flinker517 / mnist_tutorial

A tutorial for MNIST handwritten digit classification using sklearn, PyTorch and Keras.

mnist_tutorial

A tutorial for MNIST handwritten digit classification using sklearn and PyTorch.

Code structure

• numpy_matplotlib_sklearn.ipynb: for numpy, matplotlib and sklearn.
• pytorch.ipynb: for pytorch.
• keras.ipynb: for keras (not used).

Requirements (default)

Code tested on following environments, other version should also work:

• python 3.6.3
• numpy 1.13.3
• matplotlib 2.1.0
• sklearn 0.19.1
• pytorch 0.4.1

Result

For numpy_matplotlib_sklearn.ipynb

• logistic regression: Training accuracy is 95.30%, testing accuracy is 88.10%.
• Naive bayes (Bernoulli): Training accuracy is 81.78%, testing accuracy is 80.90%.
• Support vector machine: Training accuracy is 97.73%, testing accuracy is 85.40%.
• Adjusted SVM: for LinearSVC with loss='hinge', training accuracy is 95.92% and testing accuracy is 86.80%;for NuSVC, training accuracy is 88.10% and testing accuracy is 86.50%;for SVC, training accuracy is 91.25% and testing accuracy is 89.60%. For pytorch.py
• With epoch = 10, learing rate = 0.001, batch size = 128 and lenet5, training accuracy is 99.72% and test accuracy is 98.96%.