情感分析：使用递归神经网络

中文版 pytorch
1

http://zh-v2.d2l.ai/chapter_natural-language-processing-applications/sentiment-analysis-rnn.html

3

image
image1419×155 5.82 KB

:grinning:

4 Likes
4

当我第一次运行程序后，得到的结果较差，模型无法分辨部分较为显然的积极与消极的语句，但是在运行第三次后，模型分辨能力显著提升，这是什么原因呢？谁能告诉我原因呢，感谢
When I first run the code ,I got a worse result. The model can’t recognize some obvious sentence,then I ren the code twice,it perforned better a lot then the first training,can some one tell me the reason,thank you.

5

你需要在这个cell里边，添加代码net.apply(init_weights);刷新一下net的权重，否则每运行一次都是和上次运行的结果去比较。

lr, num_epochs = 0.01, 5
net.apply(init_weights);
trainer = torch.optim.Adam(net.parameters(), lr=lr)
loss = nn.CrossEntropyLoss(reduction="none")
d2l.train_ch13(net, train_iter, test_iter, loss, trainer, num_epochs,
    devices)
6

示例程序在m1跑起来超级慢，已经加了device mps，gpu 100%了。几个小时才跑了一个epoch

7

image
image782×134 8.58 KB

:thinking:

9

m2 也跑不过来。改用 3060了，100s

10

sure thing … you just do more epochs simply on the same model

11

some more questions:
a. is it possible to apply BERT embedding instead of GloVe there?