Two tiny improvements.

[maximumspatium]

• fix default path for storing trained model - it must be RELATIVE to the current path
• add an argument for specifying number of training steps in the command line
• replace scipy.misc.imread with imageio.imread because the former isn't supported in newer scipy (>= 1.3.0)

FYI, more personal improvements here: maximumspatium@f9ce0e1