File size: 17,419 Bytes
f291f4a |
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 |
"""The Projector class for visualization, serve as a helper module for evaluator and visualizer"""
from abc import ABC, abstractmethod
import os
import json
import numpy as np
import torch
class ProjectorAbstractClass(ABC):
@abstractmethod
def __init__(self, vis_model, content_path, *args, **kwargs):
pass
@abstractmethod
def load(self, *args, **kwargs):
pass
@abstractmethod
def batch_project(self, *args, **kwargs):
pass
@abstractmethod
def individual_project(self, *args, **kwargs):
pass
@abstractmethod
def batch_inverse(self, *args, **kwargs):
pass
@abstractmethod
def individual_inverse(self, *args, **kwargs):
pass
class Projector(ProjectorAbstractClass):
def __init__(self, vis_model, content_path, vis_model_name, device):
self.vis_model = vis_model
self.content_path = content_path
self.vis_model_name = vis_model_name
self.DEVICE = device
def load(self, iteration):
raise NotImplementedError
def batch_project(self, iteration, data):
self.load(iteration)
embedding = self.vis_model.encoder(torch.from_numpy(data).to(dtype=torch.float32, device=self.DEVICE)).cpu().detach().numpy()
return embedding
def individual_project(self, iteration, data):
self.load(iteration)
embedding = self.vis_model.encoder(torch.from_numpy(np.expand_dims(data, axis=0)).to(dtype=torch.float32, device=self.DEVICE)).cpu().detach().numpy()
return embedding.squeeze(axis=0)
def batch_inverse(self, iteration, embedding):
self.load(iteration)
data = self.vis_model.decoder(torch.from_numpy(embedding).to(dtype=torch.float32, device=self.DEVICE)).cpu().detach().numpy()
return data
def individual_inverse(self, iteration, embedding):
self.load(iteration)
data = self.vis_model.decoder(torch.from_numpy(np.expand_dims(embedding, axis=0)).to(dtype=torch.float32, device="cpu")).cpu().detach().numpy()
return data.squeeze(axis=0)
class DeepDebuggerProjector(Projector):
def __init__(self, vis_model, content_path, vis_model_name, segments, device):
super().__init__(vis_model, content_path, vis_model_name, device)
self.segments = segments
self.segments = segments #[(1,6),(6, 15),(15,42),(42,200)]
self.current_range = (-1,-1)
def load(self, iteration):
# [s,e)
init_e = self.segments[-1][1]
if (iteration >= self.current_range[0] and iteration <self.current_range[1]) or (iteration == init_e and self.current_range[1] == init_e):
print("Same range as current visualization model...")
return
# else
for i in range(len(self.segments)):
s = self.segments[i][0]
e = self.segments[i][1]
# range [s,e)
if (iteration >= s and iteration < e) or (iteration == init_e and e == init_e):
idx = i
break
# TODO vis model name as a hyperparameter
file_path = os.path.join(self.content_path, "Model", "{}_{}.pth".format(self.vis_model_name, idx))
save_model = torch.load(file_path, map_location="cpu")
self.vis_model.load_state_dict(save_model["state_dict"])
self.vis_model.to(self.DEVICE)
self.vis_model.eval()
self.current_range = (s, e)
print("Successfully load the visualization model for range ({},{})...".format(s,e))
class ALProjector(Projector):
def __init__(self, vis_model, content_path, vis_model_name, device) -> None:
super().__init__(vis_model, content_path,vis_model_name, device)
self.current_range = None
def load(self, iteration):
file_path=os.path.join(self.content_path, "Model", "Iteration_{}".format(iteration), self.vis_model_name+".pth")
save_model = torch.load(file_path, map_location=torch.device("cpu"))
self.vis_model.load_state_dict(save_model["state_dict"])
self.vis_model.to(self.DEVICE)
self.vis_model.eval()
print("Successfully load the visualization model for Iteration {}...".format(iteration))
class DenseALProjector(DeepDebuggerProjector):
def __init__(self, vis_model, content_path, vis_model_name, device) -> None:
super().__init__(vis_model, content_path, vis_model_name, None, device)
self.current_range = [-1,-1,-1] # iteration, e_s, e_e
def load(self, iteration, epoch):
# [s,e)
curr_iteration, curr_s, curr_e = self.current_range
segment_path = os.path.join(self.content_path, "Model", "Iteration_{}".format(iteration), "segments.json")
with open(segment_path, "r") as f:
segments = json.load(f)
init_e = segments[-1][1]
# [s,e)
if iteration == curr_iteration:
if (curr_e==init_e and epoch == curr_e) or (epoch >= curr_s and epoch < curr_e):
print("Same range as current visualization model...")
return
for i in range(len(segments)):
s = segments[i][0]
e = segments[i][1]
# range [s, e)
if (epoch >= s and epoch < e) or (e == init_e and epoch == e):
idx = i
break
file_path = os.path.join(self.content_path, "Model", "Iteration_{}".format(iteration), "{}_{}.pth".format(self.vis_model_name, idx))
save_model = torch.load(file_path, map_location=self.DEVICE)
self.vis_model.load_state_dict(save_model["state_dict"])
self.vis_model.to(self.DEVICE)
self.vis_model.eval()
self.current_range = (iteration, s, e)
print("Successfully load the visualization model in iteration {} for range ({},{}]...".format(iteration, s,e))
def batch_project(self, iteration, epoch, data):
self.load(iteration, epoch)
embedding = self.vis_model.encoder(torch.from_numpy(data).to(dtype=torch.float32, device=self.DEVICE)).cpu().detach().numpy()
return embedding
def individual_project(self, iteration, epoch, data):
self.load(iteration, epoch)
embedding = self.vis_model.encoder(torch.from_numpy(np.expand_dims(data, axis=0)).to(dtype=torch.float32, device=self.DEVICE)).cpu().detach().numpy()
return embedding.squeeze(axis=0)
def batch_inverse(self, iteration, epoch, embedding):
self.load(iteration, epoch)
data = self.vis_model.decoder(torch.from_numpy(embedding).to(dtype=torch.float32, device=self.DEVICE)).cpu().detach().numpy()
return data
def individual_inverse(self, iteration, epoch, embedding):
self.load(iteration, epoch)
data = self.vis_model.decoder(torch.from_numpy(np.expand_dims(embedding, axis=0)).to(dtype=torch.float32, device=self.DEVICE)).cpu().detach().numpy()
return data.squeeze(axis=0)
class EvalProjector(DeepDebuggerProjector):
def __init__(self, vis_model, content_path, vis_model_name, device, exp) -> None:
super().__init__(vis_model, content_path, vis_model_name, None, device)
self.exp = exp
file_path = os.path.join(content_path, "Model", "{}".format(exp), "segments.json")
with open(file_path, "r") as f:
self.segments = json.load(f)
def load(self, iteration):
# (s, e]
# (s,e]
init_s = self.segments[0][0]
if (iteration > self.current_range[0] and iteration <=self.current_range[1]) or (iteration == init_s and self.current_range[0] == init_s):
print("Same range as current visualization model...")
return
# else
for i in range(len(self.segments)):
s = self.segments[i][0]
e = self.segments[i][1]
# range (s,e]
if (iteration > s and iteration <= e) or (iteration == init_s and s == init_s):
idx = i
break
file_path = os.path.join(self.content_path, "Model", "{}".format(self.exp), "tnn_hybrid_{}.pth".format(idx))
save_model = torch.load(file_path, map_location="cpu")
self.vis_model.load_state_dict(save_model["state_dict"])
self.vis_model.to(self.DEVICE)
self.vis_model.eval()
self.current_range = (s, e)
print("Successfully load the visualization model for range ({},{})...".format(s,e))
class DVIProjector(Projector):
def __init__(self, vis_model, content_path, vis_model_name, device) -> None:
super().__init__(vis_model, content_path, vis_model_name, device)
def load(self, iteration):
file_path = os.path.join(self.content_path, "Model", "Epoch_{}".format(iteration), "{}.pth".format(self.vis_model_name))
save_model = torch.load(file_path, map_location="cpu")
self.vis_model.load_state_dict(save_model["state_dict"])
self.vis_model.to(self.DEVICE)
self.vis_model.eval()
print("Successfully load the DVI visualization model for iteration {}".format(iteration))
class TimeVisProjector(Projector):
def __init__(self, vis_model, content_path, vis_model_name, device, verbose=0) -> None:
super().__init__(vis_model, content_path, vis_model_name, device)
self.verbose = verbose
def load(self, iteration):
file_path = os.path.join(self.content_path, "Model", "{}.pth".format(self.vis_model_name))
save_model = torch.load(file_path, map_location="cpu")
self.vis_model.load_state_dict(save_model["state_dict"])
self.vis_model.to(self.DEVICE)
self.vis_model.eval()
if self.verbose>0:
print("Successfully load the TimeVis visualization model for iteration {}".format(iteration))
class TimeVisDenseALProjector(Projector):
def __init__(self, vis_model, content_path, vis_model_name, device, verbose=0) -> None:
super().__init__(vis_model, content_path, vis_model_name, device)
self.verbose = verbose
self.curr_iteration = -1
def load(self, iteration, epoch):
if iteration == self.curr_iteration:
return
file_path = os.path.join(self.content_path, "Model", f'Iteration_{iteration}', "{}.pth".format(self.vis_model_name))
save_model = torch.load(file_path, map_location="cpu")
self.vis_model.load_state_dict(save_model["state_dict"])
self.vis_model.to(self.DEVICE)
self.vis_model.eval()
if self.verbose>0:
print("Successfully load the TimeVis visualization model for iteration {}".format(iteration))
self.curr_iteration = iteration
def batch_project(self, iteration, epoch, data):
self.load(iteration, epoch)
embedding = self.vis_model.encoder(torch.from_numpy(data).to(dtype=torch.float32, device=self.DEVICE)).cpu().detach().numpy()
return embedding
def individual_project(self, iteration, epoch, data):
self.load(iteration, epoch)
embedding = self.vis_model.encoder(torch.from_numpy(np.expand_dims(data, axis=0)).to(dtype=torch.float32, device=self.DEVICE)).cpu().detach().numpy()
return embedding.squeeze(axis=0)
def batch_inverse(self, iteration, epoch, embedding):
self.load(iteration, epoch)
data = self.vis_model.decoder(torch.from_numpy(embedding).to(dtype=torch.float32, device=self.DEVICE)).cpu().detach().numpy()
return data
def individual_inverse(self, iteration, epoch, embedding):
self.load(iteration, epoch)
data = self.vis_model.decoder(torch.from_numpy(np.expand_dims(embedding, axis=0)).to(dtype=torch.float32, device=self.DEVICE)).cpu().detach().numpy()
return data.squeeze(axis=0)
import tensorflow as tf
class tfDVIProjector(ProjectorAbstractClass):
def __init__(self, content_path, flag, verbose=0):
self.content_path = content_path
self.model_path = os.path.join(self.content_path, "Model")
self.flag = flag
self.curr_iteration = -1
self.encoder = None
self.decoder = None
self.verbose = verbose
def load(self, epoch):
if self.curr_iteration == epoch:
print("Current autocoder model loaded from Epoch {}".format(epoch))
return
encoder_location = os.path.join(self.model_path, "Epoch_{:d}".format(epoch),"encoder" + self.flag)
decoder_location = os.path.join(self.model_path, "Epoch_{:d}".format(epoch),"decoder" + self.flag)
try:
self.encoder = tf.keras.models.load_model(encoder_location)
self.decoder = tf.keras.models.load_model(decoder_location)
if self.verbose>0:
print("Keras autocoder model loaded from Epoch {}".format(epoch))
self.curr_iteration = epoch
except FileNotFoundError:
print("Error! Projection function has not been initialized! Pls first visualize all.")
def batch_project(self, epoch, data):
'''
batch project data to 2D space
:param data: numpy.ndarray
:param epoch: int
:return: embedding numpy.ndarray
'''
self.load(epoch)
embedding = self.encoder(data).cpu().numpy()
return embedding
def individual_project(self, epoch, data):
'''
project a data to 2D space
:param data: numpy.ndarray
:param epoch: int
:return: embedding numpy.ndarray
'''
self.load(epoch)
data = np.expand_dims(data, axis=0)
embedding = self.encoder(data).cpu().numpy()
return embedding.squeeze(0)
def batch_inverse(self, epoch, data):
"""
map 2D points back into high dimensional space
:param data: ndarray, (n, 2)
:param epoch: num of epoch
:return: high dim representation, numpy.ndarray
"""
self.load(epoch)
representation_data = self.decoder(data).cpu().numpy()
return representation_data
def individual_inverse(self, epoch, data):
"""
map a 2D point back into high dimensional space
:param data: ndarray, (1, 2)
:param epoch: num of epoch
:return: high dim representation, numpy.ndarray
"""
self.load(epoch)
data = np.expand_dims(data, axis=0)
representation_data = self.decoder(data).cpu().numpy()
return representation_data.squeeze(0)
class tfDVIDenseALProjector(ProjectorAbstractClass):
def __init__(self, content_path, flag, verbose=0):
self.content_path = content_path
self.model_path = os.path.join(self.content_path, "Model")
self.flag = flag
self.curr_iteration = -1
self.curr_epoch = -1
self.encoder = None
self.decoder = None
self.verbose = verbose
def load(self, iteration, epoch):
if self.curr_iteration == iteration and self.curr_epoch == epoch:
print("Current autocoder model loaded from Iteration {}/Epoch {}".format(iteration, epoch))
return
encoder_location = os.path.join(self.model_path, "Iteration_{:d}".format(iteration), "Epoch_{:d}".format(epoch),"encoder" + self.flag)
decoder_location = os.path.join(self.model_path, "Iteration_{:d}".format(iteration), "Epoch_{:d}".format(epoch),"decoder" + self.flag)
try:
self.encoder = tf.keras.models.load_model(encoder_location)
self.decoder = tf.keras.models.load_model(decoder_location)
if self.verbose>0:
print("Keras autocoder model loaded from Epoch {}".format(epoch))
self.curr_iteration = iteration
self.curr_epoch = epoch
except FileNotFoundError:
print("Error! Projection function has not been initialized! Pls first visualize all.")
def batch_project(self, iteration, epoch, data):
'''
batch project data to 2D space
:param data: numpy.ndarray
:param epoch: int
:return: embedding numpy.ndarray
'''
self.load(iteration, epoch)
embedding = self.encoder(data).cpu().numpy()
return embedding
def individual_project(self, iteration, epoch, data):
'''
project a data to 2D space
:param data: numpy.ndarray
:param epoch: int
:return: embedding numpy.ndarray
'''
self.load(iteration, epoch)
data = np.expand_dims(data, axis=0)
embedding = self.encoder(data).cpu().numpy()
return embedding.squeeze(0)
def batch_inverse(self, iteration, epoch, data):
"""
map 2D points back into high dimensional space
:param data: ndarray, (n, 2)
:param epoch: num of epoch
:return: high dim representation, numpy.ndarray
"""
self.load(iteration, epoch)
representation_data = self.decoder(data).cpu().numpy()
return representation_data
def individual_inverse(self, iteration, epoch, data):
"""
map a 2D point back into high dimensional space
:param data: ndarray, (1, 2)
:param epoch: num of epoch
:return: high dim representation, numpy.ndarray
"""
self.load(iteration, epoch)
data = np.expand_dims(data, axis=0)
representation_data = self.decoder(data).cpu().numpy()
return representation_data.squeeze(0) |