kenkaneki's picture
cahages
f990f50
"""
Process CodeReview Bench leaderboard data and submissions.
"""
import json
import os
import pandas as pd
from datetime import datetime
from typing import Dict, List, Tuple, Optional
import numpy as np
from src.display.utils import (
CODEREVIEW_COLUMN, DISPLAY_COLS, CATEGORIES, COMMENT_LANGUAGES, EXAMPLE_CATEGORIES,
MULTIMETRIC_METRICS, EXACT_MATCH_METRICS
)
def process_jsonl_submission(file_path: str) -> Tuple[List[Dict], str]:
"""
Process a JSONL submission file for CodeReview Bench.
Args:
file_path: Path to the JSONL submission file
Returns:
Tuple of (entries_list, message)
"""
try:
entries = []
with open(file_path, 'r', encoding='utf-8') as f:
for line_num, line in enumerate(f, 1):
line = line.strip()
if not line:
continue
try:
entry = json.loads(line)
# Validate required fields
required_fields = ['model_name', 'programming_language', 'comment_language']
missing_fields = [field for field in required_fields if field not in entry]
if missing_fields:
return [], f"Missing required fields {missing_fields} in line {line_num}"
# Validate metrics exist
has_multimetric = any(metric in entry for metric in MULTIMETRIC_METRICS)
has_exact_match = any(metric in entry for metric in EXACT_MATCH_METRICS)
if not has_multimetric and not has_exact_match:
return [], f"No valid metrics found in line {line_num}. Required: {MULTIMETRIC_METRICS + EXACT_MATCH_METRICS}"
entries.append(entry)
except json.JSONDecodeError as e:
return [], f"Invalid JSON in line {line_num}: {e}"
if not entries:
return [], "No valid entries found in submission file"
return entries, f"Successfully processed {len(entries)} entries"
except Exception as e:
return [], f"Error processing submission: {e}"
def calculate_overall_score(entry: Dict) -> float:
"""
Calculate overall score for a CodeReview Bench entry.
Args:
entry: Dictionary containing model evaluation results
Returns:
Overall score as float
"""
# Calculate multimetric average
multimetric_scores = []
for metric in MULTIMETRIC_METRICS:
if metric in entry and isinstance(entry[metric], (int, float)):
multimetric_scores.append(entry[metric])
multimetric_avg = np.mean(multimetric_scores) if multimetric_scores else 0
# Calculate exact match average
exact_match_scores = []
for metric in EXACT_MATCH_METRICS:
if metric in entry and isinstance(entry[metric], (int, float)):
exact_match_scores.append(entry[metric])
exact_match_avg = np.mean(exact_match_scores) if exact_match_scores else 0
# Weighted combination (can be adjusted based on requirements)
overall_score = (multimetric_avg * 0.7) + (exact_match_avg * 0.3)
return overall_score
def load_leaderboard_data(file_path: str) -> Dict:
"""
Load the leaderboard data from a JSON file.
"""
if not os.path.exists(file_path):
version = "v0"
if "_v" in file_path:
version = file_path.split("_")[-1].split(".")[0]
return {"entries": [], "last_updated": datetime.now().isoformat(), "version": version}
with open(file_path, 'r') as f:
data = json.load(f)
# Ensure version field exists
if "version" not in data:
version = "v0"
if "_v" in file_path:
version = file_path.split("_")[-1].split(".")[0]
data["version"] = version
return data
def save_leaderboard_data(data: Dict, file_path: str) -> None:
"""
Save the leaderboard data to a JSON file.
"""
# Ensure the directory exists
os.makedirs(os.path.dirname(file_path), exist_ok=True)
# Update the last_updated timestamp
data["last_updated"] = datetime.now().isoformat()
# Ensure version is set
if "version" not in data:
version = "v0"
if "_v" in file_path:
version = file_path.split("_")[-1].split(".")[0]
data["version"] = version
with open(file_path, 'w') as f:
json.dump(data, f, indent=2)
def leaderboard_to_dataframe(leaderboard_data: Dict) -> pd.DataFrame:
"""
Convert leaderboard data to a pandas DataFrame for display.
"""
rows = []
for entry in leaderboard_data.get("entries", []):
model_name = entry.get("model_name", "Unknown Model")
# Extract basic metadata
row = {
"model_name": model_name,
"model_type": entry.get("model_type", "Unknown"),
"mode": entry.get("mode", "Strict"),
"submission_date": entry.get("submission_date", ""),
"version": entry.get("version", "v0"),
"review_model_type": entry.get("review_model_type", "custom").lower()
}
# Add additional metadata fields if present
for key in ["base_model", "revision", "precision", "weight_type", "topic", "programming_language", "comment_language"]:
if key in entry:
row[key] = entry[key]
# Add multimetric scores
for metric in MULTIMETRIC_METRICS:
if metric in entry:
row[metric] = entry[metric]
else:
row[metric] = pd.NA
# Add exact match metrics
for metric in EXACT_MATCH_METRICS:
if metric in entry:
row[metric] = entry[metric]
else:
row[metric] = pd.NA
# Calculate aggregated metrics
multimetric_scores = [entry.get(metric, 0) for metric in MULTIMETRIC_METRICS if metric in entry and pd.notna(entry[metric])]
exact_match_scores = [entry.get(metric, 0) for metric in EXACT_MATCH_METRICS if metric in entry and pd.notna(entry[metric])]
if multimetric_scores:
row["multimetric_average"] = np.mean(multimetric_scores)
else:
row["multimetric_average"] = pd.NA
if exact_match_scores:
row["exact_match_average"] = np.mean(exact_match_scores)
else:
row["exact_match_average"] = pd.NA
# Calculate overall score
row["overall_score"] = calculate_overall_score(entry)
# Add language-specific metrics if available
for lang in COMMENT_LANGUAGES:
for metric in ["readability", "relevance", "overall_score"]:
lang_key = f"{lang}_{metric}"
if lang_key in entry:
row[lang_key] = entry[lang_key]
else:
row[lang_key] = pd.NA
# Add evaluation count
row["total_evaluations"] = entry.get("total_evaluations", entry.get("evaluation_count", pd.NA))
rows.append(row)
# Create DataFrame and sort by overall score
df = pd.DataFrame(rows)
# Ensure all expected columns exist
for metric in MULTIMETRIC_METRICS + EXACT_MATCH_METRICS:
if metric not in df.columns:
df[metric] = pd.NA
# Sort by overall score (descending)
if not df.empty:
df = df.sort_values(by="overall_score", ascending=False, na_position='last')
# Ensure summary columns exist
summary_cols = ["overall_score", "multimetric_average", "exact_match_average", "total_evaluations"]
for col in summary_cols:
if col not in df.columns:
df[col] = pd.NA
return df
def add_entries_to_leaderboard(leaderboard_data: Dict, new_entries: List[Dict]) -> Dict:
"""
Add new entries to the leaderboard, replacing any with the same model name.
"""
# Create a mapping of existing entries by model name and version
existing_entries = {
(entry["model_name"], entry.get("version", "v0")): i
for i, entry in enumerate(leaderboard_data.get("entries", []))
}
# Process each new entry
for new_entry in new_entries:
model_name = new_entry.get("model_name")
version = new_entry.get("version", "v0")
# Add calculated metrics
new_entry["overall_score"] = calculate_overall_score(new_entry)
# Calculate averages
multimetric_scores = [new_entry.get(metric) for metric in MULTIMETRIC_METRICS if metric in new_entry and pd.notna(new_entry[metric])]
exact_match_scores = [new_entry.get(metric) for metric in EXACT_MATCH_METRICS if metric in new_entry and pd.notna(new_entry[metric])]
if multimetric_scores:
new_entry["multimetric_average"] = np.mean(multimetric_scores)
if exact_match_scores:
new_entry["exact_match_average"] = np.mean(exact_match_scores)
if (model_name, version) in existing_entries:
# Replace existing entry
leaderboard_data["entries"][existing_entries[(model_name, version)]] = new_entry
else:
# Add new entry
if "entries" not in leaderboard_data:
leaderboard_data["entries"] = []
leaderboard_data["entries"].append(new_entry)
# Update the last_updated timestamp
leaderboard_data["last_updated"] = datetime.now().isoformat()
return leaderboard_data