benchmark_evaluation.py

"""

File: benchmark_evaluation.py

------------------------------

Benchmark E. coli protein sequences with ENCOT, generate optimized DNA,

compute metrics (CAI, tAI, GC, CFD, cis-elements), and produce summary tables

and figures.

"""

import sys
import os
import argparse
import pandas as pd
import numpy as np
import torch
import json
import matplotlib.pyplot as plt
import seaborn as sns
from datetime import datetime
import time
from tqdm import tqdm
from typing import Dict, List, Tuple, Any

from CAI import CAI, relative_adaptiveness
from CodonTransformer.CodonData import (
    download_codon_frequencies_from_kazusa,
    get_codon_frequencies,
)
from CodonTransformer.CodonPrediction import (
    load_model,
    predict_dna_sequence,
)
from CodonTransformer.CodonEvaluation import (
    get_GC_content,
    get_ecoli_tai_weights,
    get_min_max_profile,
    calculate_tAI,
    count_negative_cis_elements,
)
from transformers import AutoTokenizer
sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
from evaluate_optimizer import translate_dna_to_protein


def find_longest_orf(dna_sequence: str) -> str:
    """

    Find the longest open reading frame (ORF) in a DNA sequence.



    Args:

        dna_sequence (str): Input DNA sequence (ATCGN characters).



    Returns:

        str: Longest ORF (from start to stop codon), or empty string if none.

    """
    dna_sequence = dna_sequence.upper()
    start_codons = ['ATG']
    stop_codons = ['TAA', 'TAG', 'TGA']
    
    longest_orf = ""
    
    for frame in range(3):
        current_orf = ""
        in_orf = False
        
        for i in range(frame, len(dna_sequence) - 2, 3):
            codon = dna_sequence[i:i+3]
            if len(codon) != 3:
                break
                
            if codon in start_codons and not in_orf:
                in_orf = True
                current_orf = codon
            elif in_orf:
                current_orf += codon
                if codon in stop_codons:
                    if len(current_orf) > len(longest_orf):
                        longest_orf = current_orf
                    in_orf = False
                    current_orf = ""

        if in_orf and len(current_orf) > len(longest_orf):
            longest_orf = current_orf
    
    return longest_orf


def _detect_columns(df: pd.DataFrame, name_hint: str | None = None, seq_hint: str | None = None) -> tuple[str | None, str]:
    """

    Detect name and sequence columns in a case-insensitive, robust way.



    Args:

        df (pd.DataFrame): Input DataFrame read from Excel.

        name_hint (str | None): Optional override for name/label column (case-insensitive).

        seq_hint (str | None): Optional override for sequence column (case-insensitive).



    Returns:

        tuple[str | None, str]: Detected (name_column or None, sequence_column).



    Raises:

        ValueError: If a sequence-like column cannot be found.

    """
    cols = list(df.columns)
    low_map = {c.lower().strip(): c for c in cols}

    # If hints are provided and exist (case-insensitive), honor them
    if name_hint:
        nh = name_hint.lower().strip()
        if nh in low_map:
            name_col = low_map[nh]
        else:
            name_col = None
    else:
        name_col = None

    if seq_hint:
        sh = seq_hint.lower().strip()
        if sh in low_map:
            seq_col = low_map[sh]
        else:
            seq_col = None
    else:
        seq_col = None

    # If not found, try candidates
    if name_col is None:
        name_candidates = [
            'name','id','title','gene','protein','description','label','accession','locus','entry','uniprot','ncbi','protein name'
        ]
        for k in name_candidates:
            if k in low_map:
                name_col = low_map[k]
                break

    if seq_col is None:
        seq_candidates = [
            # protein-first
            'protein sequence','protein_sequence','protein','aa sequence','aa_sequence','aa','amino acid sequence','amino_acid_sequence',
            # generic
            'sequence','seq',
            # dna/cds
            'cds','dna','coding sequence','coding_sequence','cds sequence','cds_sequence'
        ]
        for k in seq_candidates:
            if k in low_map:
                seq_col = low_map[k]
                break

    if not seq_col:
        raise ValueError(f"Could not detect sequence column. Available columns: {cols}")

    return name_col, seq_col


def parse_excel_sequences(excel_path: str, name_col: str | None = None, seq_col: str | None = None, sheet_name: str | int | None = None) -> List[Dict[str, str]]:
    """

    Parse sequences from the benchmark Excel file and auto-detect relevant columns.



    Args:

        excel_path (str): Path to the Excel file.

        name_col (str | None): Optional override for sequence name column.

        seq_col (str | None): Optional override for sequence column.

        sheet_name (str | int | None): Sheet name or index (default: first sheet).



    Returns:

        List[Dict[str, str]]: List of standardized sequence records with fields:

            id, name, protein_sequence, original_sequence (DNA or None), is_dna.



    Raises:

        ValueError: If a sequence column cannot be detected.

    """
    sn = sheet_name
    if isinstance(sn, str) and sn.isdigit():
        sn = int(sn)
    if sn is None:
        sn = 0

    df_or_dict = pd.read_excel(excel_path, sheet_name=sn)
    if isinstance(df_or_dict, dict):
        first_title, df = next(iter(df_or_dict.items()))
        print(f"Using sheet: {first_title}")
    else:
        df = df_or_dict
    sequences = []

    detected_name_col, detected_seq_col = _detect_columns(df, name_col, seq_col)
    print(f"Detected columns -> name: {detected_name_col or '[generated]'}, sequence: {detected_seq_col}")

    for idx, row in df.iterrows():
        sequence = str(row[detected_seq_col]).strip()
        if detected_name_col:
            name = str(row[detected_name_col]).strip()
        else:
            name = f"seq_{idx}"

        if name.startswith('>'):
            name = name[1:].strip()

        sequence = ''.join(filter(str.isalpha, sequence))

        dna_chars = sum(1 for c in sequence.upper() if c in 'ATCGN')
        is_dna = (dna_chars / len(sequence)) > 0.95 if len(sequence) > 0 else False

        if is_dna:
            longest_orf = find_longest_orf(sequence)

            if longest_orf and len(longest_orf) >= 30:
                original_dna = longest_orf
                protein_seq = translate_dna_to_protein(longest_orf)
            else:
                truncated_len = (len(sequence) // 3) * 3
                if truncated_len >= 30:
                    original_dna = sequence[:truncated_len]
                    protein_seq = translate_dna_to_protein(original_dna)
                else:
                    continue

            if '*' in protein_seq:
                stop_pos = protein_seq.find('*')
                if stop_pos >= 10:
                    protein_seq = protein_seq[:stop_pos]
                    original_dna = original_dna[:stop_pos*3]
                else:
                    continue

        else:
            protein_seq = sequence.upper()
            protein_seq = protein_seq.replace('*', '')
            original_dna = None

        if len(protein_seq) < 10:
            continue
        
        sequences.append({
            'id': idx,
            'name': name,
            'protein_sequence': protein_seq,
            'original_sequence': original_dna,
            'is_dna': is_dna
        })
    
    return sequences


def calculate_cfd(dna_sequence: str, codon_frequencies: Dict) -> float:
    """

    Calculate Codon Frequency Distribution (CFD) similarity to a reference.



    Args:

        dna_sequence (str): Input DNA sequence.

        codon_frequencies (Dict): Reference frequencies; accepts flattened mapping

            or an amino2codon structure (will be flattened).



    Returns:

        float: Similarity score in [0, 1] where higher is more similar.

    """
    if not dna_sequence:
        return 0.0

    codon_count = {}
    total_codons = 0

    for i in range(0, len(dna_sequence) - 2, 3):
        codon = dna_sequence[i:i+3].upper()
        if len(codon) == 3:
            codon_count[codon] = codon_count.get(codon, 0) + 1
            total_codons += 1

    seq_freq = {}
    if total_codons > 0:
        for codon, count in codon_count.items():
            seq_freq[codon] = count / total_codons

    # Flatten amino2codon frequencies if needed
    flat_codon_freq = {}
    if isinstance(codon_frequencies, dict):
        first_key = next(iter(codon_frequencies.keys()))
        if isinstance(codon_frequencies[first_key], tuple) and len(codon_frequencies[first_key]) == 2:
            for amino, (codons, freqs) in codon_frequencies.items():
                for codon, freq in zip(codons, freqs):
                    flat_codon_freq[codon] = freq
        else:
            flat_codon_freq = codon_frequencies

    similarity = 0.0
    count = 0

    for codon in set(list(seq_freq.keys()) + list(flat_codon_freq.keys())):
        seq_f = seq_freq.get(codon, 0.0)
        ref_f = flat_codon_freq.get(codon, 0.0)
        similarity += 1 - abs(seq_f - ref_f)
        count += 1

    return similarity / count if count > 0 else 0.0


def run_model_on_sequences(

    sequences: List[Dict],

    model,

    tokenizer,

    device,

    cai_weights: Dict,

    tai_weights: Dict,

    codon_frequencies: Dict,

    reference_profile: List[float],

    output_dir: str

) -> pd.DataFrame:
    """

    Run ColiFormer on protein sequences and compute metrics for optimized DNA.



    Args:

        sequences (List[Dict]): Parsed sequence records.

        model: Loaded ColiFormer model.

        tokenizer: Tokenizer used by the model.

        device: Torch device.

        cai_weights (Dict): CAI weights.

        tai_weights (Dict): tAI weights.

        codon_frequencies (Dict): Reference codon frequencies.

        reference_profile (List[float]): Reserved for DTW profile (unused here).

        output_dir (str): Directory for outputs (not written here).



    Returns:

        pd.DataFrame: Per-sequence metrics and optimized DNA.

    """
    results = []
    print(f"Processing {len(sequences)} sequences...")

    for seq_data in tqdm(sequences, desc="Optimizing sequences"):
        protein_seq = seq_data['protein_sequence']

        if len(protein_seq) < 10:
            continue

        try:
            start_time = time.time()

            output = predict_dna_sequence(
                protein=protein_seq,
                organism="Escherichia coli general",
                device=device,
                model=model,
                deterministic=True,
                match_protein=True,
            )

            runtime = time.time() - start_time

            if isinstance(output, list):
                optimized_dna = output[0].predicted_dna
            else:
                optimized_dna = output.predicted_dna

            original_metrics = {}
            if seq_data['is_dna'] and seq_data['original_sequence']:
                original_dna = seq_data['original_sequence'].upper()
                original_metrics = {
                    'original_cai': CAI(original_dna, weights=cai_weights),
                    'original_gc': get_GC_content(original_dna),
                    'original_tai': calculate_tAI(original_dna, tai_weights),
                    'original_cfd': calculate_cfd(original_dna, codon_frequencies),
                    'original_neg_cis': count_negative_cis_elements(original_dna),
                }

            optimized_metrics = {
                'optimized_cai': CAI(optimized_dna, weights=cai_weights),
                'optimized_gc': get_GC_content(optimized_dna),
                'optimized_tai': calculate_tAI(optimized_dna, tai_weights),
                'optimized_cfd': calculate_cfd(optimized_dna, codon_frequencies),
                'optimized_neg_cis': count_negative_cis_elements(optimized_dna),
                'runtime': runtime,
            }

            result = {
                'id': seq_data['id'],
                'name': seq_data['name'],
                'protein_sequence': protein_seq,
                'protein_length': len(protein_seq),
                'optimized_dna': optimized_dna,
                **original_metrics,
                **optimized_metrics,
            }
            results.append(result)

        except Exception as e:
            print(f"Error processing sequence {seq_data['id']}: {str(e)}")
            continue

    return pd.DataFrame(results)


def generate_visualizations(results_df: pd.DataFrame, output_dir: str):
    """

    Generate visualizations and a metrics summary table.



    Saves:

        - CAI before/after bar plot

        - Median CAI comparison

        - Metrics distribution panel

        - CSV summary table



    Args:

        results_df (pd.DataFrame): Results from optimization.

        output_dir (str): Output directory root.



    Returns:

        pd.DataFrame: Summary table of aggregate metrics.

    """
    plt.style.use('seaborn-v0_8-darkgrid')
    sns.set_palette("husl")

    fig_dir = os.path.join(output_dir, 'figures')
    os.makedirs(fig_dir, exist_ok=True)

    # 1. Before/After CAI Graph
    if 'original_cai' in results_df.columns:
        plt.figure(figsize=(12, 8))

        before_cai = results_df['original_cai'].dropna()
        after_cai = results_df.loc[before_cai.index, 'optimized_cai']

        x = np.arange(len(before_cai))
        width = 0.35

        fig, ax = plt.subplots(figsize=(14, 8))
        bars1 = ax.bar(x - width/2, before_cai, width, label='Before Optimization', alpha=0.8)
        bars2 = ax.bar(x + width/2, after_cai, width, label='After Optimization', alpha=0.8)

        ax.set_xlabel('Sequence Index', fontsize=12)
        ax.set_ylabel('CAI Score', fontsize=12)
        ax.set_title('ENCOT: CAI Before and After Optimization', fontsize=14, fontweight='bold')
        ax.set_xticks(x[::5])  # Show every 5th label
        ax.set_xticklabels(x[::5])
        ax.legend()
        ax.grid(axis='y', alpha=0.3)

        avg_before = before_cai.mean()
        avg_after = after_cai.mean()
        improvement = ((avg_after - avg_before) / avg_before) * 100

        ax.text(0.02, 0.98, f'Average CAI Before: {avg_before:.3f}\nAverage CAI After: {avg_after:.3f}\nImprovement: {improvement:.1f}%',
                transform=ax.transAxes, fontsize=10, verticalalignment='top',
                bbox=dict(boxstyle='round', facecolor='wheat', alpha=0.5))

        plt.tight_layout()
        plt.savefig(os.path.join(fig_dir, 'cai_before_after.png'), dpi=300, bbox_inches='tight')
        plt.close()

        print(f"CAI Before/After graph saved to {os.path.join(fig_dir, 'cai_before_after.png')}")

        # 1b. Median CAI Before/After Graph
        plt.figure(figsize=(8, 6))

        median_before = before_cai.median()
        median_after = after_cai.median()

        categories = ['Before Optimization', 'After Optimization']
        medians = [median_before, median_after]
        colors = ['#ff7f0e', '#2ca02c']

        bars = plt.bar(categories, medians, color=colors, alpha=0.8, width=0.6)
        plt.ylabel('Median CAI Score', fontsize=12)
        plt.title('ENCOT: Median CAI Before and After Optimization', fontsize=14, fontweight='bold')
        plt.ylim(0, max(medians) * 1.2)

        for bar, median in zip(bars, medians):
            plt.text(bar.get_x() + bar.get_width()/2, bar.get_height() + 0.01, 
                    f'{median:.3f}', ha='center', va='bottom', fontweight='bold')

        improvement_pct = ((median_after - median_before) / median_before) * 100
        plt.text(0.5, max(medians) * 0.95, f'Improvement: {improvement_pct:.1f}%', 
                ha='center', transform=plt.gca().transData, fontsize=12, 
                bbox=dict(boxstyle='round', facecolor='lightblue', alpha=0.7))

        plt.grid(axis='y', alpha=0.3)
        plt.tight_layout()
        plt.savefig(os.path.join(fig_dir, 'median_cai_comparison.png'), dpi=300, bbox_inches='tight')
        plt.close()

        print(f"Median CAI comparison graph saved to {os.path.join(fig_dir, 'median_cai_comparison.png')}")
    
    # 2. Summary metrics table
    metrics_summary = {}

    if 'original_cai' in results_df.columns:
        metrics_summary['CAI'] = {
            'Before': results_df['original_cai'].mean(),
            'After': results_df['optimized_cai'].mean(),
            'Improvement': ((results_df['optimized_cai'].mean() - results_df['original_cai'].mean()) / results_df['original_cai'].mean()) * 100
        }
        metrics_summary['GC Content (%)'] = {
            'Before': results_df['original_gc'].mean(),
            'After': results_df['optimized_gc'].mean(),
            'Difference': results_df['optimized_gc'].mean() - results_df['original_gc'].mean()
        }
        metrics_summary['tAI'] = {
            'Before': results_df['original_tai'].mean(),
            'After': results_df['optimized_tai'].mean(),
            'Improvement': ((results_df['optimized_tai'].mean() - results_df['original_tai'].mean()) / results_df['original_tai'].mean()) * 100
        }
        metrics_summary['CFD'] = {
            'Before': results_df['original_cfd'].mean(),
            'After': results_df['optimized_cfd'].mean(),
            'Improvement': ((results_df['optimized_cfd'].mean() - results_df['original_cfd'].mean()) / results_df['original_cfd'].mean()) * 100
        }
        metrics_summary['Negative Cis Elements'] = {
            'Before': results_df['original_neg_cis'].mean(),
            'After': results_df['optimized_neg_cis'].mean(),
            'Reduction': results_df['original_neg_cis'].mean() - results_df['optimized_neg_cis'].mean()
        }
    else:
        metrics_summary['CAI'] = {
            'Optimized': results_df['optimized_cai'].mean(),
            'Std Dev': results_df['optimized_cai'].std()
        }
        metrics_summary['GC Content (%)'] = {
            'Optimized': results_df['optimized_gc'].mean(),
            'Std Dev': results_df['optimized_gc'].std()
        }
        metrics_summary['tAI'] = {
            'Optimized': results_df['optimized_tai'].mean(),
            'Std Dev': results_df['optimized_tai'].std()
        }
        metrics_summary['CFD'] = {
            'Optimized': results_df['optimized_cfd'].mean(),
            'Std Dev': results_df['optimized_cfd'].std()
        }
        metrics_summary['Negative Cis Elements'] = {
            'Optimized': results_df['optimized_neg_cis'].mean(),
            'Std Dev': results_df['optimized_neg_cis'].std()
        }

    metrics_summary['Runtime (seconds)'] = {
        'Mean': results_df['runtime'].mean(),
        'Median': results_df['runtime'].median(),
        'Total': results_df['runtime'].sum()
    }
    
    summary_df = pd.DataFrame(metrics_summary).T
    summary_df = summary_df.round(4)
    
    summary_df.to_csv(os.path.join(output_dir, 'metrics_summary.csv'))
    print(f"\nMetrics Summary saved to {os.path.join(output_dir, 'metrics_summary.csv')}")
    print("\n" + "="*60)
    print("METRICS SUMMARY:")
    print("="*60)
    print(summary_df.to_string())
    
    fig, axes = plt.subplots(2, 3, figsize=(15, 10))
    axes = axes.flatten()
    
    metrics_to_plot = [
        ('optimized_cai', 'CAI Distribution'),
        ('optimized_gc', 'GC Content Distribution (%)'),
        ('optimized_tai', 'tAI Distribution'),
        ('optimized_cfd', 'CFD Distribution'),
        ('optimized_neg_cis', 'Negative Cis Elements'),
        ('runtime', 'Runtime Distribution (seconds)')
    ]
    
    for idx, (col, title) in enumerate(metrics_to_plot):
        if col in results_df.columns:
            axes[idx].hist(results_df[col].dropna(), bins=20, edgecolor='black', alpha=0.7)
            axes[idx].set_title(title, fontsize=10, fontweight='bold')
            axes[idx].set_xlabel(col.replace('optimized_', '').replace('_', ' ').title())
            axes[idx].set_ylabel('Frequency')
            axes[idx].grid(axis='y', alpha=0.3)
            
            mean_val = results_df[col].mean()
            axes[idx].axvline(mean_val, color='red', linestyle='--', linewidth=2, label=f'Mean: {mean_val:.3f}')
            axes[idx].legend()
    
    plt.suptitle('ENCOT: Optimization Metrics Distribution', fontsize=14, fontweight='bold', y=1.02)
    plt.tight_layout()
    plt.savefig(os.path.join(fig_dir, 'metrics_distribution.png'), dpi=300, bbox_inches='tight')
    plt.close()
    
    print(f"Metrics distribution plot saved to {os.path.join(fig_dir, 'metrics_distribution.png')}")
    
    return summary_df


def main():
    """CLI entrypoint to run the ENCOT benchmark workflow."""
    parser = argparse.ArgumentParser(description="Benchmark ENCOT on E. coli sequences")
    parser.add_argument("--excel_path", type=str, default="Benchmark 80 sequences.xlsx",
                        help="Path to benchmark Excel file")
    parser.add_argument("--checkpoint_path", type=str, default="models/ecoli-codon-optimizer/finetune_best.ckpt",
                        help="Path to fine-tuned model checkpoint")
    parser.add_argument("--natural_sequences_path", type=str, default="data/ecoli_processed_genes.csv",
                        help="Path to natural E. coli sequences for CAI calculation")
    parser.add_argument("--output_dir", type=str, default="benchmark_results",
                        help="Directory to save results")
    parser.add_argument("--use_gpu", action="store_true", help="Use GPU if available")
    parser.add_argument("--name_col", type=str, default=None, help="Optional: column name for sequence label (case-insensitive)")
    parser.add_argument("--seq_col", type=str, default=None, help="Optional: column name for sequence (case-insensitive)")
    parser.add_argument("--sheet_name", type=str, default=None, help="Optional: Excel sheet name or index")
    
    args = parser.parse_args()
    
    timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
    output_dir = os.path.join(args.output_dir, f"run_{timestamp}")
    os.makedirs(output_dir, exist_ok=True)
    
    print("="*60)
    print("ENCOT BENCHMARK EVALUATION")
    print("="*60)
    
    device = torch.device("cuda" if torch.cuda.is_available() and args.use_gpu else "cpu")
    print(f"Using device: {device}")
    
    print(f"\nLoading sequences from {args.excel_path}...")
    sequences = parse_excel_sequences(
        args.excel_path,
        name_col=args.name_col,
        seq_col=args.seq_col,
        sheet_name=args.sheet_name,
    )
    print(f"Loaded {len(sequences)} sequences")
    
    print("\nLoading ENCOT model...")
    model = load_model(model_path=args.checkpoint_path, device=device)
    tokenizer = AutoTokenizer.from_pretrained("adibvafa/CodonTransformer")
    print("Model loaded successfully")
    
    print("\nPreparing evaluation utilities...")
    
    natural_df = pd.read_csv(args.natural_sequences_path)
    ref_sequences = natural_df['dna_sequence'].tolist()
    cai_weights = relative_adaptiveness(sequences=ref_sequences)
    print("CAI weights generated")
    
    tai_weights = get_ecoli_tai_weights()
    print("tAI weights loaded")
    
    try:
        codon_frequencies = download_codon_frequencies_from_kazusa(taxonomy_id=83333)
        print("Codon frequencies loaded from Kazusa")
    except Exception as e:
        print(f"Warning: Kazusa download failed ({e}). Using local frequencies.")
        codon_frequencies = get_codon_frequencies(
            ref_sequences, organism="Escherichia coli general"
        )
    
    reference_profile = []

    print("\n" + "="*60)
    print("RUNNING OPTIMIZATION...")
    print("="*60)
    
    results_df = run_model_on_sequences(
        sequences=sequences,
        model=model,
        tokenizer=tokenizer,
        device=device,
        cai_weights=cai_weights,
        tai_weights=tai_weights,
        codon_frequencies=codon_frequencies,
        reference_profile=reference_profile,
        output_dir=output_dir
    )
    
    results_path = os.path.join(output_dir, 'optimization_results.csv')
    results_df.to_csv(results_path, index=False)
    print(f"\nRaw results saved to {results_path}")
    
    optimized_sequences = results_df[['id', 'name', 'protein_sequence', 'optimized_dna']].copy()
    optimized_sequences['protein_length'] = results_df['protein_length']
    optimized_sequences['dna_length'] = optimized_sequences['optimized_dna'].apply(len)
    optimized_sequences['optimized_cai'] = results_df['optimized_cai']
    optimized_sequences['optimized_gc'] = results_df['optimized_gc']
    optimized_sequences['optimized_tai'] = results_df['optimized_tai']
    
    if 'original_cai' in results_df.columns:
        optimized_sequences['original_cai'] = results_df['original_cai']
        optimized_sequences['cai_improvement'] = ((results_df['optimized_cai'] - results_df['original_cai']) / results_df['original_cai'] * 100).round(2)
    
    optimized_sequences_path = os.path.join(output_dir, 'optimized_dna_sequences.csv')
    optimized_sequences.to_csv(optimized_sequences_path, index=False)
    print(f"Optimized DNA sequences saved to {optimized_sequences_path}")
    
    print("\n" + "="*60)
    print("GENERATING VISUALIZATIONS...")
    print("="*60)
    
    summary_df = generate_visualizations(results_df, output_dir)
    
    print("\n" + "="*60)
    print("BENCHMARK EVALUATION COMPLETE")
    print("="*60)
    print(f"Results saved to: {output_dir}")
    print(f"Total sequences processed: {len(results_df)}")
    print(f"Average runtime per sequence: {results_df['runtime'].mean():.2f} seconds")
    print(f"Total runtime: {results_df['runtime'].sum():.2f} seconds")


if __name__ == "__main__":
    main()