* clean up how groups of tubs are handled by adding a TubGroup class (#172)

* Extra steps for Mac installation of simulator

* clean up how groups of tubs are handled by adding a TubGroup class

* hot fix to prevent pi from failing to load socketio

* clean up how groups of tubs are handled by adding a TubGroup class

* fixed split paths for multiple tub groups

* made merging dataframes faster

* made merging dataframes faster

* refactor parts

* merge refactor branch

* name changes

* updated some of the commands (tubcheck, tubhist)

* cleaned up commands

* remove extra code, add tests

* add tests

* update docs

* remove old menu buttons

* remove old menu buttons

* update tests on pi

Extra steps for Mac installation of simulator

V 2.2.0 - better tubs, opencv, bug fixes

* create a tub for training against stack of N records through time

* joystick doc fixes

* added sdsandbox integration. testing time stack.

* specify keras and tensorflow version for linux desktop install

* added donkey sim integration

* added deps to setup.py

* added simulator docs

* create a movie from a tub

* improve error handling and usage for making movie from tub

* added docs for donkey commands

* use config instead of magic num

* fix sim steering scale. added steering plot analysis.

* rename movie command to makemovie

* smaller increment on delta max throttle

* more comments. added steer scale param

* add plot_predictions method to template

* no scale by default from sim server

* fixed linear model, added cropping and image norm

* add unit test for sim server

* document PID controls

* keep wanting to turn verbose off. added now as argument to train, along with others. default to on, as before.

* added linux joystick setup guide for simulator

* move sim command from manage.py to donkey command line utils

* A tricky bug that causes the cache of generator results

* command line arg for caching training data

* default to cache on. fix gether tubs

* rename analyze to histogram. added needed pandas dependancy for plot

* move gather tubs to tools and added tests. Added docs for plot_predictions and histogram

* increment version

* Update .travis.yml

* remove opencv deps from default setup until we have it on the disc image

* pi setup needed note on updating latest code

* joystick doc fixes

* specify keras and tensorflow version for linux desktop install

* create a movie from a tub

* improve error handling and usage for making movie from tub

* added docs for donkey commands

* use config instead of magic num

* rename movie command to makemovie

* smaller increment on delta max throttle

2.3.2 - Minor improvements to docs and CLI arguments

Js updates

Fixed quotes in documentation

If the terminal supports pasting unicode, it'll look alright, but not parse and the car won't join the network.

added command line option/sticky config to use joystick in manage.py. Added option to record on non-zero throttle

Check and fix

Dynamic steps per epoch

* --tub is now optional for train subcommand. It also takes comma-separated format

* new train docopt parameters

* ... causes wierd behavior in docopt

* Dynamatically generates clips based on timestamps so that it's always up to date

* Change POST to using dynamic clip recognition

* remove animation from steering progress bar to reveal jitterness of steering angles

* car.py -> manage.py. Ghost rebase commits keep coming back

* slightly more concise code

* Minor bug fix for simulation to load a model

* Minor bug in unbin_Y

BUG: record generator needs to throw StopIterator exception so that itertool.chain can move on to next tub (#144)