Tensorflow-fashion-mnist

Getting to know DL better with practice using the fashion-MNIST dataset. More compute needed, results are not record breaking but it works !

Data structure

• 60k Training
• 5k Validation
• 5k Testing

Convnet Structure

• 5x5x32 convolution
• max pool, k=2
• relu
• 5x5x64 convolution
• max pool. k=2
• relu
• fully connected, 128 nodes
• output, 10 nodes, for 10 classes

Training

• Gradient Descent optimizer
• learning rate decay
• dropout on fc layer (0.8)
• minibatch