* Python packages.
  Ensure that you use the requirements.txt file to install the appropriate dependencies via pip.
  pip install -r requirements.txt

  If you are working on Mac, all you need to do is
  sudo easy_install pip
  pip install -r requirements.txt

* Executable scripts. See help messages by giving -h option.
  simulate.py   Generate dialogues with simulated users.
  texthub.py    Text interface

* Config files.
  There are some examples of config files in config directory.

* Configurable options.
  [DEFAULT] default configs. (can be shared with RNN tracker)
    ontology: string. path to ontology file.
    database: string. path to database file.

  [logging] context logger setting.
    file: string. log file name.
    file_level: string. logging level for file.
    screen_level: string. logging level for screen.

  [hub] configs for hubs.
    semi: string. semantic decoder name. 'PassthroughSemI' by default. Note that input string should be dialog-act type
      in the case of baseline trackers. Natural sentences are available for word-based RNN tracker.
    semo: string. output generator name. 'PassthroughSemO' by default. 'BasicSemO' for template-based output generator.

  [simulate] simulate script.
    continuewhensuccessful: boolean.
    maxturns: int.
    forcenullpositive: boolean. append the hypothesis list with null() and a tiny probability (0.001).
    confscorer: string. confidence scorer. 'additive' by default.

  [policy] dialogue manager.
    useconfreq: boolean. If true the policy is allowed to add a venue count in system actions,
      e.g. inform(count=20, food=chinese). This addition happens in summary-master mapping.
	policytype: string. 'hdc' for handcrafted policy (default), 'type' for text-input manual policy, 'gp' for gp policy and
	  'mcc' for monte carlo control policy.
    belieftype: string. 'rnn' for recurrent neural net tracker. 'baseline' for baseline tracker and
      'focus' for focus baseline tracker (default).
    startwithhello: boolean. If true the first action will be 'hello()'.
    bcm: boolean. If true the system is using a policy committee
    currpolicy: int. learning committee member


  [dbase] database.

  [em] error model.
    nbestsize: int.
    confusionmodel: string. 'RandomConfusion' is only available at this point.
    nbestgeneratormodel: string. 'UniformNBestGenerator' or 'SampledNBestGenerator'

  [eval] evaluation and reward function.
    reward_venue_recommended: int.
    penaliseallturns: int.
    wrongvenuepenalty: int.
    notmentionedvaluepenalty: int.

  [basicsemo] template-based output generator.
    templatefile: string. path to template file.
    emphasis: boolean. whether to use emphasized outputs.
    emphasisopen: string. open tag. '<EMPH>' by default.
    emphasisclose: string. close tag. '</EMPH>' by default.

  [goalgenerator] goal generator.
    patience: int. patience level.
    maxvenuespergoal: int.

  [um] user model.
    usenewgoalscenarios: boolean. If true the simulated user is allowed to change their goals during dialogues.
    answerreqalways: boolean.

  [nnpolicy] neural net policy.

  [mccpolicy] mcc policy
    gamma: float. 1.0 by default
    nu: float. 0.0001 by default
    epsilon_start: float. 1.0 by default
    epsilon_end: float. 0.1 by default
    maxIter: float. 2000 by default

  [gppolicy] gp policy
    kernel: string. 'polysort' by default
    thetafile: string. '' by default
    revertfile: string. '' by default
    replacefile: string. '' by default

  [gpsarsa] gpsarsa
    saveasprior: boolean. False by default
    random: boolean. False by default
    learning: boolean. False by default
    gamma: float. 1.0 by default
    sigma: float. 5.0 by default
    nu: float. 0.01 by default
    numprior: int. 0 by default
    scale: int. -1 by default

    
    
  [DEFAULT] [classifier] [track] [track_unsup] [train] [evaluate] [experiment] [seed] are sections for Matt's RNN tracker.