I mean, if you can do it just in python probably you don't understand it...

Also, if in the code I name the variables as "batches", in reality I used the Stochastic Gradient Descent algorithm. Maybe in the future I can modify it to implement a real mini-batch algorithm.

The plots are done using GNUplot and Perl. You should do something like:

./main >> data.log
chmod +x plot_training.pl
./plot_training.pl data.log