dataframe_comparison.py

import pandas as pd
import numpy as np
import os
import matplotlib.pyplot as plt
from fastdtw import fastdtw
from scipy.spatial.distance import euclidean
from itertools import combinations

# --- Configurações ---
DATA_DIR = 'data'
OUTPUT_DIR = 'comparativos_output'
os.makedirs(OUTPUT_DIR, exist_ok=True)

def get_all_2d_columns(df:pd.DataFrame) -> list[str]:
    return [col for col in df.columns if col.endswith(('_x', '_y'))]

def get_dtw_path(df1:pd.DataFrame, df2:pd.DataFrame, cols:list[str]):
    s1 = df1[cols].ffill().bfill().fillna(0).values
    s2 = df2[cols].ffill().bfill().fillna(0).values
    distance, path = fastdtw(s1, s2, dist=euclidean)
    return path, distance

def compare_pair(file1_info, file2_info):
    df1 = pd.read_parquet(file1_info['parquet'])
    df2 = pd.read_parquet(file2_info['parquet'])
    

    cols1 = set(get_all_2d_columns(df1))
    cols2 = set(get_all_2d_columns(df2))
    

    common_cols = sorted(list(cols1.intersection(cols2)))
    
    if not common_cols:
        print(f"⚠️ Pular: Nenhuma coluna em comum entre {file1_info['name']} e {file2_info['name']}")
        return None

    # Calculate DTW using only the shared points
    path, dist = get_dtw_path(df1, df2, common_cols)
    avg_dist = dist / len(path)
    
    pair_name = f"{file1_info['name']}_vs_{file2_info['name']}"
    print(f"Processando: {pair_name} | Pontos usados: {len(common_cols)//2} | Distância: {avg_dist:.4f}")
    
    return {"pair": pair_name, "distance": dist, "avg_distance": avg_dist}



if __name__ == "__main__":
    signals_map = []
    
    for signal_name in os.listdir(DATA_DIR):
        signal_path = os.path.join(DATA_DIR, signal_name)
        if os.path.isdir(signal_path):
            files = os.listdir(signal_path)
            parquets = [f for f in files if f.endswith('.parquet')]
            
            for p in parquets:
                base_name = p.replace('.parquet', '')
                video_file = base_name + '.mp4'
                if video_file in files:
                    signals_map.append({
                        'signal_group': signal_name,
                        'name': base_name,
                        'video': os.path.join(signal_path, video_file),
                        'parquet': os.path.join(signal_path, p)
                    })

    all_pairs = list(combinations(signals_map, 2))
    print(f"Total de sinais encontrados: {len(signals_map)}")
    print(f"Total de comparações a realizar: {len(all_pairs)}")

    results = []
    for f1, f2 in all_pairs:
        res = compare_pair(f1, f2)
        results.append(res)

    df_results = pd.DataFrame(results)
    df_results.to_csv(os.path.join(OUTPUT_DIR, 'ranking_similaridade.csv'), index=False)
    print("\nRanking de similaridade salvo em 'ranking_similaridade.csv'")