OSCR

Automatic selection of the best neural architecture for time series forecasting.

Code ↔ Paper

6 matches between paragraphs of the paper and lines of its authors' code, computed by the harvester (lexical-v1). Click a colored paragraph or line to see its counterpart.

2 blocks of code here have a permanent identifier (oscr:b:…), listed with the matches below: citable for good, with a page of its own. Select a match and press Cite to copy it. What a block identifier is.

The 6 matches
  1. [1] § Methods › Composite architectures ↔ Integrated1/NN.py, lines 120 to 268 · score 0.81 · multi head Attention, hidden state, LSTM block, layer normalization, GRU block, SSM block · block oscr:b:JY4XNJMGDD1KR-0 · Report a mismatch
  2. [2] § Methods › Composite architectures ↔ Integrated2/NN.py, lines 120 to 268 · score 0.81 · multi head Attention, hidden state, LSTM block, layer normalization, GRU block, SSM block · block oscr:b:V6P6NAXEFDFR2-F · Report a mismatch
  3. [3] § Results › Iterative architecture samplings for glucose prediction task ↔ Process/nsga2_glucose_generations.ipynb, lines 282 to 399 · score 0.72 · ground truth Pareto, crowding distance, dominated sorting, offspring, canonicalization, seeds · Report a mismatch
  4. [4] § Results › Iterative architecture samplings for glucose prediction task ↔ Process/nsga2_glucose_generations.ipynb, lines 451 to 494 · score 0.68 · 5–95 %, ground truth Pareto, band, Discovery, seed, NSGA · Report a mismatch
  5. [5] § Methods › Multi-objective evolutionary search via NSGA-II › Algorithm 2 ↔ Process/nsga2_glucose_generations.ipynb, lines 102 to 201 · score 0.61 · binary tournament selection, crowding distance, rank · Report a mismatch
  6. [6] § Methods › Multi-objective evolutionary search via NSGA-II › Algorithm 2 ↔ Process/nsga2_glucose_generations.ipynb, lines 282 to 399 · score 0.58 · crowding distance, dominated sorting, rank, fronts, NSGA, Pareto · Report a mismatch

Paper

Loaded from Europe PMC by your browser; OSCR does not store it: doi.org · Europe PMC

The paper is loaded when this pane is shown.

The authors' code

Jupyter notebook · 498 lines · 15 KB · no license · 4 matches

  1. # %% [markdown]
  2. # # NSGA-II (fixed generations) on the glucose architecture dataset (708)
  3. #
  4. # This notebook:
  5. # 1. Loads `data_glucose_2d.pkl` (708 unique architectures).
  6. # 2. Runs **NSGA-II for a fixed number of generations** (e.g., `G=10`) and plots the **Pareto front at generation 10**.
  7. # 3. Repeats NSGA-II for **20 random seeds**, and plots **# of discovered ground-truth Pareto architectures vs. generation** using **mean + 90% band (5–95%)**.
  8. #
  9. # Objectives (to minimize):
  10. # - `rela_error` (relative ℓ2 error)
  11. # - `training_time`
  12. # %%
  13. # Imports
  14. import sys
  15. import pickle
  16. from pathlib import Path
  17. from typing import Dict, List, Tuple, Optional
  18. import numpy as np
  19. import pandas as pd
  20. import matplotlib.pyplot as plt
  21. # %%
  22. # ---------- Robust loading helpers ----------
  23. # Placeholder for the class name used during pickling: "__main__.DataManual"
  24. class DataManual:
  25. def __init__(self, **kwargs):
  26. self.__dict__.update(kwargs)
  27. # Patch NumPy module aliases that can appear in pickles created with other NumPy versions.
  28. import numpy as _np
  29. sys.modules.setdefault("numpy._core", _np.core)
  30. sys.modules.setdefault("numpy._core.multiarray", _np.core.multiarray)
  31. sys.modules.setdefault("numpy._core.numeric", _np.core.numeric)
  32. if hasattr(_np.core, "_multiarray_umath"):
  33. sys.modules.setdefault("numpy._core._multiarray_umath", _np.core._multiarray_umath)
  34. # ---------- Load dataset ----------
  35. pkl_path = "data_glucose_2d.pkl" # if needed, change to the correct path
  36. with open(pkl_path, "rb") as f:
  37. data_loaded = pickle.load(f)
  38. print("Loaded:", type(data_loaded), "len =", len(data_loaded), "example type =", type(data_loaded[0]))
  39. rows = [obj.__dict__.copy() for obj in data_loaded]
  40. df = pd.DataFrame(rows)
  41. # Keep only needed fields (drop param_all)
  42. df = df[[
  43. "n_gru", "n_lstm", "n_attention", "n_ssm",
  44. "dim_hidden", "sequence",
  45. "rela_error", "training_time"
  46. ]].copy()
  47. assert len(df) == 708, "Expected 708 unique architectures."
  48. assert df.isna().sum().sum() == 0, "Found missing values."
  49. obj_cols = ["rela_error", "training_time"]
  50. df.head()
  51. # %% [markdown]
  52. # ## Canonicalization mapping (genotype → unique architecture index)
  53. #
  54. # Even though the dataset already contains 708 unique entries, during NSGA-II we generate children by
  55. # discrete crossover/mutation over the genotype `(counts, sequence, hidden_dim)` and then map back to
  56. # a unique architecture index via a canonical key.
  57. # %%
  58. TEMPLATES = {
  59. 1: ["SSM", "ATT", "GRU", "LSTM"],
  60. 2: ["ATT", "SSM", "GRU", "LSTM"],
  61. 3: ["SSM", "GRU", "ATT", "LSTM"],
  62. 4: ["GRU", "ATT", "SSM", "LSTM"],
  63. 5: ["ATT", "GRU", "SSM", "LSTM"],
  64. 6: ["GRU", "SSM", "ATT", "LSTM"],
  65. }
  66. HIDDEN_SET = [16, 32, 64]
  67. def expand_sequence(n_ssm: int, n_att: int, n_gru: int, n_lstm: int, seq_idx: int) -> Tuple[str, ...]:
  68. order = TEMPLATES[int(seq_idx)]
  69. counts = {"SSM": int(n_ssm), "ATT": int(n_att), "GRU": int(n_gru), "LSTM": int(n_lstm)}
  70. out: List[str] = []
  71. for b in order:
  72. out.extend([b] * counts[b])
  73. return tuple(out)
  74. def canonical_key_from_row(row: pd.Series) -> Tuple[Tuple[str, ...], int]:
  75. seq = expand_sequence(row["n_ssm"], row["n_attention"], row["n_gru"], row["n_lstm"], row["sequence"])
  76. return (seq, int(row["dim_hidden"]))
  77. key_to_idx: Dict[Tuple[Tuple[str, ...], int], int] = {}
  78. for idx, row in df.iterrows():
  79. k = canonical_key_from_row(row)
  80. if k in key_to_idx:
  81. raise ValueError(f"Duplicate canonical key found at idx={idx} and idx={key_to_idx[k]}: {k}")
  82. key_to_idx[k] = int(idx)
  83. print("Unique canonical keys:", len(key_to_idx))
  84. # %% [markdown]
  85. # ## NSGA-II components
  86. # - dominance / Pareto mask
  87. # - fast non-dominated sorting
  88. # - crowding distance
  89. # - binary tournament selection
  90. # %%
  91. def dominates(a: np.ndarray, b: np.ndarray) -> bool:
  92. return np.all(a <= b) and np.any(a < b)
  93. def pareto_mask(objs: np.ndarray) -> np.ndarray:
  94. n = objs.shape[0]
  95. is_eff = np.ones(n, dtype=bool)
  96. for i in range(n):
  97. if not is_eff[i]:
  98. continue
  99. for j in range(n):
  100. if i == j or not is_eff[j]:
  101. continue
  102. if dominates(objs[j], objs[i]):
  103. is_eff[i] = False
  104. break
  105. return is_eff
  106. def fast_non_dominated_sort(pop: List[int], obj: np.ndarray) -> List[List[int]]:
  107. n = len(pop)
  108. S = [[] for _ in range(n)]
  109. n_dom = np.zeros(n, dtype=int)
  110. fronts: List[List[int]] = [[]]
  111. for p in range(n):
  112. for q in range(n):
  113. if p == q:
  114. continue
  115. if dominates(obj[p], obj[q]):
  116. S[p].append(q)
  117. elif dominates(obj[q], obj[p]):
  118. n_dom[p] += 1
  119. if n_dom[p] == 0:
  120. fronts[0].append(p)
  121. i = 0
  122. while fronts[i]:
  123. next_front = []
  124. for p in fronts[i]:
  125. for q in S[p]:
  126. n_dom[q] -= 1
  127. if n_dom[q] == 0:
  128. next_front.append(q)
  129. i += 1
  130. fronts.append(next_front)
  131. if not fronts[-1]:
  132. fronts.pop()
  133. return fronts
  134. def crowding_distance(front: List[int], obj: np.ndarray) -> Dict[int, float]:
  135. if len(front) == 0:
  136. return {}
  137. M = obj.shape[1]
  138. dist = {p: 0.0 for p in front}
  139. for m in range(M):
  140. sorted_front = sorted(front, key=lambda p: obj[p, m])
  141. fmin = obj[sorted_front[0], m]
  142. fmax = obj[sorted_front[-1], m]
  143. dist[sorted_front[0]] = float("inf")
  144. dist[sorted_front[-1]] = float("inf")
  145. if fmax == fmin:
  146. continue
  147. for i in range(1, len(sorted_front) - 1):
  148. prev_p = sorted_front[i - 1]
  149. next_p = sorted_front[i + 1]
  150. dist[sorted_front[i]] += (obj[next_p, m] - obj[prev_p, m]) / (fmax - fmin)
  151. return dist
  152. def tournament_select(pop: List[int], ranks: Dict[int, int], crowds: Dict[int, float], rng: np.random.Generator) -> int:
  153. a = int(rng.integers(0, len(pop)))
  154. b = int(rng.integers(0, len(pop)))
  155. while b == a:
  156. b = int(rng.integers(0, len(pop)))
  157. ia = pop[a]
  158. ib = pop[b]
  159. ra = ranks[ia]
  160. rb = ranks[ib]
  161. if ra < rb:
  162. return ia
  163. if rb < ra:
  164. return ib
  165. ca = crowds.get(ia, 0.0)
  166. cb = crowds.get(ib, 0.0)
  167. return ia if ca >= cb else ib
  168. # %% [markdown]
  169. # ## Variation operators (discrete genotype)
  170. #
  171. # Genotype:
  172. # \[
  173. # (n_{\text{ssm}}, n_{\text{att}}, n_{\text{gru}}, n_{\text{lstm}}, \text{sequence}, \text{dim\_hidden})
  174. # \]
  175. # %%
  176. Genotype = Tuple[int, int, int, int, int, int]
  177. def row_to_genotype(row: pd.Series) -> Genotype:
  178. return (
  179. int(row["n_ssm"]),
  180. int(row["n_attention"]),
  181. int(row["n_gru"]),
  182. int(row["n_lstm"]),
  183. int(row["sequence"]),
  184. int(row["dim_hidden"]),
  185. )
  186. def repair_genotype(g: Genotype, rng: np.random.Generator) -> Genotype:
  187. n_ssm, n_att, n_gru, n_lstm, seq, h = g
  188. n_ssm = int(np.clip(n_ssm, 0, 2))
  189. n_att = int(np.clip(n_att, 0, 2))
  190. n_gru = int(np.clip(n_gru, 0, 2))
  191. n_lstm = int(np.clip(n_lstm, 0, 2))
  192. seq = int(np.clip(seq, 1, 6))
  193. if h not in HIDDEN_SET:
  194. h = int(rng.choice(HIDDEN_SET))
  195. if (n_ssm + n_att + n_gru + n_lstm) == 0:
  196. j = int(rng.integers(0, 4))
  197. if j == 0: n_ssm = 1
  198. if j == 1: n_att = 1
  199. if j == 2: n_gru = 1
  200. if j == 3: n_lstm = 1
  201. return (n_ssm, n_att, n_gru, n_lstm, seq, h)
  202. def canonicalize_to_index(g: Genotype) -> Optional[int]:
  203. n_ssm, n_att, n_gru, n_lstm, seq, h = g
  204. key = (expand_sequence(n_ssm, n_att, n_gru, n_lstm, seq), int(h))
  205. return key_to_idx.get(key, None)
  206. def uniform_crossover(g1: Genotype, g2: Genotype, rng: np.random.Generator) -> Genotype:
  207. child = []
  208. for a, b in zip(g1, g2):
  209. child.append(int(a) if rng.random() < 0.5 else int(b))
  210. return tuple(child) # type: ignore
  211. def mutate_genotype(g: Genotype, rng: np.random.Generator) -> Genotype:
  212. n_ssm, n_att, n_gru, n_lstm, seq, h = g
  213. choice = int(rng.integers(0, 6))
  214. if choice in [0, 1, 2, 3]:
  215. delta = int(rng.choice([-1, 1]))
  216. if choice == 0: n_ssm += delta
  217. if choice == 1: n_att += delta
  218. if choice == 2: n_gru += delta
  219. if choice == 3: n_lstm += delta
  220. elif choice == 4:
  221. seq = int(rng.integers(1, 7))
  222. else:
  223. h = int(rng.choice(HIDDEN_SET))
  224. return (n_ssm, n_att, n_gru, n_lstm, seq, h)
  225. def make_child(parent1_idx: int, parent2_idx: int, rng: np.random.Generator,
  226. p_c: float, p_m: float) -> Genotype:
  227. g1 = row_to_genotype(df.loc[parent1_idx])
  228. g2 = row_to_genotype(df.loc[parent2_idx])
  229. if rng.random() < p_c:
  230. child = uniform_crossover(g1, g2, rng)
  231. else:
  232. child = g1 if rng.random() < 0.5 else g2
  233. if rng.random() < p_m:
  234. child = mutate_genotype(child, rng)
  235. return repair_genotype(child, rng)
  236. # %% [markdown]
  237. # ## NSGA-II run for a fixed number of generations
  238. #
  239. # - Generation 0: after initialization (evaluate `P` architectures)
  240. # - Generation `g>=1`: after producing and evaluating up to `P` new offspring
  241. # %%
  242. def nsga2_run_generations(
  243. G: int = 10,
  244. P: int = 24,
  245. p_c: float = 0.9,
  246. p_m: float = 0.2,
  247. seed: int = 0,
  248. ) -> Dict:
  249. rng = np.random.default_rng(seed)
  250. all_indices = df.index.to_numpy()
  251. # Ground-truth Pareto set from all 708
  252. all_objs = df[obj_cols].to_numpy(dtype=float)
  253. true_pareto_idx = df.index[pareto_mask(all_objs)].to_numpy()
  254. true_pareto_set = set(map(int, true_pareto_idx))
  255. def current_pareto_in_evaluated(evaluated_set: set) -> List[int]:
  256. E_list = sorted(evaluated_set)
  257. E_objs = df.loc[E_list, obj_cols].to_numpy(dtype=float)
  258. mask = pareto_mask(E_objs)
  259. return [E_list[i] for i, ok in enumerate(mask) if ok]
  260. def true_pareto_found_in_evaluated(evaluated_set: set) -> int:
  261. return len(set(evaluated_set) & true_pareto_set)
  262. # initialization
  263. pop = rng.choice(all_indices, size=P, replace=False).tolist()
  264. evaluated = set(int(i) for i in pop)
  265. history = []
  266. history.append({
  267. "gen": 0,
  268. "evaluated_idx": sorted(evaluated),
  269. "pareto_idx": current_pareto_in_evaluated(evaluated),
  270. "true_pareto_found": true_pareto_found_in_evaluated(evaluated),
  271. })
  272. for gen in range(1, G + 1):
  273. pop_objs = df.loc[pop, obj_cols].to_numpy(dtype=float)
  274. fronts = fast_non_dominated_sort(pop, pop_objs)
  275. ranks: Dict[int, int] = {}
  276. crowds: Dict[int, float] = {}
  277. for k, front_pos in enumerate(fronts, start=1):
  278. for ppos in front_pos:
  279. ranks[int(pop[ppos])] = k
  280. cd = crowding_distance(front_pos, pop_objs)
  281. for ppos, dist in cd.items():
  282. crowds[int(pop[ppos])] = float(dist)
  283. # offspring generation
  284. offspring: List[int] = []
  285. attempts = 0
  286. max_attempts = 3000
  287. while len(offspring) < P:
  288. attempts += 1
  289. if attempts > max_attempts:
  290. remaining = list(set(map(int, all_indices)) - evaluated - set(offspring))
  291. if not remaining:
  292. break
  293. offspring.append(int(rng.choice(remaining)))
  294. continue
  295. p1 = tournament_select(pop, ranks, crowds, rng)
  296. p2 = tournament_select(pop, ranks, crowds, rng)
  297. child_g = make_child(p1, p2, rng, p_c=p_c, p_m=p_m)
  298. child_idx = canonicalize_to_index(child_g)
  299. if child_idx is None:
  300. continue
  301. child_idx = int(child_idx)
  302. if (child_idx in evaluated) or (child_idx in offspring):
  303. continue
  304. offspring.append(child_idx)
  305. for x in offspring:
  306. evaluated.add(int(x))
  307. # elitist replacement
  308. combined = pop + offspring
  309. combined_objs = df.loc[combined, obj_cols].to_numpy(dtype=float)
  310. fronts_R = fast_non_dominated_sort(combined, combined_objs)
  311. next_pop: List[int] = []
  312. for front_pos in fronts_R:
  313. if len(next_pop) + len(front_pos) <= P:
  314. next_pop.extend([int(combined[p]) for p in front_pos])
  315. else:
  316. cd = crowding_distance(front_pos, combined_objs)
  317. sorted_front = sorted(front_pos, key=lambda p: cd.get(p, 0.0), reverse=True)
  318. needed = P - len(next_pop)
  319. next_pop.extend([int(combined[p]) for p in sorted_front[:needed]])
  320. break
  321. pop = next_pop
  322. history.append({
  323. "gen": gen,
  324. "evaluated_idx": sorted(evaluated),
  325. "pareto_idx": current_pareto_in_evaluated(evaluated),
  326. "true_pareto_found": true_pareto_found_in_evaluated(evaluated),
  327. })
  328. return {
  329. "seed": seed,
  330. "P": P,
  331. "G": G,
  332. "history": history,
  333. "true_pareto_idx": list(map(int, true_pareto_idx)),
  334. }
  335. # %% [markdown]
  336. # ## (1) Run until generation 10 and plot the Pareto front at generation 10
  337. # %%
  338. from pathlib import Path
  339. import matplotlib.pyplot as plt
  340. G = 10
  341. seed = 0
  342. P = 24
  343. res = nsga2_run_generations(G=G, P=P, p_c=0.9, p_m=0.2, seed=seed)
  344. hG = res["history"][-1]
  345. true_pareto_idx = res["true_pareto_idx"]
  346. # output folder
  347. FIG_DIR = Path("figures")
  348. FIG_DIR.mkdir(exist_ok=True)
  349. # --- plot ---
  350. fig, ax = plt.subplots()
  351. ax.scatter(df["rela_error"], df["training_time"], alpha=0.25) # all 708
  352. tp = df.loc[true_pareto_idx]
  353. ax.scatter(tp["rela_error"], tp["training_time"], alpha=0.9) # ground-truth Pareto
  354. ev = df.loc[hG["evaluated_idx"]]
  355. ax.scatter(ev["rela_error"], ev["training_time"], alpha=0.9) # evaluated
  356. pr = df.loc[hG["pareto_idx"]]
  357. ax.scatter(pr["rela_error"], pr["training_time"], alpha=0.9) # Pareto in evaluated
  358. ax.set_xlabel("relative l2 error",fontsize=16)
  359. ax.set_ylabel("training time",fontsize=16)
  360. ax.set_title(f"NSGA-II Pareto front at generation {G} (seed={seed})")
  361. # --- save (PNG + PDF) ---
  362. png_path = FIG_DIR / f"nsga2_pareto_gen{G}_seed{seed}_P{P}.png"
  363. pdf_path = FIG_DIR / f"nsga2_pareto_gen{G}_seed{seed}_P{P}.pdf"
  364. fig.savefig(png_path, dpi=300, bbox_inches="tight")
  365. fig.savefig(pdf_path, bbox_inches="tight")
  366. print(f"Saved: {png_path.resolve()}")
  367. print(f"Saved: {pdf_path.resolve()}")
  368. plt.show()
  369. plt.close(fig)
  370. print(f"Generation {G}: evaluated = {len(hG['evaluated_idx'])}, "
  371. f"Pareto-in-evaluated size = {len(hG['pareto_idx'])}, "
  372. f"true Pareto found = {hG['true_pareto_found']} / {len(true_pareto_idx)}")
  373. # %% [markdown]
  374. # ## (2) 20 seeds: plot discovered ground-truth Pareto count vs generations (mean + 90% band)
  375. # %%
  376. from pathlib import Path
  377. import matplotlib.pyplot as plt
  378. n_runs = 20
  379. G = 20
  380. P = 24
  381. counts = np.zeros((n_runs, G + 1), dtype=float) # gen 0..G
  382. for s in range(n_runs):
  383. rr = nsga2_run_generations(G=G, P=P, p_c=0.9, p_m=0.2, seed=s)
  384. counts[s, :] = [hh["true_pareto_found"] for hh in rr["history"]]
  385. gens = np.arange(G + 1)
  386. mean = counts.mean(axis=0)
  387. low, high = np.percentile(counts, [5, 95], axis=0) # 90% band
  388. # output folder
  389. FIG_DIR = Path("figures")
  390. FIG_DIR.mkdir(exist_ok=True)
  391. # --- plot ---
  392. fig, ax = plt.subplots()
  393. ax.plot(gens, mean, label="mean")
  394. ax.fill_between(gens, low, high, alpha=0.25, label="90% band (5-95%)")
  395. ax.set_xlabel("generation",fontsize=16)
  396. ax.set_ylabel("# ground-truth Pareto architectures",fontsize=16)
  397. ax.set_title(f"NSGA-II (P={P}), {n_runs} seeds: discovery vs generation")
  398. ax.legend()
  399. # --- save (PNG + PDF) ---
  400. png_path = FIG_DIR / f"nsga2_true_pareto_discovery_mean90_G{G}_P{P}_runs{n_runs}.png"
  401. pdf_path = FIG_DIR / f"nsga2_true_pareto_discovery_mean90_G{G}_P{P}_runs{n_runs}.pdf"
  402. fig.savefig(png_path, dpi=300, bbox_inches="tight")
  403. fig.savefig(pdf_path, bbox_inches="tight")
  404. print(f"Saved: {png_path.resolve()}")
  405. print(f"Saved: {pdf_path.resolve()}")
  406. plt.show()
  407. plt.close(fig)
  408. # %%

nsga2_glucose_generations.ipynb at commit f07a6b7, no license · at the source

Overview

Authors: Qianying Cao1, Shanqing Liu1, Alan John Varghese2, Jérôme Darbon1, Michael S Triantafyllou3, George Em Karniadakis1
  1. Division of Applied Mathematics, Brown University, Providence, RI USA
  2. School of Engineering, Brown University, Providence, RI USA
  3. Department of Mechanical Engineering, Massachusetts Institute of Technology, Cambridge, MA USA
Institutions: Brown University (United States); Massachusetts Institute of Technology (United States)
Journal: Nature communications, volume 17, issue 1, article 7058
Dates: received 19 June 2025; accepted 15 May 2026; published online 2 June 2026
Type: Research article · Language: English
License: CC BY-NC-ND
Identifiers: DOI 10.1038/s41467-026-73687-9 · PMID 42230615 · PMCID PMC13392406 · OpenAlex W4406745912
Open access: gold, a free copy (OpenAlex)
Status: code verified (how the code is found and verified)
Categories: methods / tools (subfield)
Methods: Connectivity, Machine learning
Keywords: Computer science, Applied mathematics
Topic: Neural Networks and Applications (Artificial Intelligence, Computer Science), according to OpenAlex
Funding: United States Department of Defense | Defense Advanced Research Projects Agency (DARPA) (HR00112490484)
Citations: cited by 1 paper (Europe PMC); 69 references in the paper

Abstract

The abstract is not reproduced here: the paper's license (CC BY-NC-ND) does not allow it. Read it in the paper, at the publisher or on Europe PMC.

Repository

Its files are read in the Code ↔ Paper reader above, with 6 matches between paragraphs and lines of code.

qianyingcao/Selection_Architecture_MOO

License: none: the authors keep all their rights
State: the link answers, verified on 27 September 2026
Evidence: files inventoried
Commit: f07a6b72d93b8e04fcc8a7e564078d0743f7f763, 23 April 2026
Languages: Python (40), Jupyter (1)
Size: 44 files, 41 scripts
Software Heritage: not archived
Found in: “Code availability”
Holds: README, environment (requirements.txt), 1 notebook
Not found: license file, CITATION.cff, tests, continuous integration, documentation
Tools: PyTorch (37 files), NumPy (35 files), SciPy (31 files), Matplotlib (5 files), pandas (3 files)
Availability: 1 check, the latest on 27 September 2026: the link answers
  • 27 September 2026: the link answers
42 files

Code availability statement

The paper has a code availability statement. Its license (CC BY-NC-ND) does not allow reproducing it here; in short, from what the harvester recognized in it:

Read it in the paper: doi.org/10.1038/s41467-026-73687-9.

Tracing map

Proposed by the machine: these links were found in the paper and verified at the source, without human review. The map will receive a Zenodo DOI once one of the paper's authors has validated it with their ORCID.

What the map holds:

  • 1 repository of the authors' code, each at its verified commit, with its license and how the link was found in the paper;
  • 41 scripts, each with its path and the digest of its content;
  • 6 matches between paragraphs of the paper and lines of the code (method lexical-v1);
  • neither the text of the paper nor the code itself.

Its JSON (tracing-map.json) is deposited on Zenodo with its DOI once the map is validated.

Data

No dataset and no data link were found in the paper.

Data availability statement

The paper has a data availability statement. Its license (CC BY-NC-ND) does not allow reproducing it here; in short, from what the harvester recognized in it:

  • no repository, dataset or request procedure was recognized in it

Read it in the paper: doi.org/10.1038/s41467-026-73687-9.

Versions

The history of this record: each version stored by the harvester or made by a correction of its authors or of the maintainers of its code, and what changed in its facts. The texts of the paper (its abstract, its availability statements) are not part of it; versions that changed only those are not listed.

Version 1, 27 September 2026: the first record

Recorded: type, language, journal, volume, issue, pages, dates, 6 authors, 2 keywords, 1 funder, 17 references.

Cite

This paper

Cao, Q., Liu, S., Varghese, A. J., Darbon, J., Triantafyllou, M. S., & Karniadakis, G. E. (2026). Automatic selection of the best neural architecture for time series forecasting. Nature communications, 17(1), 7058. https://doi.org/10.1038/s41467-026-73687-9

BibTeX

@article{cao2026automatic,
author = {Cao, Qianying and Liu, Shanqing and Varghese, Alan John and Darbon, Jérôme and Triantafyllou, Michael S and Karniadakis, George Em},
title = {{Automatic selection of the best neural architecture for time series forecasting}},
journal = {Nature communications},
year = {2026},
month = jun,
volume = {17},
number = {1},
pages = {7058},
publisher = {Nature Portfolio},
issn = {2041-1723},
doi = {10.1038/s41467-026-73687-9},
url = {https://doi.org/10.1038/s41467-026-73687-9},
pmid = {42230615},
pmcid = {PMC13392406}
}

RIS

TY - JOUR
AU - Cao, Qianying
AU - Liu, Shanqing
AU - Varghese, Alan John
AU - Darbon, Jérôme
AU - Triantafyllou, Michael S
AU - Karniadakis, George Em
TI - Automatic selection of the best neural architecture for time series forecasting
T2 - Nature communications
J2 - Nat Commun
PY - 2026
DA - 2026/06/02
VL - 17
IS - 1
SP - 7058
SN - 2041-1723
PB - Nature Portfolio
DO - 10.1038/s41467-026-73687-9
UR - https://doi.org/10.1038/s41467-026-73687-9
LA - en
ER -

CSL-JSON

{
"id": "10.1038/s41467-026-73687-9",
"type": "article-journal",
"title": "Automatic selection of the best neural architecture for time series forecasting",
"container-title": "Nature communications",
"author": [
{
"family": "Cao",
"given": "Qianying"
},
{
"family": "Liu",
"given": "Shanqing"
},
{
"family": "Varghese",
"given": "Alan John"
},
{
"family": "Darbon",
"given": "Jérôme"
},
{
"family": "Triantafyllou",
"given": "Michael S"
},
{
"family": "Karniadakis",
"given": "George Em"
}
],
"container-title-short": "Nat Commun",
"volume": "17",
"issue": "1",
"page": "7058",
"DOI": "10.1038/s41467-026-73687-9",
"PMID": "42230615",
"PMCID": "PMC13392406",
"ISSN": "2041-1723",
"publisher": "Nature Portfolio",
"URL": "https://doi.org/10.1038/s41467-026-73687-9",
"language": "en",
"issued": {
"date-parts": [
[
2026,
6,
2
]
]
}
}

The tracing map gets a citation of its own once an author has validated it and it has a DOI.

Similar papers

The papers with a page that share the most with this one: the tools found in their code, their categories, datasets, cited references and authors, the rarest counting most.

[1] doi:10.1126/sciadv.aeb0404 [code]
MR-AIV reveals in vivo brain-wide fluid flow with physics-informed AI.
Journal: Science advances
In common: pandas, SciPy, Matplotlib, 1 other tool, author George Em Karniadakis
[2] doi:10.1177/20552076251398440 [code]
KAN-SleepNet: A deep learning model combining Kolmogorov-Arnold Networks and bidirectional LSTM for automated sleep staging using EEG signals.
Journal: Digital health
In common: PyTorch, pandas, Matplotlib, 1 other tool, methods / tools, 2 references
[3] doi:10.1038/s41467-025-61292-1 [code]
Utility of synthetic musculoskeletal gaits for generalizable healthcare applications.
Journal: Nature communications
In common: PyTorch, pandas, SciPy, 2 other tools, 2 references
[4] doi:10.1523/eneuro.0226-25.2025 [code]
Automated Classification of Sleep-Wake States and Seizures in Mice.
Journal: eNeuro
In common: pandas, SciPy, Matplotlib, 1 other tool, methods / tools, 2 references
[5] doi:10.3390/s25185690 [code]
Transparent EEG Analysis: Leveraging Autoencoders, Bi-LSTMs, and SHAP for Improved Neurodegenerative Diseases Detection.
Journal: Sensors (Basel, Switzerland)
In common: pandas, SciPy, Matplotlib, 1 other tool, methods / tools, 2 references
[6] doi:10.1016/j.patter.2025.101394 [code]
Decoding multi-joint hand movements from brain signals by learning a synergy-based neural manifold.
Journal: Patterns (New York, N.Y.)
In common: PyTorch, pandas, SciPy, 1 other tool, 2 references
[7] doi:10.3390/brainsci16040421 [code]
MambaKAN: An Interpretable Framework for Alzheimer's Disease Diagnosis via Selective State Space Modeling of Dynamic Functional Connectivity.
Journal: Brain sciences
In common: PyTorch, SciPy, Matplotlib, 1 other tool, 2 references
[8] doi:10.1073/pnas.2217744120
Artificial intelligence velocimetry reveals in vivo flow rates, pressure gradients, and shear stresses in murine perivascular flows.
Journal: Proceedings of the National Academy of Sciences of the United States of America
In common: author George Em Karniadakis
[9] doi:10.1371/journal.pcbi.1009728 [code]
Computational investigation of blood cell transport in retinal microaneurysms.
Journal: PLoS computational biology
In common: author George Em Karniadakis
[10] doi:10.1038/s41598-026-68186-2 [code]
NeuroStream: spectral-spatio-temporal deep learning for visual stimulus classification from EEG.
Journal: Scientific reports
In common: PyTorch, pandas, SciPy, 2 other tools, methods / tools, 1 reference

Contribute

The authors of this paper can claim it, correct its record and validate its tracing map, and the maintainers of its code (its owner, or a public member of its organization) correct what it says of their repository; anyone signed in can ask for its removal. Every request goes to OSCR's own machine, which answers it; your account page follows them.

Sign in with ORCID to claim this paper as one of its authors, correct its record or validate its tracing map: when the paper's metadata lists your ORCID iD, you are recognized at once. Maintainers of its code: sign in with GitHub, then claim the repository on your account page.

Add the badge to the code's README

OSCR: the badge of this paper

If you are an author or a maintainer of this paper's code, you can add this badge. It says what OSCR knows of the paper (its code, its validated tracing map) and links the code to this page. Paste one of these snippets under the README's title. Keep the two marks around it, so that the badge can be replaced in place and is never added twice. OSCR changes nothing in your repository.

Markdown
<!-- oscr-badge:start -->
[![OSCR](https://openscicode.org/badge/doi/10.1038/s41467-026-73687-9.svg)](https://openscicode.org/badge/doi/10.1038/s41467-026-73687-9)
<!-- oscr-badge:end -->
reStructuredText
.. oscr-badge:start

.. image:: https://openscicode.org/badge/doi/10.1038/s41467-026-73687-9.svg
   :target: https://openscicode.org/badge/doi/10.1038/s41467-026-73687-9
   :alt: OSCR

.. oscr-badge:end
HTML
<!-- oscr-badge:start -->
<a href="https://openscicode.org/badge/doi/10.1038/s41467-026-73687-9"><img src="https://openscicode.org/badge/doi/10.1038/s41467-026-73687-9.svg" alt="OSCR"></a>
<!-- oscr-badge:end -->

On GitHub, open the README, choose Edit (the pencil), paste the snippet under the title, then "Commit changes".

If you choose to, you can have the badge kept up to date for you, with oscr badge --write from the command line or with the GitHub Action in your repository's workflows. Every other form and option (a card, AsciiDoc, LaTeX, shields.io) is on the badge page.

Request its removal

To ask OSCR to remove this record, the copies of its authors' scripts or its tracing map, use the removal request page: signed in, you say who you are, what to remove and why, then review and confirm the request. Published rules decide every request (how).

Discussion, reproductions, activity

Discussion: code errors and mismatches between the code and the paper about this paper and its code, from readers and its authors, beside the paper and its tracing map.

Report a mismatch between the code and the paper · Report a code error (signed in)

Reproductions: reports from readers who ran the authors' code and did not get the paper's result, what they ran, with which environment, commit and data.

Report a reproduction failure (signed in)

Versions of its code: the releases of its code tied to a version of this paper (the preprint, the accepted manuscript, the version of record…), each with the tracing map versioned with it.

Activity: what happens around this paper: new versions of its record, its map's validation, discussions and reproductions. It opens with sign-in.