File size: 27,418 Bytes
5e90249 c511e34 5e90249 c511e34 5e90249 c511e34 5e90249 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 | import numpy as np
import torch
import torch.nn.functional as F
import numpy as np
import random as rd
from utils.app import PeptideAnalyzer
from utils.timer import StepTimer
from scoring.scoring_functions import ScoringFunctions
import noise_schedule
### for peptide multi-objective ###
def dominates(a, b):
a = np.asarray(a); b = np.asarray(b)
return np.all(a >= b) and np.any(a > b)
def dominated_by(a, b):
return dominates(b, a)
def updateParetoFront(paretoFront, node, scoreVector, totalSize=None, eps=1e-12):
"""
Maintain a non-dominated set (Pareto front) of (node -> scoreVector).
- Accept 'node' iff it is NOT dominated by any node in the set.
- Remove any nodes that ARE dominated by 'node'.
- Skip insertion if an equal point already exists (within eps).
- If totalSize is given and the archive exceeds it, drop the item
with the smallest sum(scoreVector) as a simple tie-breaker.
Args:
paretoFront (dict): {node: scoreVector}
node: candidate node (used as dict key)
scoreVector (array-like): candidate scores (to be maximized)
totalSize (int|None): optional max size for the archive
eps (float): tolerance for equality/inequality checks
Returns:
dict: updated paretoFront
"""
s = np.asarray(scoreVector, dtype=float)
def dominates(a, b):
# a >= b in all coords and > in at least one (with tolerance)
return np.all(a >= b - eps) and np.any(a > b + eps)
def equal(a, b):
return np.all(np.abs(a - b) <= eps)
# reject if candidate is dominated by any node already in the set
for v in paretoFront.values():
v = np.asarray(v, dtype=float)
if dominates(v, s):
return paretoFront # no change
# remove any nodes dominated by candidate node
survivors = {}
#has_equal = False
for k, v in paretoFront.items():
v_arr = np.asarray(v, dtype=float)
if dominates(s, v_arr):
continue # drop dominated incumbent
"""if equal(s, v_arr):
has_equal = True # skip duplicate insertion later"""
survivors[k] = v_arr
# if an equal point exists, keep survivors as-is (no duplicate)
"""if has_equal:
return survivors"""
# insert node
survivors[node] = s
# delete nodes if larger than total size
if totalSize is not None and totalSize > 0 and len(survivors) > totalSize:
# remove the item with the smallest sum(scoreVector)
keys = list(survivors.keys())
sums = np.array([np.sum(np.asarray(survivors[k], dtype=float)) for k in keys])
drop_idx = int(np.argmin(sums))
del survivors[keys[drop_idx]]
return survivors
### BEGINNING OF NODE CLASS ###
class Node:
"""
Node class: partially unmasked sequence
- parentNode: Node object at previous time step
- childNodes: set of M Node objects generated from sampling M distinct unmasking schemes
- totalReward: vector of cumulative rewards for all K objectives
- visits: number of times the node has been visited by an interation
- path: array of partially unmasked SMILES strings leading to the node from the completely masked root node
- timestep: the time step where the sequence was sampled
"""
def __init__(self, args, tokens=None, log_rnd=None, log_policy_step=None, log_pretrained_step=None, parentNode=None, childNodes=None, totalReward=None, timestep=None):
self.args = args
self.parentNode = parentNode
# fixed child node list creation
self.childNodes = [] if childNodes is None else childNodes
self.log_rnd = log_rnd # stores the log_rnd up to that step
#self.log_p0 = 0 # stores the log probabiltiy of the unmasking step from the previous iteration
self.log_policy_step = log_policy_step # stores the log probability of the unmasking step under the current policy
self.log_pretrained_step = log_pretrained_step
# initialize total rewards to the reward of the roll out unmasked sequence
if totalReward is not None:
self.totalReward = totalReward # potential reward of the node based on generated children
else:
self.totalReward = np.zeros(self.args.num_obj)
# set initial visits to 1
self.visits = 1
# set timestep (value between 0 and num_steps)
self.timestep = timestep
# dict with 'seqs' as token array and 'attention_mask'
self.tokens = tokens
def selectNode(self):
"""
Selects a node to move to among the children nodes based on select score
"""
# extract the status of the current node
nodeStatus = self.getExpandStatus()
# if the node is a legal non-leaf node
if (nodeStatus == 3):
# initialize array that will store select score vectors of each child node
paretoFront = {}
for childNode in self.childNodes:
childStatus = childNode.getExpandStatus()
# only append child if it is legal leaf node (expandable) or legal non-leaf node
if childStatus == 2 or childStatus == 3:
selectScore = childNode.calcSelectScore()
paretoFront = updateParetoFront(paretoFront, childNode, selectScore)
selected = rd.choice(list(paretoFront.keys()))
# return selected child node and status
return selected, selected.getExpandStatus()
# if node is not valid non-leaf node
return self, nodeStatus
def addChildNode(self, tokens, log_rnd, log_policy_step, log_pretrained_step, totalReward):
""""
Adds a child node:
log_rnd: log_rnd of the path up to the added child node
log_policy_step: scalar value of the log-prob of sampling the step under the policy
log_pretrained_step: scalar value of the log-prob of sampling the step under the pretrained model
"""
child = Node(args=self.args,
tokens=tokens,
log_rnd = log_rnd,
log_policy_step=log_policy_step,
log_pretrained_step=log_pretrained_step,
parentNode=self,
childNodes=[],
totalReward=totalReward,
timestep=self.timestep+1)
self.childNodes.append(child)
return child
def update_logrnd(self, log_policy_step, log_rnd):
self.log_policy_step = log_policy_step
self.log_rnd = log_rnd
def updateNode(self, rewards):
"""
Updates the cumulative rewards vector with the reward vector at a descendent leaf node.
Increments the number of visits to the node.
"""
self.visits += 1
self.totalReward += rewards # singleton tensor
def calcSelectScore(self):
"""
Calculates the select score for the node from the cumulative rewards vector and number of visits.
- c: determines the degree of exploration
- minSelectScore: determines the
"""
scaling = 0.1 # scaling of the second term in the select score
# K-dimensional vector of normalized rewards for each objective
normRewards = self.totalReward / self.visits
# scales the cumulative reward by the sampling probability
return normRewards + (scaling * self.log_policy_step.detach().cpu().item() * np.sqrt(self.parentNode.visits) / self.visits)
def getExpandStatus(self):
"""
Returns an integer indicating whether the node is a:
1. terminal node (sequence is fully unmasked)
2. legal leaf node (partially unmasked sequence that can be expanded)
3. legal non-leaf node (already expanded sequence with M child nodes)
"""
if self.timestep == self.args.total_num_steps:
return 1
elif (self.timestep < self.args.total_num_steps) and (len(self.childNodes) == 0):
return 2
return 3
### END OF NODE CLASS ###
### BEGINNING OF MCTS CLASS ###
class MCTS:
def __init__(self, args, config, policy_model, pretrained, score_func_names=[], prot_seqs=None, rootNode=None):
self.timer = StepTimer(policy_model.device)
self.device = policy_model.device
self.args = args
self.config = config
self.noise = noise_schedule.get_noise(config)
self.time_conditioning = args.time_conditioning
self.num_obj = len(score_func_names)
self.mask_index = policy_model.mask_index
masked_seq = torch.ones((self.args.seq_length), device = self.device) * self.mask_index
masked_tokens = {'seqs': masked_seq.to(dtype=torch.long), 'attention_mask': torch.ones_like(masked_seq).to(self.device)}
if rootNode is None:
self.rootNode = Node(self.args, tokens = masked_tokens,
log_rnd=torch.zeros((), device=self.device),
log_policy_step=torch.zeros((), device=self.device),
log_pretrained_step=torch.zeros((), device=self.device),
totalReward=np.zeros(self.num_obj), timestep=0)
else:
self.rootNode = rootNode # stores the root node of the tree
# dictionary:
# "seq": final unmasked sequence
# "traj": list of (N_steps, L)
# "reward": reward of the trajectory
self.buffer = [] # List[Dict[str, Any]]
self.buffer_size = args.buffer_size
self.num_steps = args.total_num_steps
#self.num_sequences = args.num_sequences
# pretrained model
self.pretrained = pretrained
# the policy model that we want to finetune
self.policy_model = policy_model
#self.tokenizer = policy_model.tokenizer
self.device = policy_model.device
self.sequence_length = args.seq_length
self.num_iter = args.num_iter
self.num_children = args.num_children
# score functions
self.rewardFunc = ScoringFunctions(score_func_names, prot_seqs, device=args.device)
self.iter_num = 0
self.reward_log = [] # stores scalarized total rewards
self.logrnd_log = []
# stores each objective
self.valid_fraction_log = []
self.affinity1_log = []
self.affinity2_log = []
self.permeability_log = []
self.sol_log = []
self.hemo_log = []
self.nf_log = []
self.policy_model.eval()
self.pretrained.eval()
# for peptides
self.analyzer = PeptideAnalyzer()
self.tokenizer = policy_model.tokenizer
def reset(self, resetTree):
self.iter_num = 0
self.buffer = []
self.reward_log = []
self.logrnd_log = []
# reset logs for each objective
self.valid_fraction_log = []
self.affinity1_log = []
self.affinity2_log = []
self.permeability_log = []
self.sol_log = []
self.hemo_log = []
self.nf_log = []
# add option to continue with the same tree
if resetTree:
masked_seq = torch.ones((self.args.seq_length), device = self.device) * self.mask_index
masked_tokens = {'seqs': masked_seq.to(dtype=torch.long), 'attention_mask': torch.ones_like(masked_seq).to(self.device)}
self.rootNode = Node(self.args, tokens = masked_tokens,
log_rnd=torch.zeros((), device=self.device),
log_policy_step=torch.zeros((), device=self.device),
log_pretrained_step=torch.zeros((), device=self.device),
totalReward=np.zeros(self.num_obj), timestep=0)
def forward(self, resetTree=False):
self.reset(resetTree)
while (self.iter_num < self.num_iter):
self.iter_num += 1
# traverse the tree form the root node until a leaf node
with self.timer.section("select"):
leafNode, _ = self.select(self.rootNode)
# expand leaf node into num_children partially unmasked sequences at the next timestep
with self.timer.section("expand"):
self.expand(leafNode)
final_x, log_rnd, final_rewards, score_vectors, sequences = self.consolidateBuffer()
# return final_seqs (B, L), log_rnd (B, ), and final rewards (B, )
rows = self.timer.summary()
print("\n=== Timing summary (by total time) ===")
for name, cnt, total, mean, p50, p95 in rows:
print(f"{name:30s} n={cnt:5d} total={total:8.3f}s mean={mean*1e3:7.2f}ms "
f"p50={p50*1e3:7.2f}ms p95={p95*1e3:7.2f}ms")
return final_x, log_rnd, final_rewards, score_vectors, sequences
# new updateBuffer
def _debug_buffer_decision(self, sv, reason, extra=None):
if extra is None: extra = {}
print(f"[BUFFER] reason={reason} sv={np.round(sv,4)} "
f"buf_len={len(self.buffer)} extra={extra}")
def updateBuffer(self, x_final, log_rnd, score_vectors, validSequences):
B = x_final.shape[0]
traj_log_rnds, scalar_rewards = [], []
for i in range(B):
sv = np.asarray(score_vectors[i], dtype=float)
# determine how to scalarize the multi-objective rewards
if self.args.scalarization == "normalized":
pass
elif self.args.scalarization == "weighted":
pass
else:
scalar_reward = float(np.sum(sv))
traj_log_rnd = log_rnd[i] + (scalar_reward / self.args.alpha) # scale down by alpha
item = {
"x_final": x_final[i].clone(), # clone?
"log_rnd": traj_log_rnd.clone(),
"final_reward": scalar_reward,
"score_vector": sv.copy(),
"seq": validSequences[i],
}
# Drop if dominated by any existing
if any(dominated_by(sv, bi["score_vector"]) for bi in self.buffer):
# for debugging
self._debug_buffer_decision(sv, "rejected_dominated")
continue
# Remove any existing that this candidate dominates
keep = []
for bi in self.buffer:
if not dominates(sv, bi["score_vector"]):
keep.append(bi)
self.buffer = keep
# Insert with capacity rule
if len(self.buffer) < self.buffer_size:
self.buffer.append(item)
else:
# tie-breaker: replace the worst by a simple heuristic (min sum)
worst_i = int(np.argmin([np.sum(bi["score_vector"]) for bi in self.buffer]))
self.buffer[worst_i] = item
# for debugging
self._debug_buffer_decision(sv, "inserted", {"new_len": len(self.buffer)})
traj_log_rnds.append(traj_log_rnd)
scalar_rewards.append(scalar_reward)
traj_log_rnds = torch.stack(traj_log_rnds, dim=0) if traj_log_rnds else torch.empty(0)
scalar_rewards = np.asarray(scalar_rewards, dtype=float)
return traj_log_rnds, scalar_rewards
def consolidateBuffer(self):
"""
returns x_final, log_rnd, and final_rewards in tensors
"""
x_final = []
log_rnd = []
final_rewards = []
score_vectors = []
sequences = []
for item in self.buffer:
x_final.append(item["x_final"])
log_rnd.append(item["log_rnd"])
final_rewards.append(item["final_reward"])
score_vectors.append(item["score_vector"])
sequences.append(item["seq"])
x_final = torch.stack(x_final, dim=0) # (B, L)
log_rnd = torch.stack(log_rnd, dim=0).to(dtype=torch.float32) # (B)
final_rewards = np.stack(final_rewards, axis=0).astype(np.float32)
score_vectors = np.stack(score_vectors, axis=0).astype(np.float32)
return x_final, log_rnd, final_rewards, score_vectors, sequences
def isPathEnd(self, path, maxDepth):
"""
Checks if the node is completely unmasked (ie. end of path)
or if the path is at the max depth
"""
if (path[-1] != self.mask_index).all():
return True
elif len(path) >= maxDepth:
return True
return False
def select(self, currNode, eps=1e-5):
"""
Traverse the tree from the root node until reaching a legal leaf node
"""
updated_log_rnd = torch.zeros((), device=self.device)
while True:
currNode, nodeStatus = currNode.selectNode()
if currNode.parentNode is not None:
# compute new log_policy
child_tokens = currNode.tokens['seqs'].to(self.device)
attn_mask = currNode.tokens['attention_mask'].to(self.device)
parent = currNode.parentNode
parent_tokens = parent.tokens['seqs'].to(self.device)
t = torch.ones(1, device = self.device)
dt = (1 - eps) / self.num_steps
with torch.no_grad():
with self.timer.section("select.compute_log_policy"):
updated_log_policy_step = self.policy_model.compute_log_policy(parent_tokens,
child_tokens,
t=t, dt=dt)
updated_log_rnd += updated_log_policy_step
currNode.update_logrnd(updated_log_policy_step, updated_log_rnd) # update log_rnd
if nodeStatus != 3:
return currNode, nodeStatus
def expand(self, parentNode, eps=1e-5):
"""
Sample unmasking steps from the pre-trained MDLM
adds num_children partially unmasked sequences to the children of the parentNode
"""
num_children = self.num_children
# initialize child rewards that will be added to total rewards
# compute number of rollout steps
# if parentNode.timestep = self.num_steps then num_rollout_steps = 1
num_rollout_steps = self.num_steps - parentNode.timestep
# array of rollout timesteps from the timestep of parent node to 0
rollout_t = torch.linspace(1, eps, self.num_steps + 1, device=self.device)
dt = (1 - eps) / self.num_steps
# initialize x and attn_mask
x = parentNode.tokens['seqs'].to(self.device)
attn_mask = parentNode.tokens['attention_mask'].to(self.device)
parent_log_rnd = parentNode.log_rnd # stores the log_rnd up to parent node
t = rollout_t[parentNode.timestep] * torch.ones(1, 1, device = self.device)
# sample M child sequences and compute their log probabilities
with torch.no_grad():
with self.timer.section("expand.batch_mcts_reverse_step"):
_, x_children, child_log_policy_step, child_log_pretrained_step = \
self.policy_model.batch_mcts_reverse_step(token_array=x,
t=t, dt=dt,
batch_size=num_children,
pretrained=self.pretrained)
# compute weight of the step (num_children, 1)
child_log_rnd = (parent_log_rnd + (child_log_pretrained_step - child_log_policy_step)).to(self.device)
x_rollout = x_children
traj_log_rnd = child_log_rnd # initialize log_rnd for entire rolled out trajectory
# rollout under the policy and compute the log ratio at each step
with self.timer.section("expand.rollout_total"):
for i in range(1, num_rollout_steps):
t = rollout_t[parentNode.timestep + i] * torch.ones(num_children, 1, device = self.device)
with torch.no_grad():
_, x_next, log_policy_step, log_pretrained_step = \
self.policy_model.mcts_reverse_step(x_rollout,
t=t, dt=dt,
pretrained=self.pretrained)
# add the rollout step
traj_log_rnd += log_pretrained_step - log_policy_step
x_rollout = x_next
# if mask token remains, fully unmask
mask_positions = (x_rollout == self.mask_index) # (B, L) bool
# does **any** mask remain in any sequence
any_mask_global = mask_positions.any().item() # true if mask remains
if any_mask_global:
with torch.no_grad():
with self.timer.section("expand.noise_removal"):
log_p, x_next, log_policy_step, log_pretrained_step = \
self.policy_model.mcts_noise_removal(x_rollout,
t=t, dt=dt,
pretrained=self.pretrained)
traj_log_rnd += log_pretrained_step - log_policy_step
x_rollout = x_next
# stores the string sequences for reward evaluation
with self.timer.section("expand.decode"):
childSequences = self.tokenizer.batch_decode(x_rollout)
## FOR PEPTIDES ONLY ##
valid_x_children = []
valid_x_final = []
validSequences = []
valid_traj_log_rnd = []
with self.timer.section("expand.filter_is_peptide"):
for i in range(num_children):
# string sequence
childSeq = childSequences[i]
# check if the peptide is valid
if self.analyzer.is_peptide(childSeq):
valid_x_children.append(x_children[i])
valid_x_final.append(x_rollout[i])
validSequences.append(childSeq)
valid_traj_log_rnd.append(traj_log_rnd[i])
else:
childTokens = {'seqs': x_children[i].to(dtype=torch.long), 'attention_mask': attn_mask}
parentNode.addChildNode(tokens=childTokens,
log_rnd=child_log_rnd[i],
log_policy_step=child_log_policy_step[i],
log_pretrained_step=child_log_pretrained_step[i],
totalReward=np.zeros(self.num_obj))
del traj_log_rnd
if (len(validSequences) != 0):
# add scores to log
with self.timer.section("expand.scoring_functions"):
score_vectors = self.rewardFunc(input_seqs=validSequences) # (num_children, num_objectives)
average_scores = score_vectors.T
self.affinity1_log.append(average_scores[0])
self.sol_log.append(average_scores[1])
self.hemo_log.append(average_scores[2])
self.nf_log.append(average_scores[3])
self.permeability_log.append(average_scores[4])
else:
# set the values added to log as 0s if there are no valid sequences
self.affinity1_log.append(np.zeros((self.num_obj, self.num_children)))
self.sol_log.append(np.zeros((self.num_obj, self.num_children)))
self.hemo_log.append(np.zeros((self.num_obj, self.num_children)))
self.nf_log.append(np.zeros((self.num_obj, self.num_children)))
self.permeability_log.append(np.zeros((self.num_obj, self.num_children)))
# convert to tensor
if len(valid_x_final) == 0:
# log and bail out gracefully for this expansion
self.valid_fraction_log.append(0.0)
return
valid_x_final = torch.stack(valid_x_final, dim=0)
valid_traj_log_rnd = torch.stack(valid_traj_log_rnd, dim=0)
# update buffer and get rewards
with self.timer.section("expand.update_buffer"):
traj_log_rnds, scalar_rewards = self.updateBuffer(valid_x_final, valid_traj_log_rnd, score_vectors, validSequences)
allChildReward = np.zeros_like(score_vectors[0])
for i in range(len(score_vectors)):
reward = score_vectors[i]
# add to all child reward vector for backprop
allChildReward += reward # (num_objectives,)
# create node for sequence and add to the children node of parent
childTokens = {'seqs': valid_x_children[i].to(dtype=torch.long), 'attention_mask': attn_mask}
parentNode.addChildNode(tokens=childTokens,
log_rnd=child_log_rnd[i],
log_policy_step=child_log_policy_step[i],
log_pretrained_step=child_log_pretrained_step[i],
totalReward=reward)
### END OF FOR PEPTIDES ONLY ###
valid_fraction = len(validSequences) / num_children
self.valid_fraction_log.append(valid_fraction)
# debugging
print(f"[EXPAND] iter={self.iter_num} parent_t={parentNode.timestep} "
f"num_children={num_children} valid={len(validSequences)} any_mask={any_mask_global}")
if score_vectors is not None:
print(f"[SCORES] min={np.min(score_vectors,0)} max={np.max(score_vectors,0)} "
f"nan_any={np.isnan(score_vectors).any()}")
# end debugging
self.reward_log.append(scalar_rewards)
self.logrnd_log.append(traj_log_rnds.detach().cpu().numpy())
allChildReward = allChildReward / len(validSequences) # normalize by number of valid children
# backpropogate all child rewards
with self.timer.section("expand.backprop"):
self.backprop(parentNode, allChildReward)
def backprop(self, node, allChildReward):
# backpropogate rewards through the path leading to the leaf node from the root
while node:
node.updateNode(allChildReward)
node = node.parentNode |