dipromats_evaluation_v2.py

import pandas as pd
import json
import numpy as np
import warnings
# Suprimir SettingWithCopyWarning
warnings.simplefilter(action='ignore', category=pd.errors.SettingWithCopyWarning)

def evaluate_results(lang, gold, test):
    def normalize_labels(df):
        # Define a function that checks if each narrative is present and assigns "yes" or "no"
        def convert_narratives(row):
            country_code = row['country'][:2].upper()  # Get the country code ('RU', 'CH', etc.)
            narratives = row['narratives']  # List of narratives for that row

            # For each N1 to N6, check if it appears in the list of narratives
            for i in range(1, 7):
                narrative_code = f"{country_code}{i}"
                row[f"N{i}"] = 'yes' if narrative_code in narratives else 'no'
            return row  
        # Apply the function to each row of the DataFrame
        data = df.apply(convert_narratives, axis=1) 
        # Drop the original 'narratives' column if no longer needed
        data.drop(columns=['narratives', 'tweet_id'], inplace=True)
        return data 
    def get_gold_lists_for_evaluation(gold_list, test_list):
        gold_strict=[]
        gold_lenient=[]
        for i in range(0,6):
            g=gold_list[i]
            t=test_list[i]
            g = 1 if g == 'yes' else 2 if g == 'no' else g
            t = 1 if t == 'yes' else 2 if t == 'no' else t
            if g==t:
                gold_strict.append(g)
                gold_lenient.append(g)
            elif g!=t:
                if g in [2, 1]:
                    gold_strict.append(g)
                    gold_lenient.append(g)
                else:
                    gold_strict.append(2)
                    gold_lenient.append(t)
        return gold_strict, gold_lenient    
    def gen_dic(lang):
        narratives_list=['CH1', 'CH2', 'CH3', 'CH4', 'CH5', 'CH6', 'CH_micro', 'RU1', 'RU2', 'RU3', 'RU4', 'RU5', 'RU6', 'RU_micro', 'EU1', 'EU2', 'EU3', 'EU4', 'EU5', 'EU6', 'EU_micro', 'US1', 'US2', 'US3', 'US4', 'US5', 'US6', 'US_micro']
        countries_dic={'China':'CH', 'Russia':'RU', 'EU':'EU', 'USA':'US'}
        dic = {}
        dic[lang] = {}  
        for ev in ['strict', 'lenient']:
            if ev not in dic[lang]:
                dic[lang][ev] = {}  
            for narr in narratives_list:
                dic[lang][ev][narr] = {'scores': {'precision': 0., 'recall': 0., 'f1-score': 0.}, 'raw_data': []}

            for code in countries_dic.values():
                dic[lang][ev][f'{code}_micro'] = {'scores': {'precision': 0., 'recall': 0., 'f1-score': 0}, 'raw_data': []}

            dic[lang][ev]['micro'] = {'scores': {'precision': 0., 'recall': 0., 'f1-score': 0}, 'raw_data': []}
        return dic  
    def convert_labels(values):
        return np.array([
            [1 if v == 'yes' else 2 if v == 'no' else 3 for v in row]
            for row in values
        ])  
    def convert_floats(dic):
        for key, value in dic.items():
            if isinstance(value, np.float64):
                dic[key] = float(value)
            elif isinstance(value, dict):  # If the value is another dictionary, apply recursion
                convert_floats(value)
            elif isinstance(value, list):  # If the value is a list, convert individual elements
                dic[key] = [float(v) if isinstance(v, np.float64) else v for v in value]    
    dic=gen_dic(lang)
    countries_dic={'China':'CH', 'Russia':'RU', 'EU':'EU', 'USA':'US'}
    cols=[f'N{i}' for i in range(1,7)]
    
    df_gold=pd.DataFrame(gold)
    df_gold["country"] = df_gold["country"].replace("European Union", "EU")
    df_gold.drop_duplicates(subset=['id', 'lang'], keep='last', inplace=True)
    df=df_gold[df_gold['lang']==lang]
    df.reset_index(inplace=True, drop=True)
    
    df_test=pd.DataFrame(test)
    df_test["country"] = df_test["country"].replace("European Union", "EU")
    df_test=normalize_labels(df_test)
    df_test.drop_duplicates(subset=['id', 'language'], keep='last', inplace=True)
    df_test.reset_index(inplace=True, drop=True)
    
    df_strict=df.copy()
    df_lenient=df.copy()
    for i in range(len(df)):
        lang=df['lang'].iloc[i] 
        id=df['id'].iloc[i]
        gold_values=df[cols].iloc[i].values
        dft=df_test[(df_test['language']==lang) & (df_test['id']==id)]

        test_values=dft[cols].iloc[0].values    
        df_strict.loc[i, cols], df_lenient.loc[i, cols]=get_gold_lists_for_evaluation(gold_values, test_values)

    countries=['China', 'Russia', 'EU', 'USA']

    df_lang=df[(df['lang']==lang)]
    df_test_lang=df_test[(df_test['language']==lang)]
    df_strict_lang=df_strict[df_strict['lang']==lang]
    df_lenient_lang=df_lenient[df_lenient['lang']==lang]
    #F1 per narrative
    for country in countries:
        df_dup_t=df[(df['country']==country) & (df['lang']==lang)]
        df_strict_t=df_strict_lang[df_strict_lang['country']==country]
        df_lenient_t=df_lenient_lang[df_lenient_lang['country']==country]
        dft=df_test_lang[(df_test_lang['country']==country)]
        real_strict=[]
        real_lenient=[]
        real=[]
        pred=[]
        for i in range(len(df_strict_t)):
            id=df_strict_t['id'].iloc[i]
            dft2=dft[dft['id']==id]
            if len(dft2)!=0:
                real_strict.append(df_strict_t[cols].iloc[i].values)
                real_lenient.append(df_lenient_t[cols].iloc[i].values)
                pred.append(dft2[cols].iloc[0].values)
                real.append(df_dup_t[df_dup_t['id']==id][cols].iloc[0].values)
        real_strict=np.array(real_strict)
        real_lenient=np.array(real_lenient)
    
        real = convert_labels(real)
        pred = convert_labels(pred)

        for i in range(0, 6):
            raw_matrix = np.zeros((2, 3), dtype=int)  # 2 filas (pred), 3 columnas (real)
            pred_options = [1, 2]  # 1 -> 'yes', 2 -> 'no'
            real_options = [1, 3, 2]  # 1 
            p=pred[:,i]
            r=real[:,i]
            for p, r in zip(p, r):
                pred_index = pred_options.index(p)  
                real_index = real_options.index(r)  
                raw_matrix[pred_index, real_index] += 1 
            tp=raw_matrix[0,0]
            yl=raw_matrix[0,1]
            fp=raw_matrix[0,2]
            fn=raw_matrix[1,0]
            nl=raw_matrix[1,1]
            tn=raw_matrix[1,2]
            dic[lang]['lenient'][f'{countries_dic[country]}{i+1}']['raw_data']=raw_matrix.tolist()
            precision=(tp+yl)/(tp+yl+fp) if (tp+yl+fp)!=0 else 0
            recall=(tp+yl)/(tp+fn+yl) if (tp+fn+yl)!=0 else 0
            dic[lang]['lenient'][f'{countries_dic[country]}{i+1}']['scores']['precision']=precision
            dic[lang]['lenient'][f'{countries_dic[country]}{i+1}']['scores']['recall']=recall
            dic[lang]['lenient'][f'{countries_dic[country]}{i+1}']['scores']['f1-score']=(2*precision*recall)/(precision+recall) if (precision+recall)!=0 else 0    
            dic[lang]['strict'][f'{countries_dic[country]}{i+1}']['raw_data']=raw_matrix.tolist()
            precision=tp/(tp+fp+yl)  if (tp+fp+yl)!=0 else 0
            recall=tp/(tp+fn) if (tp+fn)!=0 else 0
            dic[lang]['strict'][f'{countries_dic[country]}{i+1}']['scores']['precision']=precision
            dic[lang]['strict'][f'{countries_dic[country]}{i+1}']['scores']['recall']=recall
            dic[lang]['strict'][f'{countries_dic[country]}{i+1}']['scores']['f1-score']=(2*precision*recall)/(precision+recall) if (precision+recall)!=0 else 0 

        #F1 Micro
        real_strict=[]
        real_lenient=[]
        pred=[]
        not_match=[]
        real=[]
        for i in range(len(df_lang)):
            id=df_lang['id'].iloc[i]
            dft=df_test_lang[df_test_lang['id']==id][cols]
            if len(dft)!=0:
                real_strict.extend(df_strict_lang[cols].iloc[i].values)
                real_lenient.extend(df_strict_lang[cols].iloc[i].values)
                pred.extend(df_test_lang[df_test_lang['id']==id][cols].iloc[0].values)
                real.extend(df_lang[df_lang['id']==id][cols].iloc[0].values)
            else:
                not_match.append(id)

        real = convert_labels([real])[0]
        pred = convert_labels([pred])[0]
        raw_matrix=np.zeros((2,3), dtype=int)
        pred_options = [1, 2]  # 1 -> 'yes', 2 -> 'no'
        real_options = [1, 3, 2]  # 1 
        raw_matrix = np.zeros((2, 3), dtype=int)
        for p, r in zip(pred, real):
            pred_index = pred_options.index(p)  
            real_index = real_options.index(r)  
            raw_matrix[pred_index, real_index] += 1     
        tp=raw_matrix[0,0]
        yl=raw_matrix[0,1]
        fp=raw_matrix[0,2]
        fn=raw_matrix[1,0]
        nl=raw_matrix[1,1]
        tn=raw_matrix[1,2]
        dic[lang]['lenient']['micro']['raw_data']=raw_matrix.tolist()
        precision=(tp+yl)/(tp+yl+fp) if (tp+yl+fp)!=0 else 0
        recall=(tp+yl)/(tp+fn+yl) if (tp+fn+yl)!=0 else 0
        dic[lang]['lenient']['micro']['scores']['precision']=precision
        dic[lang]['lenient']['micro']['scores']['recall']=recall
        dic[lang]['lenient']['micro']['scores']['f1-score']=(2*precision*recall)/(precision+recall) if (precision+recall)!=0 else 0 
        dic[lang]['strict']['micro']['raw_data']=raw_matrix.tolist()
        precision=tp/(tp+fp+yl) if (tp+yl+fp)!=0 else 0
        recall=tp/(tp+fn) if (tp+fn)!=0 else 0
        dic[lang]['strict']['micro']['scores']['precision']=precision
        dic[lang]['strict']['micro']['scores']['recall']=recall
        dic[lang]['strict']['micro']['scores']['f1-score']=(2*precision*recall)/(precision+recall) if (precision+recall)!=0 else 0

    #Micro-Countries
    for country in countries_dic.values():
        raw_matrix = np.sum([np.array(dic[f'{lang}']['strict'][f'{country}{i}']['raw_data']) for i in range(1, 7)], axis=0)
        tp=raw_matrix[0,0]
        yl=raw_matrix[0,1]
        fp=raw_matrix[0,2]
        fn=raw_matrix[1,0]
        nl=raw_matrix[1,1]
        tn=raw_matrix[1,2]  
        precision=(tp+yl)/(tp+yl+fp) if (tp+yl+fp)!=0 else 0
        recall=(tp+yl)/(tp+fn+yl) if (tp+fn+yl)!=0 else 0
        dic[lang]['lenient'][f'{country}_micro']['scores']['precision']=precision
        dic[lang]['lenient'][f'{country}_micro']['scores']['recall']=recall
        dic[lang]['lenient'][f'{country}_micro']['scores']['f1-score']=(2*precision*recall)/(precision+recall) if (precision+recall)!=0 else 0
        dic[lang]['lenient'][f'{country}_micro']['raw_data']=raw_matrix.tolist()    
        precision=tp/(tp+fp+yl) if (tp+yl+fp)!=0 else 0
        recall=tp/(tp+fn) if (tp+fn)!=0 else 0
        dic[lang]['strict'][f'{country}_micro']['scores']['precision']=precision
        dic[lang]['strict'][f'{country}_micro']['scores']['recall']=recall
        dic[lang]['strict'][f'{country}_micro']['scores']['f1-score']=(2*precision*recall)/(precision+recall) if (precision+recall)!=0 else 0
        dic[lang]['strict'][f'{country}_micro']['raw_data']=raw_matrix.tolist()

    convert_floats(dic[lang])

    return dic[lang]

"""
strict
  narrative_country (e.g. CH1)
    scores
      precision
      recall
      f1-score
    raw_data
  country_micro (e.g. CH_micro)
    scores
      precision
      recall
      f1-score
    raw_data
  micro (global micro)
    scores
      precision
      recall
      f1-score
    raw_data

lenient
  narrative_country (e.g. CH1)
    scores
      precision
      recall
      f1-score
    raw_data
  country_micro (e.g. CH_micro)
    scores
      precision
      recall
      f1-score
    raw_data
  micro (global micro)
    scores
      precision
      recall
      f1-score
    raw_data"""