Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

如何让统计学专家发挥更多的实力呢? #6

Open
zhaipro opened this issue Jan 21, 2019 · 5 comments
Open

如何让统计学专家发挥更多的实力呢? #6

zhaipro opened this issue Jan 21, 2019 · 5 comments

Comments

@zhaipro
Copy link
Owner

zhaipro commented Jan 21, 2019

a

图片文件的命名规则:<类别>.<出现的次数>.(<在当前类别中出现的频率>).<索引>.jpg

索引只是用来防止文件重名的。

我大概估计得用于判断准确性的参数是:

  1. 出现次数必须大于15次,毕竟出现次数少,统计出来的值可信度也不够。
  2. 频率必须超过0.182,因为有某图片出现的次数足够多,但频率不够高,我猜测其原因可能是哈希算法出错了。
@zhaipro
Copy link
Owner Author

zhaipro commented Jan 21, 2019

其实,我想知道更有统计数理论的参数。但是,我的能力…

看来我需要把埋在床底的书找出来了。

@zhaipro
Copy link
Owner Author

zhaipro commented Jan 23, 2019

让统计学数据集做训练集,让此数据集做验证集

Epoch 100/100
8264/8264 [=] - ... - loss: 0.2421 - acc: 0.9211 - val_loss: 0.3360 - val_acc: 0.9462
[0.2166585269566813, 0.9507376632185858]

等等,只有在验证集上的正确率超过96.7%才算有进展,对吧

我开始觉得自己搞错了,卷积神经网络的用途在于面对前所未见的样本,这是无论如何统计学都做不到的。

@zhaipro
Copy link
Owner Author

zhaipro commented Jan 25, 2019

a3

~$ python3 main.py a.jpg
26
0 0 32
0 1 8
0 2 57
0 3 4
1 0 45
1 1 26
1 2 75
1 3 38

[19 78 73 10 26 38 26  8]

机器学习后的模型可以识别前所未见的图片,其中的蒸笼和创可贴都识别正确了。也许统计学可以在见过的图片中有98%的识别率,而学习后的框架则具有在无限空间中有94%的识别率。哦对了,那个文字的书写方式也是卷积神经网络在学习时没有见过的。

@zhaipro
Copy link
Owner Author

zhaipro commented Jan 26, 2019

那我就坐等新图出现吧

@zhaipro zhaipro closed this as completed Jan 28, 2019
@zhaipro
Copy link
Owner Author

zhaipro commented Feb 23, 2019

这里的数据集用于测试,得出的结果:
统计学专家的正确率:0.9422140966882884
从统计学专家哪里学来的深度学习模型的正确率:0.9811081335640064

可以以此证明学习后的神经网络具备识别前所未见的实力吗?

@zhaipro zhaipro reopened this Sep 8, 2019
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant