Simple SGD - Learning Problem

[amimem]

The observations suggest that the algorithm does not learn after a few iterations.

To do:

• [x] Testing the algorithm with different seeds to make sure it is not due to a bad initialization point
• [x] Testing the algorithm with different dataset
• [x] Testing our own dataset on a ready-to-run SGD code

[amimem]

Results:

• Testing with different seeds: no effect
• Testing the algorithm with different datasets: our algorithm works perfectly on MNIST
• Testing our own dataset on a ready-to-run SGD code: We get the same results as of our own algorithm

I'm almost certain that something is wrong with our dataset/architecture.

[amimem]

I'm gonna close this issue, since #4 represents better the problem and what we have to do next