def testValueBased(self):
""" Test value-based learner.
"""
mkt = SmartMarket(self.case)
exp = MarketExperiment([], [], mkt)
for g in self.case.generators:
env = DiscreteMarketEnvironment([g], mkt)
dim_state, num_actions = (10, 10)
exp.tasks.append(ProfitTask(env, dim_state, num_actions))
module = ActionValueTable(dim_state, num_actions)
module.initialize(1.0)
# module = ActionValueNetwork(dimState=1, numActions=4)
learner = SARSA() #Q() QLambda()
# learner.explorer = BoltzmannExplorer() # default is e-greedy.
exp.agents.append(LearningAgent(module, learner))
for _ in range(1000):
exp.doInteractions(24) # interact with the env in batch mode
for agent in exp.agents:
agent.learn()
agent.reset()
def testValueBased(self):
""" Test value-based learner.
"""
mkt = SmartMarket(self.case)
exp = MarketExperiment([], [], mkt)
for g in self.case.generators:
env = DiscreteMarketEnvironment([g], mkt)
dim_state, num_actions = (10, 10)
exp.tasks.append(ProfitTask(env, dim_state, num_actions))
module = ActionValueTable(dim_state, num_actions)
module.initialize(1.0)
# module = ActionValueNetwork(dimState=1, numActions=4)
learner = SARSA() #Q() QLambda()
# learner.explorer = BoltzmannExplorer() # default is e-greedy.
exp.agents.append(LearningAgent(module, learner))
for _ in range(1000):
exp.doInteractions(24) # interact with the env in batch mode
for agent in exp.agents:
agent.learn()
agent.reset()
""" Create an environment for each agent with an asset and a market. """
env = ParticipantEnvironment(g, mkt, n_offbids=2)
""" Create a task for the agent to achieve. """
task = ProfitTask(env)
""" Build an artificial neural network for the agent. """
net = buildNetwork(task.outdim, task.indim, bias=False, outputbias=False)
# net._setParameters(array([9]))
""" Create a learning agent with a learning algorithm. """
agent = LearningAgent(module=net, learner=ENAC())
""" Initialize parameters (variance). """
# agent.setSigma([-1.5])
""" Set learning options. """
agent.learner.alpha = 2.0
# agent.learner.rprop = True
agent.actaspg = False
# agent.disableLearning()
agents.append(agent)
tasks.append(task)
""" The Experiment will coordintate the interaction of the given agents and
their associated tasks. """
experiment = MarketExperiment(tasks, agents, mkt)
experiment.setRenderer(ExperimentRenderer())
""" Instruct the experiment to coordinate a set number of interactions. """
experiment.doInteractions(3)
# Create the market and associate learning agents with each generator.
market = SmartMarket(case)
# Specify the discrete set of possible markups on marginal cost.
markups = (0.1, 0.2, 0.33, 0.5, 0.6, 0.75, 1.0)
# Specify the number of offers/bids each participant can submit.
numOffbids = 4
# Specify the desired number of discrete states.
dimState = 10
dimAction = len(markups) * numOffbids
# Construct an experiment to test the market.
experiment = MarketExperiment([], [], market)
# Add the agents and their tasks.
for g in case.generators:
env = DiscreteMarketEnvironment([g], market, dimState, markups, numOffbids)
task = ProfitTask(env)
module = ActionValueTable(dimState, dimAction)
module.initialize(1.0)
# learner = SARSA(gamma=0.9)
learner = Q()
# learner = QLambda()
# learner.explorer = BoltzmannExplorer() # default is e-greedy.
agent = LearningAgent(module, learner)
agent.name = g.name
experiment.tasks.append(task)
mkt = SmartMarket(case)
agents = []
tasks = []
for g in bus1.generators:
""" Create an environment for each agent with an asset and a market. """
env = ParticipantEnvironment(g, mkt, n_offbids=2)
""" Create a task for the agent to achieve. """
task = ProfitTask(env)
""" Build an artificial neural network for the agent. """
net = buildNetwork(task.outdim, task.indim, bias=False, outputbias=False)
# net._setParameters(array([9]))
""" Create a learning agent with a learning algorithm. """
agent = LearningAgent(module=net, learner=ENAC())
""" Initialize parameters (variance). """
# agent.setSigma([-1.5])
""" Set learning options. """
agent.learner.alpha = 2.0
# agent.learner.rprop = True
agent.actaspg = False
# agent.disableLearning()
agents.append(agent)
tasks.append(task)
""" The Experiment will coordintate the interaction of the given agents and
their associated tasks. """
experiment = MarketExperiment(tasks, agents, mkt)
experiment.setRenderer(ExperimentRenderer())
""" Instruct the experiment to coordinate a set number of interactions. """
experiment.doInteractions(3)
# Create the market and associate learning agents with each generator.
market = SmartMarket(case)
# Specify the discrete set of possible markups on marginal cost.
markups = (0.1, 0.2, 0.33, 0.5, 0.6, 0.75, 1.0)
# Specify the number of offers/bids each participant can submit.
numOffbids = 4
# Specify the desired number of discrete states.
dimState = 10
dimAction = len(markups) * numOffbids
# Construct an experiment to test the market.
experiment = MarketExperiment([], [], market)
# Add the agents and their tasks.
for g in case.generators:
env = DiscreteMarketEnvironment([g], market, dimState, markups, numOffbids)
task = ProfitTask(env)
module = ActionValueTable(dimState, dimAction)
module.initialize(1.0)
# learner = SARSA(gamma=0.9)
learner = Q()
# learner = QLambda()
# learner.explorer = BoltzmannExplorer() # default is e-greedy.
agent = LearningAgent(module, learner)
agent.name = g.name
experiment.tasks.append(task)