr.update(value=None), self.create_leaderboard_data(selected_emotion_or_dataset, selected_feature, emotion_or_dataset)
        else:
            return gr.update(), leaderboard_table
        

    def update_leaderboard_data_by_feature(self, emotion, dataset, selected_feature) -> pd.DataFrame:
        '''
        Update leaderboard data based on the selected feature

        Args:
            emotion (str): Currently selected emotion to filter the leaderboard data.
            dataset (str): Currently selected dataset to filter the leaderboard data
            selected_feature (str): The selected feature to display in the leaderboard.

        Returns:
            pd.DataFrame: A dataframe containing the leaderboard data.
        '''
        if emotion != None:
            return self.create_leaderboard_data(emotion, selected_feature, 'emotion')
        else:
            return self.create_leaderboard_data(dataset, selected_feature, 'dataset')