Kaggle
Competition of classification
Data
There are 32 x 32 pixels images in color. The goal is to classify them into 10 categories.
Rules
- Do not use extern machine learning library
- Do not use more data than given
Resuls
knn
The knn method gives about 10 % of accuracy on the test data.
logistic regression - One vs all
It gives 20 %.
SVM
This method goes up to 27% with global descriptors.
Libraries
Needs cvxopt, numpy and pandas. Executed with python3.5