skip to main content
paste
bin
.ca
type · paste · share
⌘
K
Family
The bin family
pastebin.ca
hub
Share text and code with expiry and privacy controls.
imagebin.ca
Upload and share images with direct links.
filebin.ca
Drop a file and get a shareable link.
notebin.ca
Write Markdown notes with durable links.
turl.ca
Short, reputation-checked links.
attn.ca
Notifications and alerts for your services.
voicebin.ca
Record and share short voice clips.
dnsbin.ca
Inspect DNS and debug records.
Docs
Sign in
?
← back to paste
›
Edit / fork
Untitled paste
#6d8brVnfZb
public / public
new version
anonymous
created 2 days ago
Expires in 5 days
796 B
syntax:
text
Your changes create a new paste linked to this one — the original is untouched.
new version
Your changes create a new paste linked to this one — the original is untouched.
Title (optional)
Filename
Syntax
text
text
bash
c
cpp
css
diff
dockerfile
go
html
ini
java
javascript
json
kotlin
lua
makefile
markdown
nginx
php
python
ruby
rust
shellscript
sql
swift
toml
typescript
xml
yaml
Visibility
Public feed
Access
public
Expires
7 days
10 min
1 hour
1 day
7 days
30 days
90 days
custom…
Custom expiry
Change note
(optional)
This paste will be listed on the public feed. Change Visibility if you only want people with the link to see it.
Create new version
Cancel
Paste or type…
如何识别过拟合现象? 观察损失曲线:训练集损失持续下降,但是测试集(验证集)损失先下降,达到最低点之后开始上升;同时模型在训练集上准确率很高,在测试集上效果明显变差,就说明出现了过拟合。 为了避免过拟合,训练应该在哪个Epoch停止? 在测试(验证)损失达到最小值对应的Epoch停止训练,也就是测试损失即将由降转升的临界点。这就是早停法的核心依据。 除了早停法,还有哪些避免过拟合的方法? 正则化:L1、L2权重正则化 Dropout:训练时随机让一部分神经元失活 增加训练数据、数据增强 简化网络结构,减少网络参数 加入噪声、批量归一化(BatchNorm) 如果你想要更简短精简