def __init__(self,
xdim=None,
xopt=None,
xbound=5,
feasible=True,
constrained=False,
violation=False,
**args):
self.feasible = feasible
self.constrained = constrained
self.violation = violation
self.xbound = xbound
if xdim is None:
xdim = self.xdim
if xdim is None:
xdim = self.xdimMin
assert xdim >= self.xdimMin and not (self.xdimMax is not None
and xdim > self.xdimMax)
self.xdim = xdim
if xopt is None:
self.xopt = zeros(self.xdim)
else:
self.xopt = xopt
setAllArgs(self, args)
self.reset()
def __init__(self, game, **kwargs):
setAllArgs(self, kwargs)
self._game = game
self._avatar_types = []
self._abs_avatar_types = []
self._other_types = []
self._mortal_types = []
for skey in sorted(game.sprite_constr):
sclass, _, stypes = game.sprite_constr[skey]
if issubclass(sclass, Avatar):
self._abs_avatar_types += stypes[:-1]
self._avatar_types += [stypes[-1]]
if issubclass(sclass, RotatingAvatar) or issubclass(
sclass, ShootAvatar):
self.orientedAvatar = True
if skey not in game.sprite_groups:
continue
ss = game.sprite_groups[skey]
if len(ss) == 0:
continue
if isinstance(ss[0], Avatar):
assert issubclass(ss[0].physicstype, GridPhysics), \
'Not supported: Game must have grid physics, has %s'\
% (self._avatar.physicstype.__name__)
else:
self._other_types += [skey]
if not ss[0].is_static:
self.staticOther = False
assert self.staticOther, "not yet supported: all non-avatar sprites must be static. "
self._avatar_types = sorted(
set(self._avatar_types).difference(self._abs_avatar_types))
self.uniqueAvatar = (len(self._avatar_types) == 1)
#assert self.uniqueAvatar, 'not yet supported: can only have one avatar class'
# determine mortality
for skey, _, effect, _ in game.collision_eff:
if effect in kill_effects:
if skey in self._avatar_types + self._abs_avatar_types:
self.mortalAvatar = True
if skey in self._other_types:
self.mortalOther = True
self._mortal_types += [skey]
# retain observable features, and their colors
self._obstypes = {}
self._obscols = {}
for skey in self._other_types:
ss = game.sprite_groups[skey]
self._obstypes[skey] = [
self._sprite2state(sprite, oriented=False) for sprite in ss
if sprite.is_static
]
self._obscols[skey] = ss[0].color
if self.mortalOther:
self._gravepoints = {}
for skey in self._mortal_types:
for s in self._game.sprite_groups[skey]:
self._gravepoints[(skey, self._rect2pos(s.rect))] = True
def __init__(self, game, **kwargs):
setAllArgs(self, kwargs)
self._game = game
self._avatar_types = []
self._abs_avatar_types = []
self._other_types = []
self._mortal_types = []
for skey in sorted(game.sprite_constr):
sclass, _, stypes = game.sprite_constr[skey]
if issubclass(sclass, Avatar):
self._abs_avatar_types += stypes[:-1]
self._avatar_types += [stypes[-1]]
if issubclass(sclass, RotatingAvatar) or issubclass(sclass, ShootAvatar):
self.orientedAvatar = True
if skey not in game.sprite_groups:
continue
ss = game.sprite_groups[skey]
if len(ss) == 0:
continue
if isinstance(ss[0], Avatar):
assert issubclass(ss[0].physicstype, GridPhysics), \
'Not supported: Game must have grid physics, has %s'\
% (self._avatar.physicstype.__name__)
else:
self._other_types += [skey]
if not ss[0].is_static:
self.staticOther = False
assert self.staticOther, "not yet supported: all non-avatar sprites must be static. "
self._avatar_types = sorted(set(self._avatar_types).difference(self._abs_avatar_types))
self.uniqueAvatar = (len(self._avatar_types) == 1)
#assert self.uniqueAvatar, 'not yet supported: can only have one avatar class'
# determine mortality
for skey, _, effect, _ in game.collision_eff:
if effect in kill_effects:
if skey in self._avatar_types+self._abs_avatar_types:
self.mortalAvatar = True
if skey in self._other_types:
self.mortalOther = True
self._mortal_types += [skey]
# retain observable features, and their colors
self._obstypes = {}
self._obscols = {}
for skey in self._other_types:
ss = game.sprite_groups[skey]
self._obstypes[skey] = [self._sprite2state(sprite, oriented=False)
for sprite in ss if sprite.is_static]
self._obscols[skey] = ss[0].color
if self.mortalOther:
self._gravepoints = {}
for skey in self._mortal_types:
for s in self._game.sprite_groups[skey]:
self._gravepoints[(skey, self._rect2pos(s.rect))] = True
def __init__(self, agentName = "UnnamedClient", host = 'localhost', port = 4242, **otherargs):
"""General TCP Environment"""
setAllArgs(self, otherargs)
self.host = host
self.port = port
if self.verbose:
print "TCPENV: agentName ", agentName
self.client = Client(host, port, agentName)
self.connected = True
def __init__(self, provider, init_params, **kwargs):
self.provider = provider
self.parameters = init_params.copy()
self.paramdim = len(init_params)
self._num_updates = 0
setAllArgs(self, kwargs)
self._additionalInit()
if self.verbose:
self._printStuff()
def __init__(self, provider, init_params, **kwargs):
self.provider = provider
self.parameters = init_params.copy()
self.paramdim = len(init_params)
self._num_updates = 0
setAllArgs(self, kwargs)
self._additionalInit()
self.callback(self)
if self.verbose:
self._printStuff()
def __init__(self, xdim = None, xopt = None, **args):
if xdim is None:
xdim = self.xdim
if xdim is None:
xdim = self.xdimMin
assert xdim >= self.xdimMin and not (self.xdimMax is not None and xdim > self.xdimMax)
self.xdim = xdim
if xopt is None:
self.xopt = zeros(self.xdim)
else:
self.xopt = xopt
setAllArgs(self, args)
self.reset()
def __init__(self, num_actions, num_features, **kwargs):
ValueBasedLearner.__init__(self)
setAllArgs(self, kwargs)
self.explorer = None
self.num_actions = num_actions
self.num_features = num_features
if self.randomInit:
self._theta = randn(self.num_actions, self.num_features) / 10.
else:
self._theta = zeros((self.num_actions, self.num_features))
self._additionalInit()
self._behaviorPolicy = self._boltzmannPolicy
self.reset()
def __init__(self, num_actions, num_features, **kwargs):
ValueBasedLearner.__init__(self)
setAllArgs(self, kwargs)
self.explorer = None
self.num_actions = num_actions
self.num_features = num_features
if self.randomInit:
self._theta = randn(self.num_actions, self.num_features) / 10.
else:
self._theta = zeros((self.num_actions, self.num_features))
self._additionalInit()
self._behaviorPolicy = self._boltzmannPolicy
self.reset()
def __init__(self, xdim = None, xopt = None, **args):
if xdim is None:
xdim = self.xdim
if xdim is None:
xdim = self.xdimMin
assert xdim >= self.xdimMin and not (self.xdimMax is not None and xdim > self.xdimMax)
self.xdim = xdim
if xopt is None:
self.xopt = zeros(self.xdim)
else:
self.xopt = xopt
setAllArgs(self, args)
self.reset()
def __init__(self, xdim = None, xopt = None, xbound=5, feasible=True, constrained=False, violation=False, **args):
self.feasible=feasible
self.constrained=constrained
self.violation=violation
self.xbound=xbound
if xdim is None:
xdim = self.xdim
if xdim is None:
xdim = self.xdimMin
assert xdim >= self.xdimMin and not (self.xdimMax is not None and xdim > self.xdimMax)
self.xdim = xdim
if xopt is None:
self.xopt = zeros(self.xdim)
else:
self.xopt = xopt
setAllArgs(self, args)
self.reset()
def __init__(self, DS, **kwargs):
""" Initialize with the training data set DS. All keywords given are set as member variables.
The following are particularly important:
@param hidden: number of hidden units
@param TDS: test data set for checking convergence
@param VDS: validation data set for final performance evaluation
@param epoinc: number of epochs to train for, before checking convergence (default: 5)
"""
self.DS = DS
self.hidden=10
self.maxepochs=1000
self.Graph=None
self.TDS=None
self.VDS=None
self.epoinc = 5
setAllArgs( self, kwargs)
self.trainCurve = None
def __init__(self, DS, **kwargs):
""" Initialize with the training data set DS. All keywords given are set as member variables.
The following are particularly important:
@param hidden: number of hidden units
@param TDS: test data set for checking convergence
@param VDS: validation data set for final performance evaluation
@param epoinc: number of epochs to train for, before checking convergence (default: 5)
"""
self.DS = DS
self.hidden=10
self.maxepochs=1000
self.Graph=None
self.TDS=None
self.VDS=None
self.epoinc = 5
setAllArgs( self, kwargs)
self.trainCurve = None
def __init__(self, num_features, num_actions, indexOfAgent=None, **kwargs):
IndexableValueBasedLearner.__init__(self, indexOfAgent)
setAllArgs(self, kwargs)
self.explorer = None
self.indexOfAgent = indexOfAgent
self.num_actions = num_actions
self.num_features = num_features
if self.randomInit:
self._theta = randn(self.num_actions, self.num_features) / 10.
else:
self._theta = zeros((self.num_actions, self.num_features))
self._additionalInit()
self._behaviorPolicy = self._boltzmannPolicy
self.reset()
#
self.ownerAgentProperties["requireOtherAgentsState"] = False
self.ownerAgentProperties["requireJointAction"] = False
self.ownerAgentProperties["requireJointReward"] = False
def __init__(self, num_features, num_actions, indexOfAgent=None, **kwargs):
IndexableValueBasedLearner.__init__(self, indexOfAgent)
setAllArgs(self, kwargs)
self.explorer = None
self.indexOfAgent=indexOfAgent
self.num_actions = num_actions
self.num_features = num_features
if self.randomInit:
self._theta = randn(self.num_actions, self.num_features) / 10.
else:
self._theta = zeros((self.num_actions, self.num_features))
self._additionalInit()
self._behaviorPolicy = self._boltzmannPolicy
self.reset()
#
self.ownerAgentProperties["requireOtherAgentsState"]=False
self.ownerAgentProperties["requireJointAction"]=False
self.ownerAgentProperties["requireJointReward"]=False
def __init__(self, evaluator = None, initEvaluable = None, **kwargs):
""" The evaluator is any callable object (e.g. a lambda function).
Algorithm parameters can be set here if provided as keyword arguments. """
# set all algorithm-specific parameters in one go:
self.__minimize = None
self.__evaluator = None
setAllArgs(self, kwargs)
# bookkeeping
self.numEvaluations = 0
self.numLearningSteps = 0
if self.storeAllEvaluated:
self._allEvaluated = []
self._allEvaluations = []
elif self.storeAllEvaluations:
self._allEvaluations = []
if evaluator is not None:
self.setEvaluator(evaluator, initEvaluable)
def __init__(self, evaluator = None, initEvaluable = None, **kwargs):
""" The evaluator is any callable object (e.g. a lambda function).
Algorithm parameters can be set here if provided as keyword arguments. """
# set all algorithm-specific parameters in one go:
self.__minimize = None
self.__evaluator = None
setAllArgs(self, kwargs)
# bookkeeping
self.numEvaluations = 0
self.numLearningSteps = 0
if self.storeAllEvaluated:
self._allEvaluated = []
self._allEvaluations = []
elif self.storeAllEvaluations:
self._allEvaluations = []
if evaluator is not None:
self.setEvaluator(evaluator, initEvaluable)
def __init__(self, basefun, **kwargs):
""" Provide a stationary 1D base function. """
setAllArgs(self, kwargs)
self._basefun = basefun
self._optimum = 0
if not self.startatzero:
self._jump()
else:
self._timer = self.jumptime
# monkey-patching wrapper
tmp = self._basefun._newSample
def _oneStep(*args, **kwargs):
""" the timer is decremented only when _newSamples() is called """
#print self, self._timer, self._optimum
if 'override' not in kwargs or kwargs['override'] == False:
self._timer -= 1
if self._timer <= 0:
self._jump()
return tmp(*args, **kwargs)
self._basefun._newSample = _oneStep
def __init__(self, basefun, **kwargs):
""" Provide a stationary 1D base function. """
setAllArgs(self, kwargs)
self._basefun = basefun
self._optimum = 0
if not self.startatzero:
self._jump()
else:
self._timer = self.jumptime
# monkey-patching wrapper
tmp = self._basefun._newSample
def _oneStep(*args, **kwargs):
""" the timer is decremented only when _newSamples() is called """
#print self, self._timer, self._optimum
if 'override' not in kwargs or kwargs['override'] == False:
self._timer -= 1
if self._timer <= 0:
self._jump()
return tmp(*args, **kwargs)
self._basefun._newSample = _oneStep
def __init__(self, dataset, module, **kwargs):
setAllArgs(self, kwargs)
self.module = module
self.paramdim = module.paramdim
self._ready = False
self.reset(dataset)
def __init__(self, **args):
setAllArgs(self, args)
self._lastseen = None
def set(self, **kwargs):
""" convenience method to set several member variables at once """
setAllArgs(self, kwargs)
def set(self, **kwargs):
""" convenience method to set several member variables at once """
setAllArgs(self, kwargs)
def __init__(self, paramdim, loss_fun, gradient_fun, **kwargs):
self.paramdim = paramdim
self.loss_fun = loss_fun
self.gradient_fun = gradient_fun
setAllArgs(self, kwargs)
def __init__(self, dataset, module, **kwargs):
setAllArgs(self, kwargs)
self.module = module
self.paramdim = module.paramdim
self._ready = False
self.reset(dataset)
def __init__(self, **args):
setAllArgs(self, args)
self._lastseen = None
def __init__(self, paramdim, loss_fun, gradient_fun, **kwargs):
self.paramdim = paramdim
self.loss_fun = loss_fun
self.gradient_fun = gradient_fun
setAllArgs(self, kwargs)
