def fit_inverse_model(self):
# Pre-processing
self.pre_process()
has_fit_e = has_fit_f = False
# Fit change-point model for electricity consumption
print('Fitting electricity model...')
if (hasattr(self, "weather_electricity")):
self.im_electricity = model.InverseModel(self.weather_electricity.v_T_C,
self.eui_daily_electricity,
'Electricity')
has_fit_e = self.im_electricity.fit_model()
if (has_fit_e):
self.im_electricity.plot_IM(self)
# Fit change-point model for fossil fuel consumption
print('Fitting fossil fuel model...')
if (hasattr(self, "weather_fossil_fuel")):
self.im_fossil_fuel = model.InverseModel(self.weather_fossil_fuel.v_T_C,
self.eui_daily_fossil_fuel,
'Fossil Fuel')
has_fit_f = self.im_fossil_fuel.fit_model()
if (has_fit_f): self.im_fossil_fuel.plot_IM(self)
return (has_fit_e or has_fit_f)
import conf
# Tensorboard logging
import tensorboard_logger as tboard
timestamp = str(round(time.time()))[-6:]
tb_dir = conf.DIR + "/runs/es_" + timestamp
tboard.configure(tb_dir, flush_secs=5)
# gym
env = gym.make("BreakoutDeterministic-v4")
conf.OUTPUT_NUM = env.action_space.n
# models and learning
encoder = model.Encoder()
inverse_model = model.InverseModel(encoder=encoder)
criterion = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.Adadelta(
inverse_model.parameters(),
lr=0.1) #torch.optim.Adam(inverse_model.parameters(), lr=0.001)
# buffer
memory = memory.MemoryBuffer(size=20000,
batch_size=8,
replace=True,
bias_prob=0.3)
curr_state = None
episode_i = 0
while True: