def run_reg(self):
print "run_reg"
run.setEnvVar( "assimp_path", self.assimp_bin_path )
run.run_test()
rc.run()
self.b_update_.config( state=ACTIVE )
return 0
def run_reg(self):
print "run_reg"
run.setEnvVar("assimp_path", self.assimp_bin_path)
run.run_test()
rc.run()
self.b_update_.config(state=ACTIVE)
return 0
def run_sr():
blocks = [
(
range(20, 100, 10),
range(0, MAX_ERRORS + 1),
),
(
range(20, 400, 20),
range(0, MAX_ERRORS + 1),
),
(
range(20, 800, 30),
range(0, MAX_ERRORS + 1),
),
(
range(20, 1010, 40),
range(0, MAX_ERRORS + 1),
),
(
range(20, 1010, 10),
range(0, MAX_ERRORS + 1),
),
]
for dell, errl in blocks:
run.run_test(
test_url=
'http://www.quantified-mind.com/tests/simple_reaction_time/practice',
js_script_template=js_scripts.SIMPLE_REACTION_SCRIPT,
delays=list(dell),
errors=list(errl),
state_file='simple-reaction.json',
)
def run_vm():
blocks = [
(
# ok, so my delay was 742 on average... so let's do from 20 to 1000 in steps of 20?
range(20, 1200, 20),
range(0, MAX_ERRORS + 1),
),
(
range(30, 1200, 20),
range(0, MAX_ERRORS + 1),
),
]
for dell, errl in blocks:
run.run_test(
test_url=
"http://www.quantified-mind.com/tests/feature_match/practice",
js_script_template=js_scripts.VISUAL_MATCHING_SCRIPT,
delays=list(dell),
errors=list(errl),
state_file='visual-matching.json',
)
def calculate_fitness(self):
tree = self.genome
tree_string = get_node_string(tree)
#print("Tree was turned into: ", tree_string)
#input("GO ON?!")
with open("behavior_tree_bot/tree.txt", "w") as file:
file.write(tree_string)
results = run.run_test()
#print("Results: ", results)
#self._fitness = 10 + results['wins'] - 2 * results['crashes'] - results['timed_out']
self._fitness = 10 + 0.5 * results['easy_bot'] + 2 * results[
'spread_bot'] + 2 * results['aggressive_bot'] + 0.8 * results[
'production_bot'] - 2 * results['crashes'] - results[
'timed_out'] + 0.5 * results["unique_wins"]
self._results = results
#measurements = metrics.metrics(self.to_level())
# Print out the possible measurements or look at the implementation of metrics.py for other keys:
# print(measurements.keys())
# Default fitness function: Just some arbitrary combination of a few criteria. Is it good? Who knows?
# STUDENT Modify this, and possibly add more metrics. You can replace this with whatever code you like.
#####################################################################
## TODO
## We need to think of a way to calculate fitness. The naive solution is to
## literally just run the game and check our win rate. Unfortunately this would
## take a lot of time. Keep an eye out for patterns that might give us a better
## way to analyze trees success without actually running it. In the meantime,
## this should probably just create the behavior tree from our genome and
## run the planet wars program.
##
## -Damen
#####################################################################
"""
coefficients = dict(
meaningfulJumpVariance=0.5,
negativeSpace=0.6,
pathPercentage=0.5,
emptyPercentage=0.6,
linearity=-0.5,
solvability=2.0
)
self._fitness = sum(map(lambda m: coefficients[m] * measurements[m],
coefficients))
"""
return self
# 10개 이상의 카테고리를 10개로 묶는다.
shrink_cat_vars_dict = shrink_cat_vars(train, cat_vars, shrink_size=10)
shrink_cat_train = shrink_cat(train_cat, shrink_cat_vars_dict)
shrink_cat_test = shrink_cat(test_cat, shrink_cat_vars_dict)
cat_features = [
feature for feature in shrink_cat_train.columns if 'new' in feature
]
print("engineered features: ", cat_features)
# 2-4. merge
train = pd.concat([train[features], shrink_cat_train[cat_features]],
axis=1).astype(float)
test = pd.concat([test[features], shrink_cat_test[cat_features]],
axis=1).astype(float)
print(train.shape, test.shape)
# 3. train and prediction
fold = 10
seed = 2018
models = run_train(train, train_label, params, fold=fold, seed=seed)
# 4. test
prediction = run_test(test, models, fold)
# 5. save
test_submission = pd.DataFrame({'id': test_id, 'target': prediction})
test_submission.to_csv('../porto-seguro/final.csv', index=False)
print("saved result")
def run_reg(self):
log(INFO, "Starting regression test suite.")
run.run_test()
rc.run()
self.b_update_.config( state=ACTIVE )
return 0
time.sleep(5*60)
if __name__ == "__main__":
if len(sys.argv) > 1:
infrastructure = sys.argv[1]
config_path = sys.argv[2]
scenario_path = sys.argv[3]
num = sys.argv[4]
results = []
for i in xrange(int(num)):
header = '###################\n' \
'# Running test %s #\n' \
'###################'
print header % (i+1)
results_path = run_test(sys.argv)
print "Results for this test are in %s\n" % results_path
results.append(results_path)
if infrastructure == 'aws':
reboot_aws_frontends(config_path)
summary = '#######################' \
'# SUMMARY OF TESTS #' \
'#######################'
for i, r in enumerate(results):
print "Results for test %s are in: %s" % (i, r)
time.sleep(5 * 60)
if __name__ == "__main__":
if len(sys.argv) > 1:
infrastructure = sys.argv[1]
config_path = sys.argv[2]
scenario_path = sys.argv[3]
num = sys.argv[4]
results = []
for i in xrange(int(num)):
header = '###################\n' \
'# Running test %s #\n' \
'###################'
print header % (i + 1)
results_path = run_test(sys.argv)
print "Results for this test are in %s\n" % results_path
results.append(results_path)
if infrastructure == 'aws':
reboot_aws_frontends(config_path)
summary = '#######################' \
'# SUMMARY OF TESTS #' \
'#######################'
for i, r in enumerate(results):
print "Results for test %s are in: %s" % (i, r)
def run_reg(self):
log(INFO, "Starting regression test suite.")
run.run_test()
rc.run()
self.b_update_.config( state=ACTIVE )
return 0
def train_network(self, resume=False):
csv_logger = CsvLogger(filepath=self.opt.save_model)
if (not resume):
best_test = 0
start_epoch = self.opt.start_epoch
else:
checkpoint_path = os.path.join(
self.opt.resume,
'checkpoint{}.pth.tar'.format(self.opt.local_rank))
csv_path = os.path.join(
self.opt.resume, 'results{}.csv'.format(self.opt.local_rank))
print("=> loading checkpoint '{}'".format(checkpoint_path))
checkpoint = torch.load(checkpoint_path,
map_location=self.opt.device)
start_epoch = checkpoint['epoch']
start_step = len(self.train_loader) * start_epoch
self.optim, self.mixup = self.init_optimizer_and_mixup(
checkpoint['optimizer'])
best_test = checkpoint['best_prec1']
self.model.load_state_dict(checkpoint['state_dict'])
print("=> loaded checkpoint '{}' (epoch {})".format(
checkpoint_path, checkpoint['epoch']))
for epoch in range(start_epoch, self.opt.epochs + 1):
print("-------------------------", epoch,
"--------------------------------")
train_loss, train_accuracy1, train_accuracy5, = run_train(
self.model,
self.train_loader,
self.mixup,
epoch,
self.optim,
self.criterion,
self.device,
self.opt._dtype,
self.opt.train_batch_size,
log_interval=2)
test_loss, test_accuracy1, test_accuracy5 = run_test(
self.model, self.val_loader, self.criterion, self.device,
self.opt._dtype)
self.optim.epoch_step()
self.vis.plot_curves({'train_loss': train_loss},
iters=epoch,
title='train loss',
xlabel='epoch',
ylabel='train loss')
self.vis.plot_curves({'train_acc': train_accuracy1},
iters=epoch,
title='train acc',
xlabel='epoch',
ylabel='train acc')
self.vis.plot_curves({'val_loss': test_loss},
iters=epoch,
title='val loss',
xlabel='epoch',
ylabel='val loss')
self.vis.plot_curves({'val_acc': test_accuracy1},
iters=epoch,
title='val acc',
xlabel='epoch',
ylabel='val acc')
csv_logger.write({
'epoch': epoch + 1,
'val_error1': 1 - test_accuracy1,
'val_error5': 1 - test_accuracy5,
'val_loss': test_loss,
'train_error1': 1 - train_accuracy1,
'train_error5': 1 - train_accuracy5,
'train_loss': train_loss
})
save_checkpoint(
{
'epoch': epoch + 1,
'state_dict': self.model.state_dict(),
'best_prec1': best_test,
'optimizer': self.optim.state_dict()
},
test_accuracy1 > best_test,
filepath=self.opt.save_model,
local_rank=self.opt.local_rank)
# TODO: save on the end of the cycle
mem = '%.3gG' % (torch.cuda.memory_cached() /
1E9 if torch.cuda.is_available() else 0)
print("memory gpu use : ", mem)
if test_accuracy1 > best_test:
best_test = test_accuracy1
