Benchmarking for the RPI

[michelucci]

Here is a possible way of creating a benchmarking suite for the Raspberry Pi. We need something that is reproducible but that has a memory footprint that is not too big. Example:

Dataset: MNIST (can be loaded with Keras)

Run and printout the running time by training the following networks for, say, 10 epochs (this value can be changed):

Always two layers. In each layer n neurons for n=[5, 10, 100, 500, 1000, 5000, 10**4, 10**5]

The idea would be to have a python script, let's say moist_rpi_benchmark.py that printout the running times for all networks. Or even better one could print it and save it in a file.