writing down the parameter space

[yamins81]

Look at the function get_model in curiosity repo in the file

curiosity/models/obj_detector_feedforward.py

initial strategy:
1) in parallel randomly draw from parameter space and run models [do on openmind]
2) if there's any traction, we can do hyperopt (look at hyperopt package)
--> traction = does anything besides the base model train at all?
is there any range in testing performance? [needs to comparable across models]
is there any range in consistency to humans?
is there any range in consistency to base model?