Spaces:
Running
Running
File size: 21,195 Bytes
ee1b999 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 |
import logging
from typing import Optional, Any, Dict # Added Dict
from zoneinfo import ZoneInfo
# datasets import might not be strictly needed by LeaderboardViewer itself anymore,
# but _get_dataframe might still use types from it if EvalResult refers to them.
# For now, let's keep it if your EvalResult or SuiteConfig models have dependencies.
# If not, it can be removed from here.
import datasets # Potentially removable from this file
import matplotlib.pyplot as plt
import plotly.express as px
import plotly.graph_objects as go
import numpy as np
import pandas as pd
import seaborn as sns
import json # For loading the local JSON file
import os # For checking file existence
from agenteval import compute_summary_statistics
from agenteval.config import SuiteConfig
from agenteval.models import EvalResult
logger = logging.getLogger(__name__)
import logging
from typing import Optional, Any, Dict, List # Added List
from zoneinfo import ZoneInfo # Assuming this might be used by SuiteConfig/EvalResult or _get_dataframe
import json
import os
# Assuming these are correctly imported from your project
from agenteval.config import SuiteConfig
from agenteval.models import EvalResult
# from agenteval import compute_summary_statistics # Used by _get_dataframe
class DataTransformer:
"""
Load and visualize leaderboard from a single, local JSON result file.
"""
_INFORMAL_TO_FORMAL_NAME_MAP = {
"lit": "Literature Understanding",
"data": "Data Analysis",
"code": "Code Execution",
"discovery": "Discovery",
"arxivdigestables_validation": "Arxivdigestables Validation",
"sqa_dev": "Sqa Dev",
"litqa2_validation": "Litqa2 Validation",
"paper_finder_validation": "Paper Finder Validation",
"discoverybench_validation": "Discoverybench Validation",
"core_bench_validation": "Core Bench Validation",
"ds1000_validation": "DS1000 Validation",
"e2e_discovery_validation": "E2E Discovery Validation",
"super_validation": "Super Validation",
# Add any other raw names that can appear in task.name or task.tags
}
def __init__(
self,
json_file_path: str, # Mandatory: path to the local JSON file
split: str, # Still needed for context within the JSON's suite_config
is_internal: bool = False
):
self._json_file_path = json_file_path
self._split = split
self._internal = is_internal
self._loaded_json_data: Optional[Dict[str, Any]] = None
self._cfg: Optional[SuiteConfig] = None
logger.info(f"Initializing LeaderboardViewer with local JSON file: {self._json_file_path}")
# --- Load and Validate JSON data ---
if not os.path.exists(self._json_file_path):
raise FileNotFoundError(f"JSON file not found at path: {self._json_file_path}")
try:
with open(self._json_file_path, 'r', encoding='utf-8') as f:
self._loaded_json_data = json.load(f)
except json.JSONDecodeError as e:
raise ValueError(f"Failed to parse JSON from local file {self._json_file_path}: {e}")
except Exception as e:
raise ValueError(f"Error reading local file {self._json_file_path}: {e}")
if not self._loaded_json_data:
raise ValueError(f"No data loaded from JSON file {self._json_file_path}.")
try:
eval_result = EvalResult.model_validate(self._loaded_json_data)
except Exception as e:
raise ValueError(f"Failed to validate JSON data from file '{self._json_file_path}' against EvalResult model: {e}")
self._cfg = eval_result.suite_config
if not isinstance(self._cfg, SuiteConfig):
raise TypeError(f"self._cfg is not a SuiteConfig object after loading from '{self._json_file_path}', got {type(self._cfg)}.")
# --- Populate Tag Map (Corrected Placement and Helper Function Access) ---
self.tag_map: dict[str, list[str]] = {}
# Access tasks from the loaded config
tasks_for_split: List[Any] = self._cfg.get_tasks(self._split) # Assuming get_tasks returns a list of task-like objects
for task in tasks_for_split:
# Ensure task object has 'name' and 'tags' attributes
if not hasattr(task, 'name') or not hasattr(task, 'tags'):
logger.warning(f"Task object {task} is missing 'name' or 'tags' attribute. Skipping.")
continue
formal_task_display_name = self._get_formal_display_name_static(task.name) # Use the helper method
if not (task.tags or []):
continue
for raw_tag_name in task.tags:
formal_tag_display_name_key = self._get_formal_display_name_static(raw_tag_name)
self.tag_map.setdefault(formal_tag_display_name_key, []).append(formal_task_display_name)
for key in self.tag_map:
self.tag_map[key] = sorted(list(set(self.tag_map[key])))
# --- Helper function defined as a static method or regular method ---
# Option 1: Static method (doesn't need 'self', uses the class attribute)
@staticmethod
def _get_formal_display_name_static(raw_name: str) -> str:
"""
Helper function to get the formal display name for a raw tag or task name.
Uses the class's map and provides a fallback.
"""
return DataTransformer._INFORMAL_TO_FORMAL_NAME_MAP.get(raw_name, raw_name.replace("_", " ").title())
def _load(self) -> tuple[pd.DataFrame, dict[str, list[str]]]:
"""
Prepares the DataFrame from the loaded JSON data.
The JSON data is already loaded and validated in __init__.
"""
if self._loaded_json_data is None or self._cfg is None:
# This should not happen if __init__ completed successfully
raise RuntimeError("LeaderboardViewer2 not properly initialized. JSON data or SuiteConfig is missing.")
# The _get_dataframe function expects a list of records.
# Since we have a single JSON file representing one result, wrap it in a list.
records_list: list[dict] = [self._loaded_json_data]
overview_df = _get_dataframe(
records_list=records_list,
split=self._split,
is_internal=self._internal,
suite_config=self._cfg, # Pass the SuiteConfig loaded in __init__
)
return overview_df, self.tag_map
# --- view method remains the same as your last version ---
def view(
self,
tag: Optional[str] = None,
with_plots: bool = False,
use_plotly: bool = False,
) -> tuple[pd.DataFrame, dict[str, Any]]:
data, tag_map = self._load() # tag_map is also returned by _load now
print(f"AHAHASHJDBFGASJHDBJAHSDB,AHDB {tag_map}")
print(f"THIS IS THE DATA DATA DTAA {data.columns}")
if data.empty or (len(data) == 1 and data.iloc[0].get("Agent") == "No data"):
logger.warning("No data available to view. Returning empty DataFrame and plots.")
return data, {}
base_cols = ["Agent", "Submitter", "Date", "Logs"]
existing_cols = [col for col in base_cols if col in data.columns]
primary_score_col: str
group_metric_names: list[str]
if tag is None:
primary = "Overall"
group = list(tag_map.keys())
else:
primary = tag
group = tag_map.get(tag, [])
if f"{primary} Score" in data.columns:
data = data.sort_values(f"{primary} Score", ascending=False)
else:
logger.warning(f"Primary metric '{primary}' for sorting not found. Data will not be sorted by it.")
metrics_to_display = []
if f"{primary} Cost" in data.columns:
metrics_to_display.append(f"{primary} Cost")
if f"{primary} Score" in data.columns:
metrics_to_display.append(f"{primary} Score")
for g_item in group:
if g_item in data.columns:
metrics_to_display.append(g_item)
if f"{g_item} Cost" in data.columns:
metrics_to_display.append(f"{g_item} Cost")
if f"{g_item} Score" in data.columns:
metrics_to_display.append(f"{g_item} Score")
final_cols_to_display = existing_cols + [m for m in metrics_to_display if m in data.columns]
final_cols_to_display = sorted(list(set(final_cols_to_display)), key=final_cols_to_display.index)
df_view = data.loc[:, final_cols_to_display].reset_index(drop=True)
plots: dict[str, Any] = {}
if with_plots:
plot_metric_names = [primary] + [g_item for g_item in group if g_item in data.columns]
for metric_name in plot_metric_names:
score_col = f"{metric_name} Score"
cost_col = f"{metric_name} Cost"
if score_col in df_view.columns and cost_col in df_view.columns:
if use_plotly:
fig = _plot_scatter_plotly(df_view, x=cost_col, y=score_col, agent_col="Agent")
plots[f"scatter_{metric_name}"] = fig
else:
logger.warning(
f"Skipping plot for '{metric_name}': score column '{score_col}' or cost column '{cost_col}' not found."
)
return df_view, plots
def _safe_round(value, digits=2):
return round(value, digits) if isinstance(value, (float, int)) and pd.notna(value) else value
def _get_dataframe(
records_list: list[dict],
split: str,
is_internal: bool,
suite_config: SuiteConfig,
timezone: str = "US/Pacific",
) -> pd.DataFrame:
# This function remains the same as in the previous version you provided.
# It takes a list of records (which will be a list containing one item
# from the loaded JSON file) and processes it.
if not records_list:
logger.warning(f"No records provided to _get_dataframe for split '{split}'. Returning empty DataFrame with placeholder.")
expected_pretty_cols = ["Agent Name", "Submitter", "Date", "Overall Score", "Logs"]
empty_df = pd.DataFrame({p_col: ["No data"] for p_col in expected_pretty_cols})
return empty_df
cfg = suite_config
rows = []
for itm_idx, itm in enumerate(records_list):
if not isinstance(itm, dict):
logger.warning(f"Item {itm_idx} in records_list is not a dict, skipping.")
continue
try:
ev = EvalResult.model_validate(itm)
except Exception as e:
logger.error(f"Failed to validate item {itm_idx} with EvalResult: {itm}. Error: {e}")
continue
sub = ev.submission
date_str = None
if sub.submit_time is not None:
submit_dt = sub.submit_time
if not isinstance(submit_dt, pd.Timestamp):
if submit_dt.tzinfo is None:
logger.debug(f"Submission time for {sub.agent_name} is timezone-naive, assuming UTC.")
submit_dt = submit_dt.replace(tzinfo=ZoneInfo("UTC"))
date_str = pd.Timestamp(submit_dt).tz_convert(ZoneInfo(timezone)).strftime("%Y-%m-%d")
else:
date_str = None
if not ev.results:
logger.warning(
f"Skipping submission {sub.agent_name} ({sub.username or 'N/A'}) "
f"({sub.submit_time or 'N/A'}) due to no results."
)
continue
stats = compute_summary_statistics(
suite_config=cfg, split=split, results=ev.results
)
flat = {}
print(f"STATS STATS ASTATAS SD T S T A A {stats}")
for key, s_obj in stats.items():
parts = key.split("/")
if parts[0] == "overall":
flat["overall/score"] = _safe_round(getattr(s_obj, 'score', np.nan))
flat["overall/cost"] = _safe_round(getattr(s_obj, 'cost', np.nan))
elif parts[0] == "tag" and len(parts) > 1:
tag_name = parts[1]
flat[f"tag/{tag_name}/score"] = _safe_round(getattr(s_obj, 'score', np.nan))
flat[f"tag/{tag_name}/cost"] = _safe_round(getattr(s_obj, 'cost', np.nan))
elif parts[0] == "task" and len(parts) > 1:
task_name = parts[1]
score = getattr(s_obj, 'score', np.nan)
cost = getattr(s_obj, 'cost', np.nan)
score_stderr = getattr(s_obj, 'score_stderr', np.nan)
cost_stderr = getattr(s_obj, 'cost_stderr', np.nan)
flat[f"task/{task_name}/score"] = _safe_round(score)
flat[f"task/{task_name}/score_ci"] = _safe_round(score_stderr * 1.96 if pd.notna(score_stderr) else np.nan)
flat[f"task/{task_name}/cost"] = _safe_round(cost)
flat[f"task/{task_name}/cost_ci"] = _safe_round(cost_stderr * 1.96 if pd.notna(cost_stderr) else np.nan)
else:
logger.debug(f"Uncommon key structure from compute_summary_statistics: '{key}'. Attempting generic add.")
if hasattr(s_obj, 'score'):
flat[f"{key}/score"] = _safe_round(s_obj.score)
if hasattr(s_obj, 'cost'):
flat[f"{key}/cost"] = _safe_round(s_obj.cost)
current_logs_url = None
if is_internal and sub.logs_url:
current_logs_url = str(sub.logs_url)
elif not is_internal and sub.logs_url_public:
current_logs_url = str(sub.logs_url_public)
rows.append(
{
"agent_name": sub.agent_name or "N/A",
"username": sub.username or "N/A",
"submit_time": date_str,
**flat,
"logs_url": current_logs_url,
}
)
if not rows:
logger.warning(f"No valid rows generated from records_list for split '{split}'. Returning empty DataFrame with placeholder.")
expected_pretty_cols = ["Agent", "Submitter", "Date", "Overall Score", "Overall Cost", "Logs"]
empty_df = pd.DataFrame({p_col: ["No data"] for p_col in expected_pretty_cols})
return empty_df
df = pd.DataFrame(rows)
pretty_cols = {c: _pretty_column_name(c) for c in df.columns if c in df.columns}
overview = df.rename(columns=pretty_cols)
return overview
def _pretty_column_name(col: str) -> str:
"""Map raw column name to display name."""
# --- Step 1: Fixed, direct mappings ---
fixed_mappings = {
"submit_time": "Date",
"agent_name": "Agent",
"username": "Submitter",
"logs_url": "Logs",
"overall/score": "Overall Score",
"overall/cost": "Overall Cost",
}
if col in fixed_mappings:
return fixed_mappings[col]
# --- Step 2: Define your mapping for informal names to descriptive names ---
informal_map = DataTransformer._INFORMAL_TO_FORMAL_NAME_MAP
# --- Step 3: Dynamic mappings for task or tag columns using the informal_to_formal_name_map ---
parts = col.split("/")
if len(parts) == 3:
item_type, informal_name, metric_suffix = parts #
formal_name = informal_map.get(informal_name)
if formal_name is None:
formal_name = informal_name.replace("_", " ").title()
print(f"[DEBUG _pretty_column_name] Informal name '{informal_name}' not in map, using fallback: '{formal_name}'")
if metric_suffix == "score":
return f"{formal_name} Score"
if metric_suffix == "cost":
return f"{formal_name} Cost"
if metric_suffix == "score_ci":
return f"{formal_name} Score 95% CI"
if metric_suffix == "cost_ci":
return f"{formal_name} Cost 95% CI"
# --- Step 4: Fallback for columns that don't match the "type/name/metric" pattern ---
if "/" not in col:
return col.replace("_", " ").title()
else:
return parts[-1].replace("_", " ").title()
DEFAULT_Y_COLUMN = "Overall Score"
DUMMY_X_VALUE_FOR_MISSING_COSTS = 0 # Value to use if x-axis data (costs) is missing
def _plot_scatter_plotly(
data: pd.DataFrame,
x: Optional[str],
y: str,
agent_col: str = "Agent"
) -> go.Figure:
x_col_to_use = x
y_col_to_use = y
# 1. Check if y-column exists
if y_col_to_use not in data.columns:
logger.error(
f"y-axis column '{y_col_to_use}' MUST exist in DataFrame. "
f"Cannot generate plot. Available columns: {data.columns.tolist()}"
)
return go.Figure()
# 2. Check if agent_col exists
if agent_col not in data.columns:
logger.warning(
f"Agent column '{agent_col}' not found in DataFrame. "
f"Available columns: {data.columns.tolist()}. Returning empty figure."
)
return go.Figure()
# 3. Prepare data (make a copy, handle numeric conversion for y)
data_plot = data.copy()
try:
data_plot[y_col_to_use] = pd.to_numeric(data_plot[y_col_to_use], errors='coerce')
except Exception as e:
logger.error(f"Error converting y-column '{y_col_to_use}' to numeric: {e}. Returning empty figure.")
return go.Figure()
# 4. Handle x-column (costs)
x_axis_label = x_col_to_use if x_col_to_use else "Cost (Data N/A)" # Label for the x-axis
x_data_is_valid = False
if x_col_to_use and x_col_to_use in data_plot.columns:
try:
data_plot[x_col_to_use] = pd.to_numeric(data_plot[x_col_to_use], errors='coerce')
# Check if there's any non-NaN data after coercion for x
if data_plot[x_col_to_use].notna().any():
x_data_is_valid = True
else:
logger.info(f"x-axis column '{x_col_to_use}' exists but contains all NaN/None values after numeric conversion.")
except Exception as e:
logger.warning(f"Error converting x-column '{x_col_to_use}' to numeric: {e}. Will use dummy x-values.")
# x_data_is_valid remains False
else:
if x_col_to_use: # Name was provided but column doesn't exist
logger.warning(f"x-axis column '{x_col_to_use}' not found in DataFrame.")
else: # x (column name) was None
logger.info("x-axis column name was not provided (is None).")
if not x_data_is_valid:
logger.info(f"Using dummy x-value '{DUMMY_X_VALUE_FOR_MISSING_COSTS}' for all data points as x-data is missing or invalid.")
# Create a new column with the dummy x-value for all rows
# Use a unique name for this dummy column to avoid potential clashes
dummy_x_col_name = "__dummy_x_for_plotting__"
data_plot[dummy_x_col_name] = DUMMY_X_VALUE_FOR_MISSING_COSTS
x_col_to_use = dummy_x_col_name # Update x_col_to_use to point to our dummy data
x_axis_label = x if x else "Cost (Data N/A)" # Use original x name for label if provided
# or a generic label if x was None.
# Could also be f"Cost (Fixed at {DUMMY_X_VALUE_FOR_MISSING_COSTS})"
# 5. Drop rows where y is NaN (x is now guaranteed to have values, either real or dummy)
data_plot.dropna(subset=[y_col_to_use], inplace=True)
fig = go.Figure()
if data_plot.empty:
logger.warning(f"No valid data to plot for y='{y_col_to_use}' (and x='{x_col_to_use}') after cleaning NaNs from y.")
# Still return a figure object, but it will be empty. Update layout for clarity.
fig.update_layout(
title=f"{y_col_to_use} vs. {x_axis_label} (No Data)",
xaxis=dict(title=x_axis_label, range=[DUMMY_X_VALUE_FOR_MISSING_COSTS - 1, DUMMY_X_VALUE_FOR_MISSING_COSTS + 1] if not x_data_is_valid else None),
yaxis=dict(title=y_col_to_use)
)
return fig
for agent, group in data_plot.groupby(agent_col):
hover_x_display = "%{x:.2f}" if x_data_is_valid else str(DUMMY_X_VALUE_FOR_MISSING_COSTS) + " (fixed)"
fig.add_trace(go.Scatter(
x=group[x_col_to_use],
y=group[y_col_to_use],
mode='markers',
name=str(agent),
hovertemplate=f"{x_axis_label}: {hover_x_display}<br>{y_col_to_use}: %{{y:.2f}}<extra>{str(agent)}</extra>",
marker=dict(size=10)
))
# Configure layout
xaxis_config = dict(title=x_axis_label)
if not x_data_is_valid: # If using dummy x, set a tighter, fixed range for x-axis
xaxis_config['range'] = [DUMMY_X_VALUE_FOR_MISSING_COSTS - 1, DUMMY_X_VALUE_FOR_MISSING_COSTS + 1]
xaxis_config['tickvals'] = [DUMMY_X_VALUE_FOR_MISSING_COSTS] # Show only one tick at the dummy value
xaxis_config['ticktext'] = [str(DUMMY_X_VALUE_FOR_MISSING_COSTS)]
else: # Real x-data
xaxis_config['rangemode'] = "tozero"
fig.update_layout(
title=f"{y_col_to_use} vs. {x_axis_label}",
xaxis=xaxis_config,
yaxis=dict(title=y_col_to_use, rangemode="tozero"),
legend_title_text=agent_col
)
return fig |