Use Scikit-learn to do Pattern Recognition

Use Scikit-learn to do Pattern Recognition on the MNist Database of handwritten digits


To install Scikit-learn on Windows , You should intall numpy, scipy and matplotlib, first, it seems a little bit tedious. One convenient way to install those 3 commonly used packages, I suggest, is just  go to this site. http://www.lfd.uci.edu/~gohlke/pythonlibs/

Find 2 exe files

• Scipy-stack: Scipy‑stack‑14.8.27.win32‑py3.4.exe
• Scikit-learn: scikit‑learn‑0.15.2.win32‑py3.4.exe

After installation, just check it in Python

>>> import sklearn

Then goto the homepage of Scikit-learn, to get some idea of it.

Here are some pre-jobs I have done after 2 days study. Just for your reference. Enjoy it.

By the way, if you want to construct a real-time system, you need use the camera to capture a real image. You can use pygame module, which can be found here.

• pygame:  pygame‑1.9.2a0.win32‑py3.4.exe

---

---

以上程式 rySklearn001.py 會呼叫 fetch_mldata 函式，
The above python program will call fetch_mldata function,

這個函式 存於 C:\Python34\Lib\site-packages\sklearn\datasets\mldata.py

它會自動來以下網站

http://mldata.org/repository/data/download/matlab/mnist-original/

擷取這個檔案，
mnist-original.mat

放在本程式所在的檔案夾之中。

執行以下 python 指令(本程式所提供)，

就可看到隨機出現的某張影像。

>>> randomShow()
0 894 0.0
:

---

以下開始從基礎開始學一下 sklearn 模組。

An introduction to machine learning with scikit-learn

Recognizing hand-written digits

I have found some programs good for beginner.
You should be able to run them and study them first.

(1)
plot_iris_dataset.py
http://scikit-learn.org/stable/auto_examples/datasets/plot_iris_dataset.html

(2)
plot_digits_classification.py
http://scikit-learn.org/stable/auto_examples/plot_digits_classification.html#example-plot-digits-classification-py

(3)
plot_digits_last_image.py
http://scikit-learn.org/stable/auto_examples/datasets/plot_digits_last_image.html#example-datasets-plot-digits-last-image-py

(4)
plot_random_dataset.py
http://scikit-learn.org/stable/auto_examples/datasets/plot_random_dataset.html#example-datasets-plot-random-dataset-py

---

以上程式 rySklearn003_MNistSVM.py (http://goo.gl/nGZuFk)

針對 MNist，完成初步的 Pattern Recognition 任務，

先用  Sklearn 裡面的 Perceptron 來做。

再加碼用 SVM。

Perceptron 得到 65% 正確率。 

SVM 得到 100% 正確率！！我們接下來要研究 SVM 的理論。

最後這張是 multi-layer perceptron (back-propagation 的結果。 accuracy= 81%

這個 mlp 是在 github 上找到的，特別針對sklearn寫的。原版程式經 ry 抓了幾隻蟲後暫可執行。