Simple GAN to generate 'fake' digits that minic the MNIST dataset.
Note: tutorial states that they train for 100 epochs.
- It seems like after I do some training, the generated image only updates on the second call. Is that true?
- Can I disconnect from a running kernel and reconnect to it?
- What's the ratio of the time taken to do a training batch on my laptop vs on p2.xlarge? What's the optimum batch size?