MNIST training data
---------------------------------

Training samples: 60,000
Testing  samples: 10,000

feature.dat

feature vector of sample 1
feature vector of sample 2
feature vector of sample 3
..
feature vector of sample n

It consists of 60,000 records. 
Feature vectors are sequentially stored in the file. Its dimension is 576 
and data type is "float" (4 bytes)

label.dat

label of sample 1
label of sample 2
...
label of sample n

The corresponding labels of each sample in the file "feature.dat"
label: 0 -- 9. Data type is "int"(4 bytes)

For MNIST testing data, the format is the same as above.

test_feature.dat (10,000 samples)
test_label.dat 

Data are writen into a file in binary format in Intel Pc plaform. If
you access the data in Sun workstation, you need to change the data order.

If you like to test your SvM's algorithm, "C" and RBF kernel are set to be

C = 10.0; 
RBF kernel: exp(-1.667||x1-x2||^2)

If you have problem to access the data, feel free to contact me.


If you like to know how I generate the data set, you can get more details in Paper

[1] J.X. Dong, A. Krzyzak and C.Y. Suen, A fast SVM training algorithm.
    International Journal of Pattern Recognition and Artificial Intelligence,
    vol. 17, No. 3, pp. 1--18, 2003. Electronic version:
    http://www.cenparmi.concordia.ca/~people/jdong/publication.html

--------------------------------------------------------------
Jianxiong Dong

jdong@cenparmi.concordia.ca

