TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Implementing a Convolutional Neural Network from Scratch in Python

238 点作者 vzhou842大约 6 年前

6 条评论

tsumnia大约 6 年前
Scrolling HN, I saw the two words I&#x27;m always weary of - &quot;from scratch&quot;. Mostly because I will click on the link, hoping to learn the mathematics behind a particular algorithm, only to see they&#x27;ve import sklearn and skip over all the explaining of how things are ACTUALLY getting done. Not that these types of tutorials do not have their place, but its irksome to see &quot;from scratch&quot; including most of the hard part being done.<p>With that, thank you Victor. Specifically because you did not do this at all and instead wrote a very easy to follow guide. I think this type of learning material will be very useful for CS and mathematics. The idea that very complicated algorithms must be explicitly implemented and then walked through, rather than symbols in a white paper will help make the mathematics of CS more applicable to everyone.<p>And to anyone bringing up numpy, it is at a level of &quot;prepackaged&quot; I&#x27;m fine with. I&#x27;m not going to raise my own pigs and chickens to make a breakfast burrito, but saying I did it from scratch by microwaving a frozen one isn&#x27;t going to cut it either. Numpy is like the basic ingredients to the recipe. While something like sklearn or tensorflow are perfectly acceptable, I wouldn&#x27;t say that&#x27;s the best method for learning CNNs.
评论 #19984490 未加载
评论 #19984825 未加载
评论 #19986858 未加载
评论 #19984332 未加载
vzhou842大约 6 年前
Hey, author here. Any&#x2F;all feedback is welcome, and I&#x27;m happy to answer questions.<p>Previous discussion on HN of the &quot;introduction to Neural Networks&quot; referenced in this article: <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=19320217" rel="nofollow">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=19320217</a><p>Runnable code from the article: <a href="https:&#x2F;&#x2F;repl.it&#x2F;@vzhou842&#x2F;A-CNN-from-scratch-Part-1" rel="nofollow">https:&#x2F;&#x2F;repl.it&#x2F;@vzhou842&#x2F;A-CNN-from-scratch-Part-1</a><p>Github: <a href="https:&#x2F;&#x2F;github.com&#x2F;vzhou842&#x2F;cnn-from-scratch" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;vzhou842&#x2F;cnn-from-scratch</a>
评论 #19983110 未加载
评论 #19984222 未加载
评论 #19986775 未加载
评论 #20024490 未加载
duked大约 6 年前
This is a great tutorial. However, every time I see RNN&#x2F;CNN it&#x27;s always applied to some video stream or set of images. I really would like to find some tutorial but applied to event logs or other text-based input. Anyone has a good link for that?
评论 #19986322 未加载
amelius大约 6 年前
&gt; If you trained a network to detect dogs, you’d want it to be able to a detect a dog regardless of where it appears in the image. Imagine training a network that works well on a certain dog image, but then feeding it a slightly shifted version of the same image. The dog would not activate the same neurons, so the network would react completely differently!<p>But CNNs only deal with translations. What if the image of the dog is rotated?
评论 #19984340 未加载
评论 #19984691 未加载
master_yoda_1大约 6 年前
You copy code and figures from karpathy lecture notes and did not cite him. Its call plagiarism.
windsignaling大约 6 年前
Where&#x27;s part 2? That&#x27;s going to be the thing that&#x27;s actually non-trivial
评论 #19984943 未加载