VAE代码（失败记录）

Flame5376

已于 2023-09-14 17:03:02 修改

阅读量154

点赞数

文章标签：学习

于 2023-09-14 15:27:04 首次发布

本文链接：https://blog.csdn.net/m0_46114999/article/details/132858400

版权

跑了一天，出来的值一直很怪

bbc_dwf	epoch:10,acc:21.9775, nmi:0.1616, ari:-0.0322
bbc_dwf	epoch:10,acc:21.9775, nmi:0.1616, ari:-0.0322
bbc_pw	epoch:10,acc:22.9213, nmi:0.2278, ari:0.0229
cite_pw	epoch:10,acc:19.8978, nmi:0.1718, ari:0.0040
acm_pw	epoch:10,acc:22.2810, nmi:0.1085, ari:0.0112
abstract_pw	epoch:10,acc:22.4570, nmi:0.1977, ari:0.1057

记录：

正常配置
500, 500, 1000, 1000, 500, 500
n_z = 10,
n_latent = 5
lr = 0.005
n_clusters=5, n_init=20
batch_size=256,shuffle=False
Epoch 117: Train Loss: 0.9996
acc:22.5169, nmi:0.3532, ari:0.1177
z_mean聚类
Epoch 12: Train Loss: 0.9996
acc:25.2135, nmi:8.4652, ari:0.0945
修改网络结构 500, 250, 100, 100, 250, 500, 5000,
lr = 0.005
Epoch 100: Train Loss: 0.9997
acc:47.1011, nmi:28.1540, ari:18.9423
lr=0.002
loss会上升？
Epoch 100: Train Loss: 0.9996
epoch:0,acc:23.9551, nmi:1.3544, ari:0.5867
epoch:0,acc:26.6517, nmi:2.2213, ari:0.9771
epoch:0,acc:51.7303, nmi:33.4441, ari:23.8787
epoch:0,acc:33.1236, nmi:8.7497, ari:6.0201
shuffle=True 50 epoch
epoch:0,acc:26.3820, nmi:1.8062, ari:1.0806
epoch:0,acc:24.9888, nmi:1.3512, ari:0.8620
epoch:0,acc:26.3371, nmi:2.8599, ari:1.1493

在这里插入图片描述

lr=0.01
epoch:0,acc:38.8315, nmi:35.1708, ari:22.3065
epoch:0,acc:25.8876, nmi:11.3914, ari:0.6150
epoch:0,acc:29.9775, nmi:3.9909, ari:3.7405
epoch:0,acc:45.4382, nmi:28.5032, ari:18.9452
epoch:0,acc:35.2809, nmi:17.6080, ari:10.3822
epoch:0,acc:32.9888, nmi:8.1380, ari:4.4337
epoch:0,acc:22.9663, nmi:0.0000, ari:0.0000
C:\Users\19920.conda\envs\pytorch\lib\site-packages\sklearn\base.py:1151: ConvergenceWarning: Number of distinct clusters (1) found smaller than n_clusters (5). Possibly due to duplicate points in X.
return fit_method(estimator, *args, **kwargs)
epoch:0,acc:22.0674, nmi:0.7222, ari:-0.0831
epoch:0,acc:23.3258, nmi:1.1470, ari:-0.0626
MSE->CE
epoch:0,acc:34.1124, nmi:13.1627, ari:8.5426
epoch:0,acc:28.3596, nmi:6.3709, ari:3.9546
500, 100, 250, 250, 100, 500, 5000 batch_size=256, shuffle=True
epoch:0,acc:25.9326, nmi:3.3499, ari:1.3552
500, 500, 1000, 500, 500, 1000
n_z = 10,
n_latent = 128
lr = 0.002
batch_size=256, shuffle=True
loss_CE
epoch:0,acc:23.9101, nmi:1.1215, ari:-0.0012
bbc from rln
epoch:0,acc:23.6404, nmi:0.7112, ari:0.2215
???

成功！

在这里插入图片描述
vae = VAE(128, 256,256, 128, 9635,
n_z = 10,
n_latent = 3
)
acc:59.0562, nmi:39.2102, ari:26.7944
acc:80.4944, nmi:62.3218, ari:58.1344
acc:81.0337, nmi:60.6005, ari:59.7550
acc:73.3933, nmi:51.2723, ari:44.0603
acc:83.3258, nmi:61.1844, ari:63.2995
acc:75.2360, nmi:56.2646, ari:53.6817
Epoch 300: Train Loss: 426.0915

vae = VAE(500, 1000, 1000, 500, 9635,
n_z = 10,
n_latent = 3
)
Epoch 300: Train Loss: 416.8523
acc:63.2809, nmi:45.4132, ari:39.6262
acc:76.0000, nmi:54.7599, ari:50.3674
acc:60.4045, nmi:44.7804, ari:38.7348