Closed alepacheco closed 7 years ago
I tried the MNIST Classifier and I have 96% of accuracy just after the very first epoch. Which backend do you use ? Do you train your classifier on CPU or GPU ? Which version of Keras do you use ?
It turns out I didn’t normalized the input
x_train /= 255
x_test /= 255
After that It’s working as expected, it turns out normalization is very important.
I have look up the keras docummentaion and build a basic MNIST Classifier on Kaggle but it does't go past 30% accuracy. I'm using this model: https://github.com/fchollet/keras/blob/master/examples/mnist_mlp.py (98.40%) with the Kaggle sample data and keras.
https://www.kaggle.com/alepacheco/simple-mnist