Update and rename readme.txt to readme.md

Author: soloice

readme.txt renamed to readme.md

@@ -12,3 +12,13 @@ Finally, svm_test_full.py trains a SVM classifier on the whole MNIST data.
 In my experiment, I found training an SVM with 'RBF' kernel is much faster than that with linear kernel.  I don't why.  Perhaps it is because in RKHS the data points are more separable thus facilitates the training procedure.
 For your reference, Training a MultiSVM classifier with 'RBF' kernel on 6/7 MNIST data (i.e., using 60k examples as the training set) takes 11462s on my workstation (32GB RAM, 1 CPU with 8 Intel(R) Xeon(R) CPU E5-1620 v2 @ 3.70GHz cores.)
 
+Overall, the can be summarized as follows:
+
+| algorithm	| running time(s)	| average precision	| average recall	| average F1-score |
+|:-------------:|:-------------:|:-------------:|:-------------:|:------------:|
+| SMO + Linear Kernel	| 9684 + 12	| 0.91	| 0.91	| 0.91 |
+| SMO + RBF Kernel	| 666 + 54	| 0.92	| 0.92	| 0.92 |
+| QP + Linear Kernel	| 225 + 11	| 0.91	| 0.91	| 0.91 |
+| QP + RBF Kernel	| 267 + 58	| 0.92	| 0.92	| 0.92 |
+| Sklearn svm + Linear Kernel	| 12 + 118	| 0.92	| 0.92	| 0.92 |
+| Sklearn svm + RBF Kernel	| 30 + 232	| 0.92	| 0.92	| 0.92 |