lemonade117/mnist

使用多种方法解决MNSIT问题

MNIST

Introduction

MNIST is a classic data set in the field of machine learning and deep learning. There are many methods for this data set. Here are some methods for reference.

the data resource is kaggle Digit Recognizer

Algorithms

• ML

  • SVM
  • DecesionTree
  • RandomForest
  • KNeighbors
  • Adaboost
  • XGBoost
  • catboost
  • lgbm

• DL

  • FC
  • CNN（VGG16）
  • LSTM
  • BLS

Score on test data

here is the kaggle Digit Recongizer

The accuracy is verified in the kaggle competition, and all algorithms have not been adjusted or optimized.

Test on: GPU: Tesla P100 x1 CPU: 8 kernels, 64G RAM

Algorithm	Score	Time Cost for training/s
SVM	0.11614	8504
DecisionTree	0.85585	10.90
RandomForest	0.94142	2.61
KNeighbors	0.96800	5.63
Adaboost	0.72914	23.32
XGBoost
catboost
lgbm
neural network with numpy	0.92214	662 (10 epochs)
VGG16	0.98828	676 (32 epochs)
LSTM	0.90785	1447 (32 epochs)
BLS	0.93471	30.43