Even if you say machine learning, what should you put in and what to do, so keep a log of trial and error I will also post a reference link When you review it later, it will be a memo that you can remember how much you searched (desire)
20 selections of content and libraries that beginners of machine learning can learn quickly (Python edition) % E6% A9% 9F% E6% A2% B0% E5% AD% A6% E7% BF% 92% E5% 88% 9D% E5% BF% 83% E8% 80% 85% E3% 81% 8C% E3 % 80% 81% E7% B4% A0% E6% 97% A9% E3% 81% 8F% E5% AD% A6% E3% 81% B9% E3% 82% 8B% E3% 82% B3% E3% 83 % B3% E3% 83% 86% E3% 83% B3% E3% 83% 84% E3% 81% A8% E3% 83% A9) It's Python Also, I remember hearing the name "scikit-learn" in a library for machine learning.
It's hard to build an environment from scratch (it's fun if you have the time), so I want to use a package that feels good. Do you like this area? Is it?
About 3 years ago, I thought I should study Python, so I dropped it. Portable Python seems to have stopped developing. Python (x, y) has stopped updating on 06/30/2015, and 2.x series is excluded by saying ... With anaconda, I thought it would be okay to change it after moving to Linux, but I chose WinPython because it is a portable ski that does not mess with the registry. It seems that NumPy and SciPy, which I had a hard time installing on Linux, as well as the scikit-learn I was looking for are packaged, isn't it?
There was a great comparison page, so I used it as a reference. "Comparison of Python packages", minus9d's diary
It says that there is a VC ++ 2015 redistributable package, so drop it in. Is WinPython maintained for 2 generations? The latest of now (2016/7/12) seems to be the following two generations
I decided to use the larger version without thinking deeply I think there was a language like the development version when the minor version was odd, but what kind of rule does Python use?
For the time being, I made a py directory directly under the C drive and put it in, whether it was installation or just extraction. If you want to change the installation directory, you can cut it and move it, so portable type is easy.
You can proceed with the tutorial on the scikit-learn page, but find a page where you can proceed with the tutorial in Japanese for yourself, saying that English is ... and step by step. "Introduction to Machine Learning with scikit-learn", Takezo dying blog
There are a lot of exes in the directory where WinPython is put, and I'm wondering which one to start, but Jupyter Notebook seems to be more convenient than the interactive shell, so I will move it and type in the code like that
At the end, it's nice that the judgment result 8 is returned, but I got a warning.
C:\py\python-3.5.1.amd64\lib\site-packages\sklearn\utils\validation.py:386: DeprecationWarning: Passing 1d arrays as data is deprecated in 0.17 and willraise ValueError in 0.19. Reshape your data either using X.reshape(-1, 1) if your data has a single feature or X.reshape(1, -1) if it contains a single sample.
DeprecationWarning)
Is passing data as a one-dimensional array abolished from scikit-learn version 0.17? I think I wrote that it will be an error when it becomes 0.19, but even if you look at the tutorial page of the scikit-learn head family, the same code remains, so it is a mystery
If you change to the code below, the warning will stop, but what is the principle ... I wonder if that side is related because it has changed to a two-dimensional array.
clf.predict(digits.data[-1].reshape(1,-1))
I was able to confirm that shortening the array passed to fit reduces the accuracy, and it seems that I learned for the time being? So up to this area The continuation is coming again