if len(s) > 0:
for i, d in enumerate(s):
s[i] = d.split(os.getcwd() + '/' + DB_DIR + '/')[1]
s[i] = int(s[i].split('.csv')[0])
ID = max(s) + 1 ## Initial ID
else:
ID = 0
for repetiton_id in range(REPETITIONS):
## Initialize Gym problem
problem = Gym(ENV_NAME,
iterations=ITERATIONS,
max_steps=MAX_STEPS,
action_mode=ACTION_MODE)
## Initialize UMDAc
umdac = UMDAc(model, problem=problem, gen_size=GEN_SIZE)
## Write to db
db = open(DB_DIR + '/' + str(ID) + '.csv', 'w')
dbwriter = csv.DictWriter(db, fieldnames=DB_FIELDS)
dbwriter.writeheader()
print('Repetition: ', repetiton_id)
print('id: ', ID)
### TRAINING ###
from UMDAc.Wrappers.Gym import Gym
### HYPERPARAMETERS ###
GENERATIONS = 500
GEN_SIZE = 200
SURV = .5
RAND_SURV = .3
NOISE = None
FILENAME = 'lunar_result.h5' # Filename of best specimen
# Problem specific
MAX_STEPS = 400
ITERATIONS = 3
## Initialize Gym problem
problem = Gym('LunarLander-v2', iterations=ITERATIONS, max_steps=MAX_STEPS)
## Initialize model
a = Input(shape=(8, ))
b = Dense(4)(a)
model = Model(inputs=a, outputs=b)
## Initialize UMDAc
umdac = UMDAc(model, problem=problem, gen_size=GEN_SIZE)
model.summary()
### TRAINING ###
for generation in range(GENERATIONS):
### HYPERPARAMETERS ###
GENERATIONS = 1000
GEN_SIZE = 100
SURV = .5
RAND_SURV = .3
NOISE = None
FILENAME = 'car_result.h5' # Filename of best specimen
ITERATIONS = 1
MAX_STEPS = 200
## Initialize Gym problem
problem = Gym('CarRacing-v0',
iterations=ITERATIONS,
max_steps=MAX_STEPS,
action_mode='raw')
in_shape = (96, 96, 3)
action_size = problem.env.action_space.shape[0]
## Build Network
model = Sequential()
model.add(
Convolution2D(5,
kernel_size=(10, 10),
strides=(10, 10),
activation='relu',
data_format="channels_last",
input_shape=in_shape))
def evaluate(self, ids):
from tqdm import tqdm
from UMDAc.UMDAc import UMDAc
from UMDAc.Wrappers.Gym import Gym
from GA.GA import GA
max_steps = input('Maximum step number [None]:')
if max_steps == '':
max_steps = None
else:
max_steps = int(max_steps)
action_mode = None
while action_mode not in range(2):
action_mode = input('action mode [0:raw/1:argmax]:')
action_mode = int(action_mode)
if action_mode == 0:
action_mode = 'raw'
elif action_mode == 1:
action_mode = 'argmax'
repetitions = input('Repetitions [100]:')
if repetitions == '':
repetitions = 100
else:
repetitions = int(repetitions)
print('')
print('--- Selected options ---')
print('Maximum steps: ', max_steps)
print('Action mode: ', action_mode)
print('Repetitions: ', repetitions)
print('')
if input('Are you sure to continue? [N/y] >') != 'y':
print('aborting...')
quit()
print('')
## Open evaluation log
evalog = open('evaluation.csv', 'a')
evalwriter = csv.DictWriter(
evalog,
fieldnames=['id', 'Algorithm', 'Environment', 'Total reward'])
for ID in ids:
print('Experiment: ', ID)
## Get experiment's environment
mainlog = self.get_main()
env = mainlog.loc[mainlog['id'] == ID]['Environment']
env = list(env)[0]
## Get algorithm used in experiment
alg = mainlog.loc[mainlog['id'] == ID]['Algorithm']
alg = list(alg)[0]
## Init env
problem = Gym(env,
iterations=1,
max_steps=max_steps,
action_mode=action_mode)
## Select algorithm
if alg == 'UMDAc':
algorithm = UMDAc(model=None, problem=problem, gen_size=None)
elif alg == 'GA':
algorithm = GA(model=None, problem=problem, gen_size=None)
else:
print('Error, invalid algorithm name, name found:', alg)
quit()
algorithm.load_model('db/' + str(ID) + '.h5')
for i in tqdm(range(repetitions)):
## Evaluate specimen, render enabled
tr = problem.evaluate(specimen=None,
model=algorithm.model,
render=False,
verbose=False)
evalwriter.writerow({
'id': ID,
'Algorithm': alg,
'Environment': env,
'Total reward': tr
})
evalog.close()
print('')
act_sel = None
while act_sel not in range(len(envs)):
act_sel = int(input('Selection > '))
action_mode = action_modes[act_sel]
from UMDAc.UMDAc import UMDAc
from UMDAc.Wrappers.Gym import Gym
### INITIALIZATION ###
## Init env
problem = Gym(envs[sel],
iterations=1,
max_steps=MAX_STEPS,
action_mode=action_mode)
## Init UMDAc
umdac = UMDAc(model=None, problem=problem, gen_size=None)
umdac.load_model(sname)
from tqdm import tqdm
log = []
print('')
if input('Render environment? [Y/n]') == 'n':
RENDER = False
else: