File size: 12,529 Bytes
5a87d8d |
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 |
import os
import sys
sys.path.append("./BranchSBM")
import torch
import wandb
import matplotlib.pyplot as plt
import pytorch_lightning as pl
from torch.optim import AdamW
from torchmetrics.functional import mean_squared_error
from torchdyn.core import NeuralODE
from networks.utils import flow_model_torch_wrapper
from utils import wasserstein_distance, plot_lidar
from branchsbm.ema import EMA
class BranchFlowNetTrainBase(pl.LightningModule):
def __init__(
self,
flow_matcher,
flow_nets,
skipped_time_points=None,
ot_sampler=None,
args=None,
):
super().__init__()
self.args = args
self.flow_matcher = flow_matcher
self.flow_nets = flow_nets # list of flow networks for each branch
self.ot_sampler = ot_sampler
self.skipped_time_points = skipped_time_points
self.optimizer_name = args.flow_optimizer
self.lr = args.flow_lr
self.weight_decay = args.flow_weight_decay
self.whiten = args.whiten
self.working_dir = args.working_dir
#branching
self.branches = len(flow_nets)
def forward(self, t, xt, branch_idx):
# output velocity given branch_idx
return self.flow_nets[branch_idx](t, xt)
def _compute_loss(self, main_batch):
x0s = [main_batch["x0"][0]]
w0s = [main_batch["x0"][1]]
x1s_list = []
w1s_list = []
if self.branches > 1:
for i in range(self.branches):
x1s_list.append([main_batch[f"x1_{i+1}"][0]])
w1s_list.append([main_batch[f"x1_{i+1}"][1]])
else:
x1s_list.append([main_batch["x1"][0]])
w1s_list.append([main_batch["x1"][1]])
assert len(x1s_list) == self.branches, "Mismatch between x1s_list and expected branches"
loss = 0
for branch_idx in range(self.branches):
ts, xts, uts = self._process_flow(x0s, x1s_list[branch_idx], branch_idx)
t = torch.cat(ts)
xt = torch.cat(xts)
ut = torch.cat(uts)
vt = self(t[:, None], xt, branch_idx)
loss += mean_squared_error(vt, ut)
return loss
def _process_flow(self, x0s, x1s, branch_idx):
ts, xts, uts = [], [], []
t_start = self.timesteps[0]
for i, (x0, x1) in enumerate(zip(x0s, x1s)):
x0, x1 = torch.squeeze(x0), torch.squeeze(x1)
if self.ot_sampler is not None:
x0, x1 = self.ot_sampler.sample_plan(
x0,
x1,
replace=True,
)
if self.skipped_time_points and i + 1 >= self.skipped_time_points[0]:
t_start_next = self.timesteps[i + 2]
else:
t_start_next = self.timesteps[i + 1]
# edit to sample from correct flow matcher
t, xt, ut = self.flow_matcher.sample_location_and_conditional_flow(
x0, x1, t_start, t_start_next, branch_idx
)
ts.append(t)
xts.append(xt)
uts.append(ut)
t_start = t_start_next
return ts, xts, uts
def training_step(self, batch, batch_idx):
if self.args.data_type in ["scrna", "tahoe"]:
main_batch = batch[0]["train_samples"][0]
else:
main_batch = batch["train_samples"][0]
print("Main batch length")
print(len(main_batch["x0"]))
self.timesteps = torch.linspace(0.0, 1.0, len(main_batch["x0"])).tolist()
loss = self._compute_loss(main_batch)
if self.flow_matcher.alpha != 0:
self.log(
"FlowNet/mean_geopath_cfm",
(self.flow_matcher.geopath_net_output.abs().mean()),
on_step=False,
on_epoch=True,
prog_bar=True,
)
self.log(
"FlowNet/train_loss_cfm",
loss,
on_step=False,
on_epoch=True,
prog_bar=True,
logger=True,
)
return loss
def validation_step(self, batch, batch_idx):
if self.args.data_type in ["scrna", "tahoe"]:
main_batch = batch[0]["val_samples"][0]
else:
main_batch = batch["val_samples"][0]
self.timesteps = torch.linspace(0.0, 1.0, len(main_batch["x0"])).tolist()
val_loss = self._compute_loss(main_batch)
self.log(
"FlowNet/val_loss_cfm",
val_loss,
on_step=False,
on_epoch=True,
prog_bar=True,
logger=True,
)
return val_loss
def optimizer_step(self, *args, **kwargs):
super().optimizer_step(*args, **kwargs)
for net in self.flow_nets:
if isinstance(net, EMA):
net.update_ema()
def configure_optimizers(self):
if self.optimizer_name == "adamw":
optimizer = AdamW(
self.parameters(),
lr=self.lr,
weight_decay=self.weight_decay,
)
elif self.optimizer_name == "adam":
optimizer = torch.optim.Adam(
self.parameters(),
lr=self.lr,
)
return optimizer
class FlowNetTrainTrajectory(BranchFlowNetTrainBase):
def test_step(self, batch, batch_idx):
data_type = self.args.data_type
node = NeuralODE(
flow_model_torch_wrapper(self.flow_nets),
solver="euler",
sensitivity="adjoint",
atol=1e-5,
rtol=1e-5,
)
t_exclude = self.skipped_time_points[0] if self.skipped_time_points else None
if t_exclude is not None:
traj = node.trajectory(
batch[t_exclude - 1],
t_span=torch.linspace(
self.timesteps[t_exclude - 1], self.timesteps[t_exclude], 101
),
)
X_mid_pred = traj[-1]
traj = node.trajectory(
batch[t_exclude - 1],
t_span=torch.linspace(
self.timesteps[t_exclude - 1],
self.timesteps[t_exclude + 1],
101,
),
)
EMD = wasserstein_distance(X_mid_pred, batch[t_exclude], p=1)
self.final_EMD = EMD
self.log("test_EMD", EMD, on_step=False, on_epoch=True, prog_bar=True)
class FlowNetTrainCell(BranchFlowNetTrainBase):
def test_step(self, batch, batch_idx):
x0 = batch[0]["test_samples"][0]["x0"][0] # [B, D]
dataset_points = batch[0]["test_samples"][0]["dataset"][0] # full dataset, [N, D]
t_span = torch.linspace(0, 1, 101)
all_trajs = []
for i, flow_net in enumerate(self.flow_nets):
node = NeuralODE(
flow_model_torch_wrapper(flow_net),
solver="euler",
sensitivity="adjoint",
)
with torch.no_grad():
traj = node.trajectory(x0, t_span).cpu() # [T, B, D]
if self.whiten:
traj_shape = traj.shape
traj = traj.reshape(-1, traj.shape[-1])
traj = self.trainer.datamodule.scaler.inverse_transform(
traj.cpu().detach().numpy()
).reshape(traj_shape)
dataset_points = self.trainer.datamodule.scaler.inverse_transform(
dataset_points.cpu().detach().numpy()
)
traj = torch.tensor(traj)
traj = torch.transpose(traj, 0, 1) # [B, T, D]
all_trajs.append(traj)
dataset_2d = dataset_points[:, :2] if isinstance(dataset_points, torch.Tensor) else dataset_points[:, :2]
# ===== Plot all 2D trajectories together with dataset and start/end points =====
fig, ax = plt.subplots(figsize=(6, 5))
dataset_2d = dataset_2d.cpu().numpy()
ax.scatter(dataset_2d[:, 0], dataset_2d[:, 1], c="gray", s=1, alpha=0.5, label="Dataset", zorder=1)
for traj in all_trajs:
traj_2d = traj[..., :2] # [B, T, 2]
for i in range(traj_2d.shape[0]):
ax.plot(traj_2d[i, :, 0], traj_2d[i, :, 1], alpha=0.8, zorder=2)
ax.scatter(traj_2d[i, 0, 0], traj_2d[i, 0, 1], c='green', s=10, label="t=0" if i == 0 else "", zorder=3)
ax.scatter(traj_2d[i, -1, 0], traj_2d[i, -1, 1], c='red', s=10, label="t=1" if i == 0 else "", zorder=3)
ax.set_title("All Branch Trajectories (2D) with Dataset")
ax.set_xlabel("x")
ax.set_ylabel("y")
plt.axis("equal")
handles, labels = ax.get_legend_handles_labels()
if labels:
ax.legend()
save_path = f'./figures/{self.args.data_name}'
os.makedirs(save_path, exist_ok=True)
plt.savefig(f'{save_path}/{self.args.data_name}_all_branches.png', dpi=300)
plt.close()
# ===== Plot each 2D trajectory separately with dataset and endpoints =====
for i, traj in enumerate(all_trajs):
traj_2d = traj[..., :2]
fig, ax = plt.subplots(figsize=(6, 5))
ax.scatter(dataset_2d[:, 0], dataset_2d[:, 1], c="gray", s=1, alpha=0.5, label="Dataset", zorder=1)
for j in range(traj_2d.shape[0]):
ax.plot(traj_2d[j, :, 0], traj_2d[j, :, 1], alpha=0.9, zorder=2)
ax.scatter(traj_2d[j, 0, 0], traj_2d[j, 0, 1], c='green', s=12, label="t=0" if j == 0 else "", zorder=3)
ax.scatter(traj_2d[j, -1, 0], traj_2d[j, -1, 1], c='red', s=12, label="t=1" if j == 0 else "", zorder=3)
ax.set_title(f"Branch {i + 1} Trajectories (2D) with Dataset")
ax.set_xlabel("x")
ax.set_ylabel("y")
plt.axis("equal")
handles, labels = ax.get_legend_handles_labels()
if labels:
ax.legend()
plt.savefig(f'{save_path}/{self.args.data_name}_branch_{i + 1}.png', dpi=300)
plt.close()
class FlowNetTrainLidar(BranchFlowNetTrainBase):
def test_step(self, batch, batch_idx):
main_batch = batch["test_samples"][0]
metric_batch = batch["metric_samples"][0]
x0 = main_batch["x0"][0] # [B, D]
cloud_points = main_batch["dataset"][0] # full dataset, [N, D]
t_span = torch.linspace(0, 1, 101)
all_trajs = []
for i, flow_net in enumerate(self.flow_nets):
node = NeuralODE(
flow_model_torch_wrapper(flow_net),
solver="euler",
sensitivity="adjoint",
)
with torch.no_grad():
traj = node.trajectory(x0, t_span).cpu() # [T, B, D]
if self.whiten:
traj_shape = traj.shape
traj = traj.reshape(-1, 3)
traj = self.trainer.datamodule.scaler.inverse_transform(
traj.cpu().detach().numpy()
).reshape(traj_shape)
traj = torch.tensor(traj)
traj = torch.transpose(traj, 0, 1) # [B, T, D]
all_trajs.append(traj)
# Inverse-transform the point cloud once
if self.whiten:
cloud_points = torch.tensor(
self.trainer.datamodule.scaler.inverse_transform(
cloud_points.cpu().detach().numpy()
)
)
# ===== Plot all trajectories together =====
fig = plt.figure(figsize=(6, 5))
ax = fig.add_subplot(111, projection="3d", computed_zorder=False)
ax.view_init(elev=30, azim=-115, roll=0)
for i, traj in enumerate(all_trajs):
plot_lidar(ax, cloud_points, xs=traj, branch_idx=i)
plt.savefig('./figures/lidar/lidar_all_branches.png', dpi=300)
plt.close()
# ===== Plot each trajectory separately =====
for i, traj in enumerate(all_trajs):
fig = plt.figure(figsize=(6, 5))
ax = fig.add_subplot(111, projection="3d", computed_zorder=False)
ax.view_init(elev=30, azim=-115, roll=0)
plot_lidar(ax, cloud_points, xs=traj, branch_idx=i)
plt.savefig(f'./figures/lidar/lidar_branch_{i + 1}.png', dpi=300)
plt.close() |