File size: 15,706 Bytes
5686f5b | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 | from __future__ import annotations
import math
from typing import Any, Collection, Dict, List, Tuple, Optional
import pandas as pd # if you have it in this module already
# 1. Standardization dictionary
_DRUG_NAME_MAP = {
"1-hydroxymidazolam": "1-hydroxy-midazolam",
"4-hydroxytolbutamide": "4-hydroxy-tolbutamide",
"5-hydroxyomeprazole": "5-hydroxy-omeprazole",
"caffeine (137X)": "caffeine",
"dextromethorphan": "dextromethorphan",
"dextrorphan": "dextrorphan",
"digoxin": "digoxin",
"hydroxy repaglinide": "hydroxy-repaglinide",
"memantine": "memantine",
"midazolam": "midazolam",
"omeprazole": "omeprazole",
"omeprazole sulfone": "omeprazole sulfone",
"paracetamol": "paracetamol",
"paracetamol glucuronide": "paracetamol glucuronide",
"paraxanthine (17X)": "paraxanthine",
"repaglinide": "repaglinide",
"rosuvastatin": "rosuvastatin",
"tolbutamide": "tolbutamide",
"Indometacin": "indometacin",
"Theophylline": "theophylline",
}
reference_results = {
"drug": [
"caffeine",
"dextromethorphan",
"digoxin",
"memantine",
"midazolam",
"omeprazole",
"paracetamol",
"repaglinide",
"rosuvastatin",
"tolbutamide",
"1-hydroxy-midazolam",
"4-hydroxy-tolbutamide",
"5-hydroxy-omeprazole",
"dextrorphan",
"hydroxy-repaglinide",
"omeprazole sulfone",
"paracetamol glucuronide",
"paraxanthine",
],
"NLME": [
0.356,
0.796,
0.315,
0.411,
0.674,
1.470,
0.319,
0.632,
0.470,
0.766,
math.nan,
math.nan,
math.nan,
math.nan,
math.nan,
math.nan,
math.nan,
math.nan,
],
"NODE-PK": [
0.914,
0.668,
1.403,
0.549,
0.456,
1.940,
1.094,
0.879,
0.471,
0.683,
0.741,
0.871,
2.014,
0.723,
0.340,
1.992,
0.509,
1.648,
],
"T-PK": [
0.575,
0.630,
0.717,
0.799,
0.735,
1.864,
0.825,
0.846,
0.748,
0.816,
0.678,
0.898,
1.683,
1.001,
0.532,
1.620,
0.423,
0.646,
],
"SNODE-PK": [
0.780,
1.702,
0.501,
0.580,
0.874,
1.267,
1.115,
1.514,
0.624,
0.949,
1.395,
0.524,
1.811,
0.904,
0.059,
1.529,
0.823,
0.653,
],
"ST-PK": [
0.984,
1.412,
0.421,
0.869,
0.817,
1.078,
1.050,
1.246,
0.604,
0.998,
1.216,
0.742,
1.600,
0.860,
0.336,
1.294,
1.057,
0.858,
],
"AICME-RNN": [
0.646,
0.640,
0.569,
0.534,
0.548,
1.395,
0.691,
0.562,
0.578,
0.854,
0.935,
0.274,
1.575,
0.614,
0.095,
1.438,
0.365,
0.409,
],
"AICMET": [
0.477,
0.437,
0.457,
0.362,
0.366,
1.139,
0.406,
0.583,
0.396,
0.691,
0.729,
0.265,
1.615,
0.374,
0.113,
1.366,
0.295,
0.266,
],
}
reference_data_nme = {
"drug": [
"caffeine (137X)",
"dextromethorphan",
"digoxin",
"memantine",
"midazolam",
"omeprazole",
"paracetamol",
"repaglinide",
"rosuvastatin",
"tolbutamide",
"indometacin",
"theophylline",
],
"log-rmse": [0.356, 0.796, 0.315, 0.411, 0.674, 1.47, 0.319, 0.632, 0.470, 0.766, 0.604, 0.754],
"log-r2": [0.820, 0.556, 0.482, 0.740, 0.344, -0.75, 0.905, 0.561, 0.557, 0.506, 100.0, 100.0],
}
reference_df = pd.DataFrame(reference_results)
def normalize_drug_name(raw: str) -> str:
"""
Normalize drug names from comet logs to match reference table names.
Falls back to the raw name if no mapping exists.
"""
return _DRUG_NAME_MAP.get(raw, raw)
def _extract_drug_from_metric_name(
metric_name_full: str,
metric_name: str,
top_level: str | None = None,
) -> str | None:
"""
Extract drug name from metricName.
Handles patterns like:
"Empirical/Synthetic/paracetamol glucuronide/r2"
"Synthetic/Synthetic/substance_16/rmse"
(and ignores things like "Empirical/epoch_399/r2")
top_level:
If given, require metricName to start with this first segment, e.g. "Empirical".
"""
parts = metric_name_full.split("/")
if not parts:
return None
# Require that the last segment matches the metric_name we're interested in
if parts[-1] != metric_name:
return None
# Optional filter on the very first segment: "Empirical", "Synthetic", etc.
if top_level is not None and parts[0] != top_level:
return None
# Drop the metric name at the end
core = parts[:-1]
# Old-style names might have a trailing "epoch_399" segment; drop it if present
if core and core[-1].startswith("epoch_"):
core = core[:-1]
# We expect at least [prefix, drug] -> length >= 2
if len(core) < 2:
return None
raw_drug = core[-1]
if not raw_drug:
return None
# Don't treat these prefixes as drugs
if raw_drug.lower() in {"empirical", "synthetic", "train", "val", "test"}:
return None
return normalize_drug_name(raw_drug)
def metrics_list_to_pandas(
metrics_list: List[Dict[str, Any]],
model_name: str,
metric_name: str,
epoch: int | str,
top_level: str | None = None,
) -> pd.DataFrame:
"""
Convert comet_ml metrics to a per-drug DataFrame for a given metric and epoch.
metrics_list entries look like:
{
"metricName": "Empirical/Synthetic/paracetamol glucuronide/r2",
"metricValue": "-0.09778215289115906",
"timestamp": 1764093835814,
"step": 2,
"epoch": 0,
...
}
top_level:
Optional filter on the first path segment in metricName, e.g. "Empirical" or "Synthetic".
"""
# -----------------------
# 1) Resolve target epoch
# -----------------------
target_epoch: int | None
if isinstance(epoch, str):
if epoch == "last":
epochs: List[int] = []
for m in metrics_list:
e = m.get("epoch")
try:
if e is not None:
epochs.append(int(e))
except (TypeError, ValueError):
continue
target_epoch = max(epochs) if epochs else None
else:
# Unknown epoch label → nothing to do
return pd.DataFrame(columns=["drug", model_name])
else:
target_epoch = int(epoch)
if target_epoch is None:
return pd.DataFrame(columns=["drug", model_name])
# -----------------------
# 2) Collect rows
# -----------------------
rows: list[tuple[str, float, int]] = []
for m in metrics_list:
name = m.get("metricName") or ""
drug = _extract_drug_from_metric_name(
metric_name_full=name,
metric_name=metric_name,
top_level=top_level,
)
if not drug:
continue
# Filter by epoch field (new comet format)
e_raw = m.get("epoch")
try:
e_val = int(e_raw)
except (TypeError, ValueError):
continue
if e_val != target_epoch:
continue
# Metric value
try:
value = float(m.get("metricValue"))
except (TypeError, ValueError):
continue
ts = int(m.get("timestamp", 0))
rows.append((drug, value, ts))
if not rows:
return pd.DataFrame(columns=["drug", model_name])
# -----------------------
# 3) Keep latest per drug
# -----------------------
latest: dict[str, tuple[float, int]] = {}
for drug, value, ts in rows:
cur = latest.get(drug)
if cur is None or ts > cur[1]:
latest[drug] = (value, ts)
data = [{"drug": d, model_name: vts[0]} for d, vts in latest.items()]
df = pd.DataFrame(data).sort_values("drug").reset_index(drop=True)
return df
def empirical_batches_to_pandas(
all_empirical_batches: Dict[str, List["AICMECompartmentsDataBatch"]],
model: Any,
model_name: str,
metric_name: str,
repo_filter: Optional[Collection[str]] = None,
) -> pd.DataFrame:
"""
Aggregate per-drug metrics computed from all_empirical_batches into a
DataFrame with columns ["drug", model_name], analogous to metrics_list_to_pandas.
Parameters
----------
all_empirical_batches : Dict[str, List[AICMECompartmentsDataBatch]]
Mapping repo_id -> list of batches.
model : Any
Model instance exposing `_compute_metrics_from_batch_list(batch_list, repo_id)`.
model_name : str
Name of the model; becomes the metric column name in the DataFrame.
metric_name : str
Which metric to extract ("rmse", "log_rmse", "r2", "log_r2", ...).
repo_filter : Optional[Collection[str]]
If given, only these repo_ids are processed.
Returns
-------
pd.DataFrame
Columns: ["drug", model_name], sorted by drug.
"""
rows: list[tuple[str, float, str]] = []
for repo_id, batch_list in all_empirical_batches.items():
if repo_filter is not None and repo_id not in repo_filter:
continue
# metrics: dict[raw_drug -> dict[metric_name -> value, ...]]
metrics, _prediction_cache = model._compute_metrics_from_batch_list(batch_list, repo_id)
for raw_drug, metric_dict in metrics.items():
if metric_dict is None:
continue
if metric_name not in metric_dict:
continue
value = metric_dict[metric_name]
if value is None:
continue
try:
v = float(value)
except (TypeError, ValueError):
continue
drug = normalize_drug_name(raw_drug)
rows.append((drug, v, repo_id))
if not rows:
return pd.DataFrame(columns=["drug", model_name])
# If a drug appears multiple times (e.g. in several repos), keep the last one.
latest_by_drug: Dict[str, float] = {}
for drug, value, _repo_id in rows:
latest_by_drug[drug] = value
data = [{"drug": d, model_name: v} for d, v in latest_by_drug.items()]
df = pd.DataFrame(data).sort_values("drug").reset_index(drop=True)
return df
def reference_dict_to_pandas(
reference_data: Dict[str, list],
model_name: str,
metric_name: str,
) -> pd.DataFrame:
"""
Convert a reference dictionary with drug-level metrics into a pandas DataFrame.
The dictionary must have at least the keys:
- "drug": list[str]
- <metric_name>: list[float]
Applies normalization of drug names to ensure consistency.
Parameters
----------
reference_data : dict
Dictionary with keys "drug" and metric names (e.g., "log-rmse", "log-r2").
model_name : str
Name for the output value column (like "NodePK" or "GP").
metric_name : str
Which metric to extract (must be in the dict).
Returns
-------
pd.DataFrame
Two-column DataFrame with:
- "drug": standardized drug names
- model_name: metric values
Sorted by drug name.
"""
if metric_name not in reference_data:
raise ValueError(
f"Metric '{metric_name}' not in reference_data keys {list(reference_data.keys())}"
)
drugs = [normalize_drug_name(d) for d in reference_data["drug"]]
values = reference_data[metric_name]
df = pd.DataFrame({"drug": drugs, model_name: values})
return df.sort_values("drug").reset_index(drop=True)
def available_epochs_and_metrics(metrics_list: List[Dict[str, Any]]) -> Dict[str, list[str]]:
"""
Summarize which epochs, metrics and top-level prefixes are available in
a comet_ml metrics list.
This handles both:
- New-style: epoch is in the 'epoch' field and metricName is something like
"Empirical/Synthetic/paracetamol glucuronide/r2"
- Old-style: epoch encoded in metricName, e.g.
"Empirical/epoch_399/r2"
Returns
-------
Dict[str, list[str]]
{
"epochs_available": list of unique epoch identifiers (strings),
"metrics_available": list of unique metric names (last path segment),
"top_levels_available": list of unique top-level prefixes (first path segment)
}
"""
epochs: set[str] = set()
metrics: set[str] = set()
top_levels: set[str] = set()
for m in metrics_list:
name = m.get("metricName") or ""
if not name:
continue
parts = name.split("/")
if not parts:
continue
# top-level, e.g. "Empirical" or "Synthetic"
top_levels.add(parts[0])
# metric name is always the last segment, e.g. "rmse", "r2"
metric = parts[-1]
metrics.add(metric)
# --- New-style: epoch field present ---
e_field = m.get("epoch", None)
if e_field is not None:
try:
epochs.add(str(int(e_field)))
except (TypeError, ValueError):
pass
else:
# --- Fallback: old-style epoch encoded in the parent segment ---
if len(parts) >= 2:
parent = parts[-2]
if parent.startswith("epoch_"):
epochs.add(parent.replace("epoch_", ""))
return {
"epochs_available": sorted(epochs, key=lambda x: (x != "last", x)),
"metrics_available": sorted(metrics),
"top_levels_available": sorted(top_levels),
}
def count_model_wins(
df: pd.DataFrame,
model_a: str,
model_b: str,
*,
smaller_is_better: bool = True,
) -> Tuple[int, int, int]:
"""
Compare two models column-by-column in a merged DataFrame and count wins.
Parameters
----------
df : pd.DataFrame
Must contain the two columns `model_a` and `model_b` with numeric values.
model_a : str
Name of the first model column in df.
model_b : str
Name of the second model column in df.
smaller_is_better : bool, default=True
If True, smaller values are considered better (e.g. RMSE).
If False, larger values are considered better (e.g. R^2).
Returns
-------
wins_a : int
Number of rows where model_a outperforms model_b.
wins_b : int
Number of rows where model_b outperforms model_a.
ties : int
Number of rows where they are equal (after dropping NaNs).
"""
# Select valid rows only
valid = df[[model_a, model_b]].dropna()
if smaller_is_better:
wins_a = (valid[model_a] < valid[model_b]).sum()
wins_b = (valid[model_b] < valid[model_a]).sum()
else:
wins_a = (valid[model_a] > valid[model_b]).sum()
wins_b = (valid[model_b] > valid[model_a]).sum()
ties = (valid[model_a] == valid[model_b]).sum()
return wins_a, wins_b, ties
|