-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathbak_code.py
executable file
·315 lines (315 loc) · 17.1 KB
/
bak_code.py
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
# def train_and_evaluate(self, model, train_dataloader, val_dataloader,
# optimizer, loss_fn, metrics, params, model_dir,
# restore_file=None):
# # reload weights from restore_file if specified
# if restore_file is not None: # TODO 有bug要修
# restore_path = os.path.join(model_dir, restore_file + '.pth.tar')
# logging.info("Restoring parameters from {}".format(restore_path))
# utils.load_checkpoint(restore_path, model, optimizer)
#
# best_val_acc = 0.0
#
# cnt_step = 0
# for epoch in range(params.epoch):
#
# # Run one epoch
# logging.info("Epoch {}/{}".format(epoch + 1, params.epoch))
#
# # 主要是看下这里model是否共享，即train好的model是否给了test，
# # 下一个epoch的model是否来自上一个epoch train的 model
# logging.info("model id before train: ", id(model))
#
# # compute number of batches in one epoch (one full pass over the training set)
# self.train_joint_AE_deepSVDD(model, optimizer, loss_fn, train_dataloader, metrics, params)
#
# #
# logging.info("model id before test: ", id(model))
#
# # Evaluate for one epoch on validation set
# val_metrics = self.test_joint_AE_deepSVDD(model, loss_fn, val_dataloader, metrics, params)
#
# val_acc = val_metrics['auc']
# is_best = val_acc >= best_val_acc
#
# # Save weights
# utils.save_checkpoint({'epoch': epoch + 1,
# 'state_dict': model.state_dict(),
# 'optim_dict': optimizer.state_dict(),
# 'c': self.c,
# 'R': self.R,
# 'nu': self.nu,
# 'lam_rec': self.lam_rec,
# 'lam_svdd': self.lam_svdd
# },
# is_best=is_best,
# checkpoint=model_dir)
#
# # If best_eval, best_save_path
# if is_best:
# logging.info("- Found new best accuracy")
# best_val_acc = val_acc
#
# # Save best val metrics in a json file in the model directory
# best_json_path = os.path.join(model_dir, "metrics_val_best_weights.json")
# utils.save_dict_to_json(val_metrics, best_json_path)
#
# # Save latest val metrics in a json file in the model directory
# last_json_path = os.path.join(model_dir, "metrics_val_last_weights.json")
# utils.save_dict_to_json(val_metrics, last_json_path)
#
# def train_joint_AE_deepSVDD(self,
# model, optimizer, loss_fn, train_dataloader, metrics, params):
# c, t, h, w = train_dataloader.raw_shape
# #
# model.train() # set model to training mode
# optimizer.zero_grad() # clear previous gradients
# #
# cnt_step = 0
# # summary for current training loop and a running average object for loss
# summ = []
# # loss_avg = utils.RunningAverage()
# with SummaryWriter(log_dir="summary/train_deepSVDD/train_{0}_lr={1}_lam_rec={2}_"
# "lam_svdd={3}_nu={4}_{5}_{6}".format(
# self.dataset_name, self.LR, self.lam_rec, self.lam_svdd, self.nu,
# params.normal_or_dist, time.strftime('%m%d_%H%M')),
# comment="{}".format(self.dataset_name)) as writer:
# for cl_idx, video_id in enumerate(train_dataloader.train_videos):
# # 逐个子目录处理 from Train001~Train016
# # Run the train
# train_dataloader.train(video_id) # 加载当前子目录的所有帧组成一个大clip到内存
# #
# loader = DataLoader(train_dataloader,
# collate_fn=train_dataloader.collate_fn,
# num_workers=self.num_workers,
# batch_size=self.batch_size,
# shuffle=True)
#
# for i, (x, y) in tqdm(enumerate(loader),
# desc=f'Training for {self.dataset_name}'):
# #
# cnt_step = cnt_step + 1 # 一个step 一个 batch: 1380张
# # print("x, y .shape: ", x.shape, y.shape) # 1380, 1, 8, 32, 32
# x = x.to(self.device)
# x_r, z = model(x)
# #
# z = z.view(-1, 690,128)
# # print("z.size: ", z.size())
# #
# dist = torch.sum((z - self.c) ** 2, dim=1) # points to center
# #
# total_loss_bp = loss_fn(x, x_r, z) # 返回的是一个tensor
# # print("z, z_dist: ", z, z_dist)0
# reconstruction_loss = loss_fn.reconstruction_loss
# # print("reconstruction_loss: ", reconstruction_loss)
# deepSVDD_loss = loss_fn.deepSVDD_loss
# # print("deepSVDD_loss: ", deepSVDD_loss)
# total_loss = loss_fn.total_loss
# print("\ntotal_loss: ", total_loss)
# if cnt_step % params.plot_every == 0:
# writer.add_scalars("train_loss",
# {'total_loss': total_loss,
# 'reconstruction_loss': reconstruction_loss,
# 'deepSVDD_loss': deepSVDD_loss
# },
# cnt_step)
# # if cnt_step % params.save_ckpt_every == 0:
# # # 保存模型 （在每个epoch结束时保存）# 或者根据 cnt_step设置
# # ckpt_path = '{prefix}{dataset}_{time}.pkl'.format(
# # prefix=Config.prefix,
# # dataset=Config.dataset_name,
# # time=time.strftime('%m%d_%H%M') # 这个要和下面 save()无限近
# # )
# # net_dict = self.model.state_dict()
# # #
# # torch.save({'R': self.R,
# # 'c': self.c,
# # 'net_dict': net_dict, }, ckpt_path)
# # print("epoch {} complete !".format(epoch))
# total_loss_bp.backward() # 确保 optimizer.zero_grad()
# # 梯度裁剪
# torch.nn.utils.clip_grad_norm_(model.parameters(),
# max_norm=20)
# optimizer.step()
# # Update hypersphere radius R on mini-batch distances
# # if (self.objective == 'soft-boundary') and (epoch >= params.warm_up_n_steps):
# # self.R.data = torch.tensor(get_radius(dist, self.nu), device=self.device)
# #
# # 由于没有 label，无法计算acc，所以直接打印loss
# # Evaluate summaries only once in a while
# if i % params.save_summary_steps == 0:
# summary_batch = {}
# summary_batch['loss'] = total_loss
# summ.append(summary_batch)
# # compute mean of all metrics in summary
# metrics_mean = {metric: np.mean([x[metric] for x in summ]) for metric in summ[0]}
# metrics_string = " ; ".join("{}: {:05.3f}".format(k, v) for k, v in metrics_mean.items())
# logging.info("- Train metrics: " + metrics_string)
#
# @torch.no_grad() # 在这里以装饰器的方法，静止back propogation，方便代码重用
# def test_joint_AE_deepSVDD(self, model, loss_fn, val_dataloader, metrics, params):
# # type: () -> None
# """
# Actually performs tests.
# """
# c, t, h, w = val_dataloader.raw_shape
#
# # set model to evaluation mode
# model.eval()
#
# # summary for current eval loop
# # summ = []
# metrics = {}
#
# # Load the checkpoint
# # self.model.load_w(self.checkpoint)
# # self.ckpt = torch.load(self.checkpoint)
# # self.model.load_state_dict(self.ckpt['net_dict'])
# # self.R = self.ckpt['R']
# # self.c = self.ckpt['c']
#
# # Prepare a table to show results
# vad_table = self.empty_table
#
# # Set up container for novelty scores from all test videos
# global_llk = []
# global_rec = []
# global_ns = []
# global_y = []
#
# # Get accumulators，干嘛的？答：get frame-level scores from clip-level scores
# results_accumulator_llk = ResultsAccumulator(time_steps=t)
# results_accumulator_rec = ResultsAccumulator(time_steps=t)
#
# cnt_step = 0 # global_step
# with SummaryWriter(log_dir="summary/test_{0}".format(
# params.output_file.split('.')[0]),
# comment="{}".format(params.dataset_name)) as writer:
# # Start iteration over test videos
# for cl_idx, video_id in enumerate(val_dataloader.test_videos):
# # test_videos 的内容是：TestXXX(XXX：001~012) 这些目录名，每个目录名保存有一个
# # 视频的所有帧，所以代表一个视频，即 video_id
#
# # Run the test
# val_dataloader.test(video_id) # 设置好cur_video_frames【其实是整个视频的全部clips】，
# # cur_video_gt，cur_len【其实是clips number】
# loader = DataLoader(val_dataloader,
# num_workers=1,
# shuffle=False,
# batch_size=1,
# collate_fn=val_dataloader.collate_fn) # 临时构建loader
# # 因为是 inference，所以没有 batch_size (或者说==1)
# # collate_fn：实际作用是：TODO
#
# # Build score containers
# sample_llk = np.zeros(shape=(len(loader) + t - 1,))
# sample_rec = np.zeros(shape=(len(loader) + t - 1,))
# # print("len(loader): ", len(loader)) # len(self.batch_sampler)
# # 因为loader会把所有Dataset的所有item都做登记，而len(dataset) ==
# # num_frames - t + 1，即所有的clips (带overlap的)，要恢复就是：
# # len(loader) + t - 1
# # print("len(loader) + t - 1: ", len(loader) + t - 1)
# sample_y = val_dataloader.load_test_sequence_gt(video_id) # (n_frames,)
# # print("len(sample_y): ", len(sample_y))
# # 事实证明：(len(loader) + t - 1) == len(sample_y), len(loader) =
# #
# for i, (x, y) in tqdm(enumerate(loader),
# desc=f'Computing scores for {self.dataset_name}'):
# #
# cnt_step = cnt_step + 1
#
# x = x.to(self.device)
#
# x_r, z = model(x)
# z = z.view(-1, 690, 128)
# # print("in 327 line, z.size: ", z.size())
#
# ttloss = loss_fn(x, x_r, z) # 记住，self.loss其实一个 object，这里
# # 被执行了 forwrd()，所以等于修改了 object (即 self.loss被修改了)
# total_loss = loss_fn.total_loss
# reconstruction_loss = loss_fn.reconstruction_loss
# deepSVDD_loss = loss_fn.deepSVDD_loss
# # write all loss
# # if cnt_step % Config.plot_every == 0:
# # writer.add_scalars("test_loss",
# # {'total_loss': total_loss,
# # 'reconstruction_loss': reconstruction_loss,
# # 'autoregression_loss': autoregression_loss
# # },
# # cnt_step)
#
# # Feed results accumulators: 模仿一个队列，队尾进，队头出
# # 我的办法：通过设置断点，进去看results_accumulator_llk是怎么工作的？
# # 因为 batch_szie == 1, 所以push了 it(==num_clips==num_frames-t+1)次，
# # 所以还有 (t - 1) 帧没有计算 loss，留到 下面的 while
# results_accumulator_llk.push(loss_fn.deepSVDD_loss)
# results_accumulator_rec.push(loss_fn.reconstruction_loss)
# sample_llk[i] = results_accumulator_llk.get_next()
# sample_rec[i] = results_accumulator_rec.get_next()
#
# # Get last results
# # 计算剩下的 (t-1)帧各自的 loss
# while results_accumulator_llk.results_left != 0:
# index = (- results_accumulator_llk.results_left)
# sample_llk[index] = results_accumulator_llk.get_next()
# sample_rec[index] = results_accumulator_rec.get_next()
#
# min_llk, max_llk, min_rec, max_rec = self.compute_normalizing_coefficients(
# sample_llk, sample_rec)
#
# # Compute the normalized scores and novelty score
# sample_llk = normalize(sample_llk, min_llk, max_llk)
# sample_rec = normalize(sample_rec, min_rec, max_rec)
# sample_ns = novelty_score(sample_llk, sample_rec)
# # # 绘制 score-map
# # # print("len of sample_ns:", len(sample_ns))
# # fig_novelty_score = plt.figure()
# # plt.title('novelty_score of {}'.format(video_id))
# # plt.plot(range(len(sample_ns)), sample_ns, color='green',
# # label='novelty_score')
# # plt.xlabel('frames')
# # plt.ylabel('novelty_score')
# # writer.add_figure('Novelty Score', fig_novelty_score, global_step=cl_idx)
#
# # Update global scores (used for global metrics)
# global_llk.append(sample_llk)
# global_rec.append(sample_rec)
# global_ns.append(sample_ns)
# global_y.append(sample_y)
#
# try:
# # Compute AUROC for this video
# this_video_metrics = [
# roc_auc_score(sample_y, sample_llk), # likelihood metric
# roc_auc_score(sample_y, sample_rec), # reconstruction metric
# roc_auc_score(sample_y, sample_ns) # novelty score
# ]
# vad_table.add_row([video_id] + this_video_metrics)
# except ValueError:
# # This happens for sequences in which all frames are abnormal
# # Skipping this row in the table (the sequence will still count for global metrics)
# continue
#
# # Compute global AUROC and print table
# global_llk = np.concatenate(global_llk)
# global_rec = np.concatenate(global_rec)
# global_ns = np.concatenate(global_ns)
# global_y = np.concatenate(global_y)
# global_metrics = [
# roc_auc_score(global_y, global_llk), # likelihood metric
# roc_auc_score(global_y, global_rec), # reconstruction metric
# roc_auc_score(global_y, global_ns) # novelty score
# ]
# vad_table.add_row(['avg'] + list(global_metrics))
# print(vad_table)
#
# # # Save table
# # with open(self.output_file, mode='w') as f:
# # f.write(str(vad_table))
# # #
# # # 查看下网络
# # # model_input = torch.rand([1380, 1, 8, 32, 32])
# # # writer.add_graph(self.model, input_to_model=model_input)
# # print("ag_auc: ", list(global_metrics)[2])
# logging.info("avg_auc: ", list(global_metrics)[2])
# metrics['auc'] = list(global_metrics)[2]
# return metrics# 返回 avg_auc