def __init__(self, actions, mapname, contextD, context_rewards, **kwargs):
"""Initialize the environment variables.
:param actions: actions available to the system
:type actions: list of tuples (action_name,action_vector)
:param mapname: filename for map file
:param contextD: dimension of vector representing context
:param context_rewards: mapping from region labels to rewards for being
in that region (each entry represents one context)
:type context_rewards: dict {"regionlabel":rewardval,...}
:param **kwargs: see PlaceCellEnvironment.__init__
"""
PlaceCellEnvironment.__init__(self,
actions,
mapname,
name="ContextEnvironment",
**kwargs)
self.rewards = context_rewards
# generate vectors representing each context
self.contexts = {} # mapping from region label to context vector
for i, r in enumerate(self.rewards):
self.contexts[r] = list(MU.I(contextD)[i])
self.context = self.contexts[random.choice(self.contexts.keys())]
# randomly pick a new context every context_delay seconds
self.context_delay = 60
self.context_update = self.context_delay
self.create_origin("placewcontext",
lambda: self.place_activations + self.context)
self.create_origin("context", lambda: self.context)
def __init__(self, actions, mapname, contextD, context_rewards, **kwargs):
"""Initialize the environment variables.
:param actions: actions available to the system
:type actions: list of tuples (action_name,action_vector)
:param mapname: filename for map file
:param contextD: dimension of vector representing context
:param context_rewards: mapping from region labels to rewards for being in that
region (each entry represents one context)
:type context_rewards: dict {"regionlabel":rewardval,...}
:param **kwargs: see PlaceCellEnvironment.__init__
"""
PlaceCellEnvironment.__init__(self, actions, mapname, name="ContextEnvironment", **kwargs)
self.rewards = context_rewards
# generate vectors representing each context
self.contexts = {} # mapping from region label to context vector
for i, r in enumerate(self.rewards):
# self.contexts[r] = list(RandomHypersphereVG().genVectors(1, contextD)[0])
self.contexts[r] = list(MU.I(contextD)[i])
self.context = self.contexts[random.choice(self.contexts.keys())]
# randomly pick a new context every context_delay seconds
self.context_delay = 60
self.context_update = self.context_delay
self.create_origin("placewcontext", lambda: self.place_activations + self.context)
self.create_origin("context", lambda: self.context)
def tick(self):
if self.is_in(self.state, "a"):
self.in_hand = True
elif self.rewardamount > self.rewardresetamount:
self.in_hand = False
PlaceCellEnvironment.tick(self)
def __init__(self, *args, **kwargs):
"""Initialize environment variables.
:param name: name for environment
:param *args: see PlaceCellEnvironment.__init__
:param **kwargs: see PlaceCellEnvironment.__init__
"""
PlaceCellEnvironment.__init__(self, name="DeliveryEnvironment", *args,
**kwargs)
# reward value when no reward condition is met
self.defaultreward = -0.05
self.contexts = {"in_hand": [1, 0], "out_hand": [0, 1]}
self.in_hand = False
self.create_origin("placewcontext",
lambda: (self.place_activations +
self.contexts["in_hand"] if self.in_hand
else self.place_activations +
self.contexts["out_hand"]))
self.create_origin("context", lambda: (self.contexts["in_hand"]
if self.in_hand else
self.contexts["out_hand"]))
def tick(self):
if self.is_in(self.state, "a"):
self.in_hand = True
elif self.rewardamount > self.rewardresetamount:
self.in_hand = False
PlaceCellEnvironment.tick(self)
def __init__(self, *args, **kwargs):
"""Initialize environment variables.
:param name: name for environment
:param *args: see PlaceCellEnvironment.__init__
:param **kwargs: see PlaceCellEnvironment.__init__
"""
PlaceCellEnvironment.__init__(self,
name="DeliveryEnvironment",
*args,
**kwargs)
# reward value when no reward condition is met
self.defaultreward = -0.05
self.contexts = {"in_hand": [1, 0], "out_hand": [0, 1]}
self.in_hand = False
self.create_origin(
"placewcontext", lambda:
(self.place_activations + self.contexts["in_hand"] if self.in_hand
else self.place_activations + self.contexts["out_hand"]))
self.create_origin(
"context", lambda:
(self.contexts["in_hand"]
if self.in_hand else self.contexts["out_hand"]))
def gen_encoders(self, N, contextD, context_scale):
"""Generates encoders for state population in RL agent.
State aspect of encoders comes from PlaceCellEnvironment. Context component is a
unit vector with contextD dimensions and length context_scale.
"""
s_encoders = PlaceCellEnvironment.gen_encoders(self, N)
# c_encoders = RandomHypersphereVG(True,1.0,1.0).genVectors(N, d) #0.7
c_encoders = [random.choice(MU.I(contextD)) for _ in range(N)]
c_encoders = [[x * context_scale for x in enc] for enc in c_encoders]
encoders = [s + list(c) for s, c in zip(s_encoders, c_encoders)]
encoders = [[x / math.sqrt(sum([y ** 2 for y in e])) for x in e] for e in encoders]
return encoders
def gen_encoders(self, N, contextD, context_scale):
"""Generates encoders for state population in RL agent.
State aspect of encoders comes from PlaceCellEnvironment. Context
component is a unit vector with contextD dimensions and length
context_scale.
"""
s_encoders = PlaceCellEnvironment.gen_encoders(self, N)
c_encoders = [random.choice(MU.I(contextD)) for _ in range(N)]
c_encoders = [[x * context_scale for x in enc] for enc in c_encoders]
encoders = [s + list(c) for s, c in zip(s_encoders, c_encoders)]
encoders = [[x / math.sqrt(sum([y**2 for y in e])) for x in e]
for e in encoders]
return encoders
def tick(self):
PlaceCellEnvironment.tick(self)
self.update_context()
def tick(self):
PlaceCellEnvironment.tick(self)
self.update_context()
