As a student, I don't feel like I can be so confused anymore. I heard that the blockchain is very popular, I thought about it for a long time, so I decided to devote myself to machine learning on the road of no return (it means that the skin is very open). Xiaobai, I didn't understand anything, so I found a blind search on Baidu and saw an article on Zhihu. The introduction was not bad. There is also a link to jump to station B below. I skipped it without hesitation, but what came out was an open class of a weak middle-aged uncle, still in mosaic quality. After a search, it turned out to be a big cow in the machine learning world - Wu Enda . I looked for it on station B, and I really found a new video of the open class. The picture quality improvement is not a little bit. With great reverence, I started watching the open class ... (English, just time to practice listening, I haven't passed the sixth grade yet, hey~ It's just with Chinese subtitles)
Later, I went online again, and I plan to find some open source projects for machine learning to practice. After all, practice is very important. There are still quite a few open source projects on the Internet. I chose a MNIST dataset of handwritten numbers . After understanding it, I found that I need to use the TensoFlow library. What the hell is this! Baidu it. Initially, I wanted to configure it on my win10-32-bit machine, and then I found out that I was still too young. After two days of hard work, all kinds of information, all kinds of attempts, and finally failed. Throwing away my naive idea, I upgraded the machine to win10-64-bit system, and then installed a virtual machine with Ubuntu installed in it. In the Ubuntu system, I first installed an Anaconda3 , corresponding to python3.5. Link: https://repo.continuum.io/archive/ (there are various versions of this) How to get your own Baidu. After Anaconda is installed, test it and find that python can be used. Then start to configure tensorflow, there are many online tutorials, no longer wordy.
After all these are no problem, start learning MNIST! I found the mnist code of yhlleo on Github to learn to figure it out. (The Github community is quite useful) I downloaded the code of yhleo to the local (the four packages of MNIST should be placed in the local Mnist_data folder), and then opened the jupyter notebook (you can enter jupyter in Terminal after Anaconda is installed) The notebook is opened directly ) in the upper right corner of the mnist directory, create a new python notebook, and then use the %run command to test the program written by Daniel. But it took me two full days just to run the program. Due to version problems, many functions have to be updated, so errors are reported all the time. There is a problem, go online... In short, I read a lot of articles, and then various modifications tried to run the program. So, you can also refer to my revised Code . If it still doesn't work, it must not be my problem! There must be a mistake there, find the reason for yourself!
I have learned this level at present, and I don't understand a lot, but I still write some records to share. It's the first time to write something, it's not well written, it's just for reference.