File size: 12,153 Bytes
ee6da62 | 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 | """
TD3B-specific MCTS modifications.
Extends the base MCTS to support directional rewards and confidence weighting.
"""
import numpy as np
import torch
from peptide_mcts import MCTS as BaseMCTS
from .td3b_scoring import TD3BRewardFunction, TD3BConfidenceWeighting
class TD3B_MCTS(BaseMCTS):
"""
TD3B version of MCTS that:
1. Uses gated directional rewards instead of multi-objective scalarization
2. Stores directional labels and confidence scores in the buffer
3. Applies confidence-weighted importance sampling
"""
def __init__(
self,
args,
diffusion_model,
td3b_reward_function: TD3BRewardFunction,
confidence_weighting: TD3BConfidenceWeighting,
mask_index: int,
buffer_size: int = 100,
noise=None,
tokenizer=None
):
"""
Args:
args: Configuration arguments
diffusion_model: MDLM model for sampling
td3b_reward_function: TD3BRewardFunction instance
confidence_weighting: TD3BConfidenceWeighting instance
mask_index: Token ID for masked positions
buffer_size: Maximum buffer size
noise: Noise schedule
tokenizer: Peptide tokenizer
"""
# Initialize base MCTS (will set self.rewardFunc later)
# Note: base MCTS expects 'policy_model' not 'diffusion_model'
# Create a minimal config object for base MCTS
class MinimalConfig:
def __init__(self):
self.noise = type('obj', (object,), {
'type': 'loglinear',
'sigma_min': 1e-4,
'sigma_max': 20
})()
config = MinimalConfig()
super().__init__(
args=args,
config=config,
policy_model=diffusion_model,
pretrained=diffusion_model, # Use same model
score_func_names=['affinity', 'gated_reward', 'placeholder1', 'placeholder2', 'placeholder3'] # 5 objectives
)
# Set TD3B-specific attributes
self.td3b_reward_func = td3b_reward_function
self.confidence_weighting = confidence_weighting
self.mask_index = mask_index
self.buffer_size = buffer_size
self.noise = noise
self.tokenizer = tokenizer if tokenizer is not None else diffusion_model.tokenizer
# Override num_obj to ensure it's 5 (matching our padded rewards)
self.num_obj = 5
# Override rewardFunc for compatibility
self.rewardFunc = self._td3b_reward_wrapper
def _td3b_reward_wrapper(self, input_seqs):
"""
Wrapper to make TD3BRewardFunction compatible with existing MCTS interface.
Returns (N, 5) array to match base MCTS expectations.
The 5 columns are: [affinity, gated_reward, 0, 0, 0] (padding last 3)
"""
import numpy as np
total_rewards, info = self.td3b_reward_func(input_seqs)
# info contains: 'affinities', 'confidences', 'score_vectors'
# Store confidences for later use (attach to self for access in updateBuffer)
self._last_confidences = info['confidences']
# Pad score_vectors from (N, 2) to (N, 5) to match base MCTS
# Original columns: [affinity, gated_reward]
# Padded to: [affinity, gated_reward, 0, 0, 0]
score_vectors = info['score_vectors'] # (N, 2)
padded = np.zeros((score_vectors.shape[0], 5))
padded[:, :2] = score_vectors # Copy affinity and gated_reward
return padded
def updateBuffer(self, x_final, log_rnd, score_vectors, childSequences):
"""
TD3B version: stores directional labels and confidence scores.
Args:
x_final: (B, L) final sequence tokens
log_rnd: (B,) log importance weights (trajectory-level)
score_vectors: (B, K) score arrays
childSequences: List of B SMILES strings
Returns:
traj_log_rnds: (B,) updated log importance weights
scalar_rewards: (B,) scalar rewards
"""
B = x_final.shape[0]
traj_log_rnds, scalar_rewards = [], []
# Get confidences from last reward computation
confidences = getattr(self, '_last_confidences', np.ones(B))
for i in range(B):
sv = np.asarray(score_vectors[i], dtype=float) # [affinity, gated_reward]
confidence = confidences[i]
# For TD3B, the "scalar reward" is the gated reward (second element)
scalar_reward = float(sv[1]) # gated_reward = g_蠄 路 (d* 路 sigmoid(f_蠁-0.5)/伪)
# Compute confidence-weighted importance weight
# w(y) = 魏(y) 路 exp(S_total / 伪)
# In log space: log w(y) = log 魏(y) + S_total / 伪
log_confidence = np.log(np.maximum(confidence, self.confidence_weighting.min_confidence))
traj_log_rnd = log_rnd[i] + (scalar_reward / self.args.alpha) + log_confidence
# Infer directional label from oracle (sign of gated reward)
# If gated_reward > 0, peptide is predicted as target direction
# This is approximate; in practice you might want to query f_蠁 directly
directional_label = np.sign(scalar_reward) if scalar_reward != 0 else 0.0
item = {
"x_final": x_final[i].clone(),
"log_rnd": traj_log_rnd.clone() if isinstance(traj_log_rnd, torch.Tensor) else torch.tensor(traj_log_rnd),
"final_reward": scalar_reward,
"score_vector": sv.copy(),
"seq": childSequences[i],
# TD3B-specific additions
"directional_label": directional_label,
"confidence": confidence,
}
# Pareto dominance filtering (same as base class)
from peptide_mcts import dominated_by, dominates
if any(dominated_by(sv, bi["score_vector"]) for bi in self.buffer):
self._debug_buffer_decision(sv, "rejected_dominated")
continue
# Remove dominated items
keep = []
for bi in self.buffer:
if not dominates(sv, bi["score_vector"]):
keep.append(bi)
self.buffer = keep
# Insert with capacity constraint
if len(self.buffer) < self.buffer_size:
self.buffer.append(item)
else:
# Replace worst item
worst_i = int(np.argmin([np.sum(bi["score_vector"]) for bi in self.buffer]))
self.buffer[worst_i] = item
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([torch.tensor(x) if not isinstance(x, torch.Tensor) else x for x in 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 forward(self, resetTree=False):
"""
TD3B version of forward that returns 7 values.
Returns:
x_final: (N, L) sequence tokens
log_rnd: (N,) log importance weights
final_rewards: (N,) scalar rewards
score_vectors: (N, K) score arrays
sequences: List of N SMILES strings
directional_labels: (N,) directional labels
confidences: (N,) confidence scores
"""
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, directional_labels, confidences = self.consolidateBuffer()
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, directional_labels, confidences
def consolidateBuffer(self):
"""
TD3B version: includes directional labels and confidences.
Returns:
x_final: (N, L) sequence tokens
log_rnd: (N,) log importance weights
final_rewards: (N,) scalar rewards
score_vectors: (N, K) score arrays
sequences: List of N SMILES strings
directional_labels: (N,) directional labels
confidences: (N,) confidence scores
"""
# Handle empty buffer case - return empty tensors/arrays
if len(self.buffer) == 0:
import logging
logger = logging.getLogger(__name__)
logger.warning("MCTS buffer is empty - no valid sequences found. Returning empty results.")
# Return empty tensors/arrays with correct shapes
# Use policy_model (set by base MCTS class) to get device
device = self.policy_model.device if hasattr(self.policy_model, 'device') else 'cpu'
return (
torch.empty(0, 0, dtype=torch.long, device=device), # x_final: (0, 0)
torch.empty(0, dtype=torch.float32, device=device), # log_rnd: (0,)
np.empty(0, dtype=np.float32), # final_rewards: (0,)
np.empty((0, 0), dtype=np.float32), # score_vectors: (0, 0)
[], # sequences: empty list
np.empty(0, dtype=np.float32), # directional_labels: (0,)
np.empty(0, dtype=np.float32) # confidences: (0,)
)
x_final = []
log_rnd = []
final_rewards = []
score_vectors = []
sequences = []
directional_labels = []
confidences = []
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"])
directional_labels.append(item.get("directional_label", 0.0))
confidences.append(item.get("confidence", 1.0))
x_final = torch.stack(x_final, dim=0) # (N, L)
log_rnd = torch.stack(log_rnd, dim=0).to(dtype=torch.float32) # (N,)
final_rewards = np.stack(final_rewards, axis=0).astype(np.float32)
score_vectors = np.stack(score_vectors, axis=0).astype(np.float32)
directional_labels = np.array(directional_labels, dtype=np.float32)
confidences = np.array(confidences, dtype=np.float32)
return x_final, log_rnd, final_rewards, score_vectors, sequences, directional_labels, confidences
def create_td3b_mcts(
args,
diffusion_model,
td3b_reward_function: TD3BRewardFunction,
alpha: float = 0.1,
**kwargs
) -> TD3B_MCTS:
"""
Factory function to create TD3B MCTS instance.
Args:
args: Configuration arguments
diffusion_model: MDLM model
td3b_reward_function: TD3BRewardFunction instance
alpha: Temperature for importance weighting
**kwargs: Additional MCTS arguments
Returns:
mcts: TD3B_MCTS instance
"""
# Create confidence weighting module
confidence_weighting = TD3BConfidenceWeighting(
alpha=alpha,
min_confidence=0.1
)
# Create TD3B MCTS
mcts = TD3B_MCTS(
args=args,
diffusion_model=diffusion_model,
td3b_reward_function=td3b_reward_function,
confidence_weighting=confidence_weighting,
**kwargs
)
return mcts
|