def __init__(self, logging=True, confirm=True):
log_help = "\nexport NSPR_LOG_MODULES=timestamp,nsHttp:5,nsSocketTransport:5,nsStreamPump:5,nsHostResolver:5\n"
log_help += "export NSPR_LOG_FILE={}\n".format(FFLOG)
if confirm:
print(log_help)
ans = input("Have you set the environment variables for Firefox httplogging [y]? ")
print()
if ans != 'y':
sys.exit(0)
self.logger = simple_logger.SimpleLogger()
if logging:
self.logger.activate()
else:
self.logger.deactivate()
type=int,
default=3000)
args = parser.parse_args()
if args.test_checkpoints:
checkpoint_ix = ast.literal_eval(args.checkpoint_range)
else:
checkpoint_ix = [0]
model_name = args.modelname
test_checkpoints = args.test_checkpoints
sl = simple_logger.SimpleLogger(f"eval_csv/{model_name}.csv",
['checkpoint', 'score', 'episodes', 'steps'])
# All the evaluations will be evaluated on MineRLObtainDiamondVectorObf-v0 environment
MINERL_GYM_ENV = os.getenv('MINERL_GYM_ENV', 'MineRLTreechop-v0')
MINERL_MAX_EVALUATION_EPISODES = args.no_episodes #int(os.getenv('MINERL_MAX_EVALUATION_EPISODES', 10))
# Parallel testing/inference, **you can override** below value based on compute
# requirements, etc to save OOM in this phase.
EVALUATION_THREAD_COUNT = args.no_cpu #int(os.getenv('EPISODES_EVALUATION_THREAD_COUNT', 4))
device = "cuda" if torch.cuda.is_available() else "cpu"
class EpisodeDone(Exception):
pass
def run_agent_on_episode(self, single_episode_env: Episode, index):
"""Runs the agent on a SINGLE episode.
Args:
single_episode_env (Episode): The episode on which to run the agent.
"""
reward_sum = 0
counter = 0
er = EpisodeRecorder()
max_steps = args.max_steps
seeds = [
2, 12345, 45678, 303544, 744421, 816128, 406373, 99999, 88888,
76543, 67345, 11342, 24337, 45547, 82666, 128939, 494668, 927113,
869989, 873766, 708928, 786039, 371309, 281286
]
with torch.no_grad():
seed = random.sample(seeds, 1)
single_episode_env.seed(seed)
obs = single_episode_env.reset()
er.record_frame(obs['pov'])
done = False
state = self.model.get_zero_state(1, device=device)
s = torch.zeros((1, 1, 64), dtype=torch.float32, device=device)
while not done:
# what is happening here?
spatial = torch.tensor(
obs["pov"], device=device,
dtype=torch.float32).unsqueeze(0).unsqueeze(0).transpose(
2, 4)
#cv2.imshow("xdd", obs["pov"])
# cv2.waitKey(30)
#nonspatial = torch.cat([torch.tensor(obs["vector"], device=device, dtype=torch.float32),
# torch.ones((2,), device=device,dtype=torch.float32)], dim=0).unsqueeze(0).unsqueeze(0)
nonspatial = torch.zeros((1, 1, 64))
s, state = self.model.sample(
spatial, nonspatial, s, state,
torch.zeros((1, 1, 64), dtype=torch.float32,
device=device))
action = transform_int_to_actions([int(s)])
for i in range(1):
obs, reward, done, _ = single_episode_env.step(action)
er.record_frame(obs['pov'])
counter += 1
reward_sum += reward
if reward > 0:
rewards.append(reward)
if done:
break
if max_steps and counter > max_steps:
done = True
if args.save_vids:
er.save_vid(
f'train/{model_name}/eval_vids/checkpoint={self.checkpoint_number}/episode={index}_seed={seed}.avi'
)
sl = simple_logger.SimpleLogger(
f"eval_csv/{model_name}.csv",
['checkpoint', 'score', 'episodes', 'steps'])
sl.log([self.checkpoint_number, reward_sum, 1, max_steps])
import keywords
import ocr
import os
import re
import requests
import simple_logger
import sys
import time
import urllib
import utils
import website_fetcher
import goslate
# import ngrams
logger = simple_logger.SimpleLogger()
logger.activate()
#############
# ARGUMENTS #
#############
# header data for google search
HEADERS = {
'user-agent':
'Mozilla/5.0 (Macintosh; Intel Mac OS X 10.9; rv:32.0) Gecko/20100101 Firefox/32.0',
}
# regular expression for extracting urls
GOOGLERX = re.compile("""http[s]?://[^\s'"]*""")