def main(config):
prepare_dirs_and_logger(config)
tf.compat.v1.set_random_seed(config.random_seed)
if 'nn' in config.arch:
from data_nn import BatchManager
else:
from data import BatchManager
batch_manager = BatchManager(config)
if config.is_3d:
trainer = Trainer3(config, batch_manager)
else:
trainer = Trainer(config, batch_manager)
print("---------------------------------")
print("| |")
print("| |")
print("| prepare trainer |")
print("| is done |")
print("| |")
print("| |")
print("---------------------------------")
if config.is_train:
save_config(config)
trainer.train()
else:
if not config.load_path:
raise Exception(
"[!] You should specify `load_path` to load a pretrained model"
)
trainer.test()
def test_create_client_with_alias(self):
with temppath() as tpath:
config = Config(path=tpath)
section = 'dev.alias'
config.add_section(section)
config.set(section, 'url', 'http://host:port')
save_config(config)
Config(path=tpath).get_client('dev')
def test_disable_file_logging(self):
with temppath() as tpath:
config = Config(tpath)
config.add_section('cmd.command')
config.set('cmd.command', 'log.disable', 'true')
save_config(config)
config = Config(tpath)
handler = config.get_log_handler('cmd')
ok_(not isinstance(handler, TimedRotatingFileHandler))
def drive_changelog(yesterday, html):
""" Do something """
drive = util.get_driveclient()
start_change_id = util.CONFIG.get("changestamp", "1")
html += """<p><table border="1" cellpadding="3" cellspacing="0">
<thead>
<tr><th>Changestamp</th><th>Time</th><th>Author</th><th>Resource</th></tr>
</thead>
<tbody>"""
largestChangeId = -1
hits = 0
page_token = None
while True:
param = {}
if start_change_id:
param['startChangeId'] = start_change_id
if page_token:
param['pageToken'] = page_token
print(("Requesting start_change_id: %s "
"largestChangeId: %s page_token: %s"
) % (start_change_id, largestChangeId, page_token))
response = drive.changes().list(**param).execute()
largestChangeId = response['largestChangeId']
page_token = response.get('nextPageToken')
for item in response['items']:
changestamp = item['id']
if item['deleted']:
continue
modifiedDate = datetime.datetime.strptime(
item['file']['modifiedDate'][:19], '%Y-%m-%dT%H:%M:%S')
modifiedDate = modifiedDate.replace(tzinfo=pytz.timezone("UTC"))
if modifiedDate < yesterday:
continue
uri = item['file']['alternateLink']
title = item['file']['title']
author = item['file']['lastModifyingUserName']
localts = modifiedDate.astimezone(pytz.timezone("America/Chicago"))
hits += 1
html += """
<tr><td>%s</td><td>%s</td><td>%s</td><td><a href="%s">%s</a></td></tr>
""" % (changestamp, localts.strftime("%-d %b %I:%M %P"),
author, uri, title)
if not page_token:
break
util.CONFIG['changestamp'] = changestamp
if hits == 0:
html += """<tr><td colspan="4">No Changes Found...</td></tr>\n"""
html += """</tbody></table>"""
util.save_config()
return html
def config(access_key_id, secret_access_key):
"""
配置密钥信息
ACCESS_KEY_ID 申请的API密钥ID
SECRET_ACCESS_KEY 申请的API密钥
"""
save_config(access_key_id=access_key_id, secret_access_key=secret_access_key)
click.echo("密钥ID配置成功")
def test_create_client_with_alias_and_timeout(self):
with temppath() as tpath:
config = Config(path=tpath)
section = 'dev.alias'
config.add_section(section)
config.set(section, 'url', 'http://host:port')
config.set(section, 'timeout', '1')
save_config(config)
eq_(Config(path=tpath).get_client('dev')._timeout, 1)
config.set(section, 'timeout', '1,2')
save_config(config)
eq_(Config(path=tpath).get_client('dev')._timeout, (1,2))
def convert_legacy_config(trial_dir, t_agent):
legacy_config_file = os.path.join(trial_dir, "config.pk")
config_file = os.path.join(trial_dir, "config.yaml")
if not os.path.exists(legacy_config_file):
raise ValueError("No config file found in {}".format(trial_dir))
else:
legacy_config = util.load_legacy_config(legacy_config_file)
default_config = util.load_config("default.yaml")[t_agent]
for k in default_config:
if k in legacy_config:
default_config[k] = legacy_config[k]
default_config["agent"] = t_agent
util.save_config(config_file, default_config)
def settings():
tables = models.components.keys()
form = forms.create_prefs_form()
if form.validate_on_submit():
form.populate_obj(util.AttributeWrapper(app.config))
util.save_config(app.config, CONFIG_PATH)
warning = library.check()
if warning:
flash(warning, "error")
flash("Your settings have been saved.", "success")
models.create()
return redirect(request.referrer)
return render_template('settings.html', form=form, tables=tables)
def event_reconfig():
log.debug(f"Handling 'request={data['request']}'.")
lkp.clear_template_info_cache()
if lkp.instance_role == "controller":
# Inactive all partitions to prevent further scheduling
partitions = get_partitions()
update_partitions(partitions, "INACTIVE")
# Fetch and write new config.yaml
util.cfg = util.config_from_metadata()
if not util.cfg.pubsub_topic_id:
log.info("Auto reconfigure is disabled. Aborting...")
exit(0)
util.save_config(util.cfg, util.CONFIG_FILE)
util.lkp = util.Lookup(util.cfg)
# Regenerate *.conf files
log.info("Clean install custom scripts")
setup.install_custom_scripts(clean=True)
log.info("Generating new cloud.conf for slurm.conf")
setup.gen_cloud_conf(util.lkp)
log.info("Generating new slurm.conf")
setup.install_slurm_conf(util.lkp)
log.info("Generating new slurmdbd.conf")
setup.install_slurmdbd_conf(util.lkp)
log.info("Generating new cloud_gres.conf")
setup.gen_cloud_gres_conf(util.lkp)
log.info("Generating new cgroup.conf")
setup.install_cgroup_conf()
# Send restart message to cluster topic
message_json = json.dumps({
"request":
"restart",
"timestamp":
datetime.utcnow().isoformat(),
})
publish_message(project, util.cfg.pubsub_topic_id, message_json)
elif lkp.instance_role == "compute":
log.info(f"NO-OP for 'Request={data['request']}'.")
elif lkp.instance_role == "login":
log.info(f"NO-OP for 'Request={data['request']}'.")
else:
log.error(f"Unknown node role: {lkp.instance_role}")
def main(config):
prepare_dirs_and_logger(config)
tf.set_random_seed(config.random_seed)
from data import BatchManager
batch_manager = BatchManager(config)
trainer = Trainer_tumor(config, batch_manager)
if config.is_train:
save_config(config)
trainer.train()
else:
if not config.load_path:
raise Exception(
"[!] You shou/home/tudorld specify `load_path` to load a pretrained model"
)
trainer.test()
def setup_model(self):
"""Creates a SFUN object.
Returns: keras model
"""
self.checkpoints_path = os.path.join(self._config['training']['session_dir'], 'checkpoints')
if not os.path.exists(self.checkpoints_path):
os.mkdir(self.checkpoints_path)
self.history_filename = 'history_' + self._config['training']['session_dir'][self._config['training']['session_dir'].rindex('/') + 1:] + '.csv'
self.model, inputs_mask = self.build_model(train_bn=self.train_bn)
self.compile_sfun(self.model, inputs_mask, self._config['training']['lr'])
self._config['dataset']['num_freq'] = self.num_freq
config_path = os.path.join(self._config['training']['session_dir'], 'config.json')
if os.path.exists(self.checkpoints_path) and util.dir_contains_files(self.checkpoints_path):
checkpoints = os.listdir(self.checkpoints_path)
checkpoints.sort(key=lambda x: os.stat(os.path.join(self.checkpoints_path, x)).st_mtime)
last_checkpoint = checkpoints[-1]
last_checkpoint_path = os.path.join(self.checkpoints_path, last_checkpoint)
self.epoch_num = int(last_checkpoint[11:16])
print('Loading Sound Field Network model from epoch: %d' % self.epoch_num)
self.model.load_weights(last_checkpoint_path)
else:
print('Building new Sound Field Network model...')
self.epoch_num = 0
self.model.summary()
if not os.path.exists(config_path):
util.save_config(config_path, self._config)
return self.model
import os, datetime
from flask import Flask
from flask.ext.sqlalchemy import SQLAlchemy
from session import SqliteSessionInterface
import util
CONFIG_FILE = 'altium.cfg'
app = Flask(__name__)
CONFIG_PATH = os.path.join(app.root_path, CONFIG_FILE)
app.config.from_object('altium.config')
app.config.from_pyfile(CONFIG_PATH, silent=True)
util.save_config(app.config, CONFIG_PATH)
# Server-side sessions
path = app.config['SESSION_PATH']
path = os.path.join(app.root_path, '.sessions')
if not os.path.exists(path):
os.mkdir(path)
os.chmod(path, int('700', 8))
app.session_interface = SqliteSessionInterface(path)
# Initial check of the library to establish SVN data
library = util.SVNLibrary()
#library.check()
db = SQLAlchemy(app)
import hooks
def run_training():
'''train the Neural Network'''
# sanity check
assert (FLAGS.input_data_type == 'float' or FLAGS.input_data_type == 'int')
assert (FLAGS.output_data_type == 'float'
or FLAGS.output_data_type == 'int')
# import the dataset
data_sets = dataset.Datasets(FLAGS.data_dir, FLAGS.separate_file,
FLAGS.input_data_type, FLAGS.output_data_type)
#for hotspot training
'''
data_sets = dataset.Datasets(FLAGS.data_dir,
FLAGS.separate_file,
FLAGS.input_data_type, FLAGS.output_data_type,
FLAGS.tile_size, FLAGS.num_maps)
'''
with tf.Graph().as_default():
# placeholder
input_pl, golden_pl = util.generate_placeholder(
data_sets.num_in_neuron, data_sets.num_out_neuron,
FLAGS.batch_size, FLAGS.input_data_type, FLAGS.output_data_type)
# build graph
if FLAGS.hidden1 == 0:
assert (FLAGS.hidden2 == 0)
outputs = util.layer('output_layer', input_pl,
data_sets.num_in_neuron,
data_sets.num_out_neuron, None)
else:
hidden1 = util.layer('hidden1', input_pl, data_sets.num_in_neuron,
FLAGS.hidden1, util.fast_sigmoid)
if FLAGS.hidden2 == 0:
outputs = util.layer('output_layer', hidden1, FLAGS.hidden1,
data_sets.num_out_neuron, None)
else:
hidden2 = util.layer('hidden2', hidden1, FLAGS.hidden1,
FLAGS.hidden2, util.fast_sigmoid)
outputs = util.layer('output_layer', hidden2, FLAGS.hidden2,
data_sets.num_out_neuron, None)
# loss
#loss = bm.loss(outputs, golden_pl)
loss = util.loss(outputs, golden_pl, FLAGS.benchmark)
# train
#train_op = bm.training(loss, FLAGS.learning_rate)
train_op = util.training(loss, FLAGS.learning_rate)
# accumulated error for one batch of data
error = util.error(outputs, golden_pl, FLAGS.benchmark)
# summary - not necessary
summary = tf.merge_all_summaries()
# init
init = tf.initialize_all_variables()
# sess
sess = tf.Session()
# summary writer - not necessary
summary_writer = tf.train.SummaryWriter(FLAGS.log_dir, sess.graph)
# everything built, run init
sess.run(init)
# start training
#_, max_steps = data_sets.train.max_steps(FLAGS.batch_size)
for step in xrange(FLAGS.max_steps):
feed_dict = util.fill_feed_dict(data_sets.train, input_pl,
golden_pl, FLAGS.batch_size)
sess.run(train_op, feed_dict=feed_dict)
# print the loss every 100 steps
# write the summary
# evaluate the model
if not step % 100:
print('step %d: loss = %.2f' %
(step, sess.run(loss, feed_dict=feed_dict)))
summary_str = sess.run(summary, feed_dict=feed_dict)
summary_writer.add_summary(summary_str, step)
summary_writer.flush()
'''
print('training data evaluation')
util.do_eval(sess, error,
input_pl, golden_pl,
FLAGS.batch_size, data_sets.train)
'''
print('validation data evaluation')
util.do_eval(sess, error, input_pl, golden_pl,
FLAGS.batch_size, data_sets.validate)
# final accuracy
print('test data evaluation')
util.do_eval(sess, error, input_pl, golden_pl, FLAGS.batch_size,
data_sets.test)
# filename for saving
savefile = str(data_sets.num_in_neuron) + "_" + str(
FLAGS.hidden1) + "_" + str(FLAGS.hidden2) + "_" + str(
data_sets.num_out_neuron) + ".txt"
# save weights and biases
util.save_config(sess, NUM_LAYERS, FLAGS.config_dir, savefile)
# save trained output
#util.save_output(sess, data_sets.train, outputs, FLAGS.data_dir)
#need to fetch original input data
output_save = sess.run(outputs,
feed_dict={input_pl: data_sets.input_data})
np.savetxt(FLAGS.data_dir + "train_result/" + savefile,
output_save,
delimiter=" ")
for index, playlist in enumerate(playlists):
print(str(index) + " | " + playlist)
elif (arg == 'remote'):
playlists = user.get_playlist_names()
for index, playlist in enumerate(playlists):
print(str(index) + " | " + playlist)
else:
print('Not yet implemented.')
print('Show all playlists, local and remote')
elif (cmd == 'select' and arg != None):
ids = user.get_playlist_id(arg)
config["current_playlist"]["uid"] = ids["uid"]
config["current_playlist"]["pid"] = ids["pid"]
config["current_playlist"]["name"] = user.get_playlist_name(arg)
util.save_config(config)
print('Set current playlist to ' + config["current_playlist"]["name"])
elif (cmd == 'clone' and arg != None):
ids = user.get_playlist_id(arg)
config["current_playlist"]["uid"] = ids["uid"]
config["current_playlist"]["pid"] = ids["pid"]
config["current_playlist"]["name"] = user.get_playlist_name(arg)
util.save_config(config)
try:
user.init_git_playlist(ids["uid"], ids["pid"])
print('Cloned playlist ' + config["current_playlist"]["name"])
except:
print(config["current_playlist"]["name"] + ' already cloned.')
def init(self, settings, settingsFile):
for k in self._baseTypes.keys():
self._codeTables[k] = self._baseTypes[k].init(settings)
util.save_config(settings, settingsFile)
def run(self):
done = False
while not done:
self.screen.blit(Options.sky, self.screen.get_rect())
self.water.update()
self.water_sprite.draw(self.screen)
for i in xrange(len(self.menu)):
self.render(i)
cloud.update()
cloud.draw(self.screen)
rect = Options.logo.get_rect()
rect.centerx = self.screen.get_rect().centerx
rect.top = 0
self.screen.blit(Options.logo, rect)
image = util.smallfont.render("http://funnyboat.sourceforge.net/",
True, (0, 0, 0))
rect = image.get_rect()
rect.midbottom = self.screen.get_rect().midbottom
self.screen.blit(image, rect)
pygame.display.flip()
self.t += 1
nextframe = False
while not nextframe:
pygame.event.post(pygame.event.wait())
for event in pygame.event.get():
if event.type == QUIT or \
event.type == KEYDOWN and event.key == K_ESCAPE:
self.selection = -1
done = True
nextframe = True
elif event.type == NEXTFRAME:
nextframe = True
elif event.type == JOYAXISMOTION:
if event.axis == 1:
if event.value < -0.5:
self.move_up()
if event.value > 0.5:
self.move_down()
elif event.axis == 0:
if event.value < -0.5:
self.change_left()
if event.value > 0.5:
self.change_right()
elif event.type == JOYBUTTONDOWN:
if self.gamepad:
if self.gamepad.is_pressed('a', event):
self.change_right()
elif self.gamepad.is_pressed('b', event):
done = True
elif event.button == 0:
#done = True
self.change_right()
elif event.button == 1:
done = True
elif event.type == KEYDOWN:
if event.key == K_UP:
self.move_up()
elif event.key == K_DOWN:
self.move_down()
elif event.key == K_LEFT:
self.change_left()
elif event.key == K_RIGHT:
self.change_right()
elif self.selection == Options.NAME:
if event.key == K_BACKSPACE:
if len(Variables.name) != 0:
Variables.name = Variables.name[:-1]
elif event.key == K_SPACE or event.unicode != " " and event.unicode >= u' ':
if len(Variables.name) < 32:
Variables.name += event.unicode
self.refresh()
elif event.key == K_SPACE or event.key == K_RETURN:
#done = True
self.change_right()
util.save_config()
return self.selection
def init(self, settings, settingsFile):
for k in self._baseTypes.keys():
self._codeTables[k] = self._baseTypes[k].init(settings)
util.save_config(settings, settingsFile)
def train(config, trial_dir=None, visualize=False, overwrite_config=False):
t_agent = config["agent"]
assert t_agent in SUPPORTED_AGENTS, "Agent type {} not supported".format(
t_agent)
# prepare trial environment
pid = os.getpid()
trial_name = "{}_pid{}".format(t_agent, pid)
logger, log_dir = prepare_for_logging(trial_name)
# create agent
if "max_obstacles" not in config:
config["max_obstacles"] = 3
env = NIPS(visualize, max_obstacles=config["max_obstacles"])
logger.info("pid={}, env={}".format(pid, id(env)))
# to train from scratch or fine tune
fine_tuning = False
if trial_dir is not None:
config_file = os.path.join(trial_dir, "config.yaml")
if not os.path.exists(config_file):
convert_legacy_config(trial_dir, t_agent)
existing_config = util.load_config(config_file)
fine_tuning = True
if overwrite_config:
logger.info("Overwrite config from file {}".format(trial_dir))
for k, v in config.iteritems():
existing_config[k] = v
config = existing_config
config["model_dir"] = trial_dir
# save config to the trial folder
util.print_settings(logger, config, env)
config_file = os.path.join(log_dir, "config.yaml")
util.save_config(config_file, config)
# instantiate an agent
config["logger"] = logger
config["log_dir"] = log_dir
if t_agent == "DDPG":
from ddpg import DDPG
agent = DDPG(env, config)
elif t_agent == "TRPO":
from trpo import TRPO
agent = TRPO(env, config)
else:
# because of the assertion above, this should never happen
raise ValueError("Unsupported agent type: {}".format(t_agent))
# learn
if fine_tuning:
util.print_sec_header(logger, "Continual training")
agent.set_state(config)
else:
util.print_sec_header(logger, "Training from scratch")
reward_hist, steps_hist = agent.learn(
total_episodes=config["total_episodes"])
env.close()
# send result
img_file = os.path.join(log_dir, "train_stats.png")
util.plot_stats(reward_hist, steps_hist, img_file)
log_file = os.path.join(log_dir, "train.log")
util.send_email(log_dir, [img_file], [log_file], config)
logger.info("Finished (pid={}).".format(pid))
def drive_changelog(regime, yesterday, html):
""" Do something """
drive = util.get_driveclient()
start_change_id = util.CONFIG.get("changestamp_"+regime, "1")
html += """<p><table border="1" cellpadding="3" cellspacing="0">
<thead>
<tr><th>Changestamp</th><th>Time</th><th>Author</th><th>Resource</th></tr>
</thead>
<tbody>"""
largestChangeId = -1
hits = 0
page_token = None
while True:
param = {}
if start_change_id:
param['startChangeId'] = start_change_id
if page_token:
param['pageToken'] = page_token
print(("Requesting start_change_id: %s "
"largestChangeId: %s page_token: %s"
) % (start_change_id, largestChangeId, page_token))
response = drive.changes().list(**param).execute()
largestChangeId = response['largestChangeId']
page_token = response.get('nextPageToken')
for item in response['items']:
changestamp = item['id']
if item['deleted']:
continue
# don't do more work when this file actually did not change
modifiedDate = datetime.datetime.strptime(
item['file']['modifiedDate'][:19], '%Y-%m-%dT%H:%M:%S')
modifiedDate = modifiedDate.replace(tzinfo=pytz.timezone("UTC"))
if modifiedDate < yesterday:
continue
# Need to see which base folder this file is in!
parentid = get_base_folder_id(regime, drive, item['fileId'])
if parentid != util.CONFIG[regime]['basefolder']:
print(('Skipping %s as it is other project'
) % (repr(item['file']['title']), ))
continue
uri = item['file']['alternateLink']
title = item['file']['title'].encode('ascii', 'ignore')
author = item['file']['lastModifyingUserName']
localts = modifiedDate.astimezone(pytz.timezone("America/Chicago"))
hits += 1
html += """
<tr><td>%s</td><td>%s</td><td>%s</td><td><a href="%s">%s</a></td></tr>
""" % (changestamp, localts.strftime("%-d %b %I:%M %P"),
author, uri, title)
if not page_token:
break
util.CONFIG['changestamp_'+regime] = changestamp
if hits == 0:
html += """<tr><td colspan="4">No Changes Found...</td></tr>\n"""
html += """</tbody></table>"""
util.save_config()
return html
def main(config):
logger = util.create_logger(name='train_log', log_dir=config.log_dir)
if not os.path.exists(config.log_dir):
os.makedirs(config.log_dir, exist_ok=True)
util.save_config(config.log_dir, config.config)
logger.info('Logs and models will be save in {}.'.format(config.log_dir))
rnd = np.random.RandomState(seed=config.seed)
solution = util.create_solution(device='cpu:0')
num_params = solution.get_num_params()
if config.load_model is not None:
solution.load(config.load_model)
print('Loaded model from {}'.format(config.load_model))
init_params = solution.get_params()
else:
init_params = None
solver = cma.CMAEvolutionStrategy(
x0=np.zeros(num_params) if init_params is None else init_params,
sigma0=config.init_sigma,
inopts={
'popsize': config.population_size,
'seed': config.seed if config.seed > 0 else 42,
'randn': np.random.randn,
},
)
best_so_far = -float('Inf')
ii32 = np.iinfo(np.int32)
repeats = [config.reps] * config.population_size
device_type = 'cpu' if args.num_gpus <= 0 else 'cuda'
num_devices = mp.cpu_count() if args.num_gpus <= 0 else args.num_gpus
with mp.get_context('spawn').Pool(
initializer=worker_init,
initargs=(args.config, device_type, num_devices),
processes=config.num_workers,
) as pool:
for n_iter in range(config.max_iter):
params_set = solver.ask()
task_seeds = [rnd.randint(0, ii32.max)] * config.population_size
fitnesses = []
ss = 0
while ss < config.population_size:
ee = ss + min(config.num_workers, config.population_size - ss)
fitnesses.append(
pool.map(func=get_fitness,
iterable=zip(params_set[ss:ee], task_seeds[ss:ee],
repeats[ss:ee])))
ss = ee
fitnesses = np.concatenate(fitnesses)
if isinstance(solver, cma.CMAEvolutionStrategy):
# CMA minimizes.
solver.tell(params_set, -fitnesses)
else:
solver.tell(fitnesses)
logger.info(
'Iter={0}, '
'max={1:.2f}, avg={2:.2f}, min={3:.2f}, std={4:.2f}'.format(
n_iter, np.max(fitnesses), np.mean(fitnesses),
np.min(fitnesses), np.std(fitnesses)))
best_fitness = max(fitnesses)
if best_fitness > best_so_far:
best_so_far = best_fitness
model_path = os.path.join(config.log_dir, 'best.npz')
save_params(solver=solver,
solution=solution,
model_path=model_path)
logger.info(
'Best model updated, score={}'.format(best_fitness))
if (n_iter + 1) % config.save_interval == 0:
model_path = os.path.join(config.log_dir,
'iter_{}.npz'.format(n_iter + 1))
save_params(solver=solver,
solution=solution,
model_path=model_path)
def run(self):
done = False
while not done:
self.screen.blit(Options.sky, self.screen.get_rect())
self.water.update()
self.water_sprite.draw(self.screen)
for i in xrange(len(self.menu)):
self.render(i)
cloud.update()
cloud.draw(self.screen)
rect = Options.logo.get_rect()
rect.centerx = self.screen.get_rect().centerx
rect.top = 0
self.screen.blit(Options.logo, rect)
image = util.smallfont.render("http://funnyboat.sourceforge.net/", True, (0,0,0))
rect = image.get_rect()
rect.midbottom = self.screen.get_rect().midbottom
self.screen.blit(image, rect)
pygame.display.flip()
self.t += 1
nextframe = False
while not nextframe:
pygame.event.post(pygame.event.wait())
for event in pygame.event.get():
if event.type == QUIT or \
event.type == KEYDOWN and event.key == K_ESCAPE:
self.selection = -1
done = True
nextframe = True
elif event.type == NEXTFRAME:
nextframe = True
elif event.type == JOYAXISMOTION:
if event.axis == 1:
if event.value < -0.5:
self.move_up()
if event.value > 0.5:
self.move_down()
elif event.axis == 0:
if event.value < -0.5:
self.change_left()
if event.value > 0.5:
self.change_right()
elif event.type == JOYBUTTONDOWN:
if event.button == 0:
#done = True
self.change_right()
elif event.button == 1:
done = True
elif event.type == KEYDOWN:
if event.key == K_UP:
self.move_up()
elif event.key == K_DOWN:
self.move_down()
elif event.key == K_LEFT:
self.change_left()
elif event.key == K_RIGHT:
self.change_right()
elif self.selection == Options.NAME:
if event.key == K_BACKSPACE:
if len(Variables.name) != 0:
Variables.name = Variables.name[:-1]
elif event.key == K_SPACE or event.unicode != " " and event.unicode>=u' ':
if len(Variables.name) < 32:
Variables.name += event.unicode
self.refresh()
elif event.key == K_SPACE or event.key == K_RETURN:
#done = True
self.change_right()
util.save_config()
return self.selection
def save(self):
util.save_config("telegram_bot.json", self.config)
def drive_changelog(regime, yesterday, html):
""" Do something """
drive = util.get_driveclient()
folders = util.get_folders(drive)
start_change_id = util.CONFIG.get("changestamp_" + regime, "1")
html += """<p><table border="1" cellpadding="3" cellspacing="0">
<thead>
<tr><th>Folder</th><th>Resource</th></tr>
</thead>
<tbody>"""
largestChangeId = -1
hits = 0
page_token = None
param = {"includeDeleted": False, "maxResults": 1000}
while True:
if start_change_id:
param["startChangeId"] = start_change_id
if page_token:
param["pageToken"] = page_token
print(
("[%s] start_change_id: %s largestChangeId: %s page_token: %s")
% (regime, start_change_id, largestChangeId, page_token)
)
response = drive.changes().list(**param).execute()
largestChangeId = response["largestChangeId"]
page_token = response.get("nextPageToken")
for item in response["items"]:
changestamp = item["id"]
if item["deleted"]:
continue
# don't do more work when this file actually did not change
modifiedDate = datetime.datetime.strptime(item["file"]["modifiedDate"][:19], "%Y-%m-%dT%H:%M:%S")
modifiedDate = modifiedDate.replace(tzinfo=pytz.timezone("UTC"))
if modifiedDate < yesterday:
continue
# Need to see which base folder this file is in!
isproject = False
for parent in item["file"]["parents"]:
if parent["id"] not in folders:
print(("[%s] file: %s has unknown parent: %s") % (regime, item["id"], parent["id"]))
continue
if folders[parent["id"]]["basefolder"] == util.CONFIG[regime]["basefolder"]:
isproject = True
if not isproject:
print(("[%s] %s skipped") % (regime, repr(item["file"]["title"])))
continue
uri = item["file"]["alternateLink"]
title = item["file"]["title"].encode("ascii", "ignore")
localts = modifiedDate.astimezone(pytz.timezone("America/Chicago"))
hits += 1
pfolder = item["file"]["parents"][0]["id"]
html += """
<tr>
<td><a href="https://docs.google.com/folderview?id=%s&usp=drivesdk">%s</a></td>
<td><a href="%s">%s</a></td></tr>
""" % (
pfolder,
folders[pfolder]["title"],
uri,
title,
)
hit = False
if "version" in item["file"] and item["file"]["mimeType"] != FMIME:
lastmsg = ""
try:
revisions = drive.revisions().list(fileId=item["file"]["id"]).execute()
except:
print(("[%s] file %s (%s) failed revisions") % (regime, title, item["file"]["mimeType"]))
revisions = {"items": []}
for item2 in revisions["items"]:
md = datetime.datetime.strptime(item2["modifiedDate"][:19], "%Y-%m-%dT%H:%M:%S")
md = md.replace(tzinfo=pytz.timezone("UTC"))
if md < yesterday:
continue
localts = md.astimezone(pytz.timezone("America/Chicago"))
if "lastModifyingUser" not in item2:
print(("[%s] file: %s has no User? %s") % (regime, title, item2))
continue
luser = item2["lastModifyingUser"]
hit = True
thismsg = """
<tr><td colspan="2"><img src="%s" style="height:25px;"/> %s by
%s (%s)</td></tr>
""" % (
(luser["picture"]["url"] if "picture" in luser else ""),
localts.strftime("%-d %b %-I:%M %p"),
luser["displayName"],
luser["emailAddress"],
)
if thismsg != lastmsg:
html += thismsg
lastmsg = thismsg
# Now we check revisions
if not hit:
luser = item["file"]["lastModifyingUser"]
html += """
<tr><td colspan="2"><img src="%s" style="height:25px;"/> %s by
%s (%s)</td></tr>
""" % (
luser["picture"]["url"] if "picture" in luser else "",
localts.strftime("%-d %b %-I:%M %p"),
luser["displayName"],
luser.get("emailAddress", "n/a"),
)
if not page_token:
break
util.CONFIG["changestamp_" + regime] = changestamp
if hits == 0:
html += """<tr><td colspan="5">No Changes Found...</td></tr>\n"""
html += """</tbody></table>"""
util.save_config()
return html
def train(config_path):
""" Trains a model for a maximum of config.max_epochs epochs
Args:
config_path: string, path to a config.json file
"""
# Load configuration
if not os.path.exists(config_path):
print 'Error: No configuration file present at specified path.'
return
config = util.load_config(config_path)
print 'Loaded configuration from: %s' % config_path
# Create session directory
if 'session_dir' not in config['training'] or os.path.exists(config['training']['session_dir']): create_new_session(config)
# Direct all output to screen and log file
util.set_print_to_screen_and_file(
os.path.join(config['training']['session_dir'], 'session.log'))
model = fcpn.FCPN(config)
dataset = data.Dataset(config)
dataset.prepare(config['dataset']['refresh_cache'])
config['model']['pointnet']['num'] = np.prod(model.get_feature_volume_shape(
config['dataset']['training_samples']['spatial_size'], config['model']['pointnet']['spacing'], 1))
enqueue_op, queue_placeholders, queue_batch_placeholders, get_queue_size_op = setup_queue(
config['dataset']['training_samples']['num_points'] + config['model']['pointnet']['num'], dataset.get_num_output_voxels(), config['training']['batch_size'])
tf_config = tf.ConfigProto()
tf_config.gpu_options.allow_growth = config['training']['gpu']['allow_growth']
tf_config.allow_soft_placement = config['training']['gpu']['allow_soft_placement']
sess = tf.Session(config=tf_config)
with sess.as_default():
with tf.device('/gpu:' + str(config['training']['gpu']['id'])):
# Batch normalization
batch_i = tf.Variable(0, name='batch_i')
batch_normalization_decay = util.get_batch_normalization_decay(
batch_i, config['training']['batch_size'], config['training']['optimizer']['batch_normalization']['initial_decay'], config['training']['optimizer']['batch_normalization']['decay_rate'], config['training']['optimizer']['batch_normalization']['decay_step'])
tf.summary.scalar('batch_normalization_decay',
batch_normalization_decay)
is_training_pl = tf.placeholder(tf.bool, shape=())
# Build model
pred_op = model.build_model(config['training']['batch_size'], config['dataset']['training_samples']['spatial_size'], queue_batch_placeholders['input_points_pl'],
queue_batch_placeholders['input_features_pl'], is_training_pl, dataset.get_num_learnable_classes(), batch_normalization_decay)
# Loss
loss_op = model.get_loss(
pred_op, queue_batch_placeholders['output_voxels_pl'], queue_batch_placeholders['output_voxel_weights_pl'])
model.print_num_parameters()
model.print_layer_weights()
# Confusion matrix
confusion_matrix_op, confusion_matrix_update_op, confusion_matrix_clear_op = model.get_confusion_matrix_ops(
pred_op, queue_batch_placeholders['output_voxels_pl'], dataset.get_num_learnable_classes(), dataset.get_empty_class())
# Optimizer
learning_rate_op = util.get_learning_rate(
batch_i, config['training']['batch_size'], config['training']['optimizer']['learning_rate']['initial'], config['training']['optimizer']['learning_rate']['decay_rate'], config['training']['optimizer']['learning_rate']['decay_step'])
tf.summary.scalar('learning_rate', learning_rate_op)
optimizer_op = tf.train.AdamOptimizer(learning_rate_op)
if config['training']['train_upsampling_only']:
upsampling_weights = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, "upsampling")
optimization_op = optimizer_op.minimize(loss_op, var_list=upsampling_weights, global_step=batch_i)
else:
optimization_op = optimizer_op.minimize(loss_op, global_step=batch_i)
# Summary and Saving
saver = tf.train.Saver(max_to_keep=config['training']['checkpoints_to_keep'])
merged_summary_op = tf.summary.merge_all()
summary_writers = {
'train': tf.summary.FileWriter(os.path.join(config['training']['session_dir'], 'train'), sess.graph),
'val': tf.summary.FileWriter(os.path.join(config['training']['session_dir'], 'val'))
}
# Initialize variables in graph
init_g = tf.global_variables_initializer()
init_l = tf.local_variables_initializer()
sess.run([init_g, init_l], {is_training_pl: True})
# Restore model weights from disk
if config['training']['checkpoint_path']:
weights_to_be_restored = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES)
# If finetuning on a new dataset, don't load last layer weights or confusion matrix
if config['training']['finetune_new_classes']:
final_layer_weights = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, scope="upsampling/15cm_to_5cm/final_conv")
confusion_variables = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, scope="confusion")
weights_to_be_restored = list(set(weights_to_be_restored) - set(final_layer_weights) - set(confusion_variables))
restorer = tf.train.Saver(var_list=weights_to_be_restored)
restorer.restore(sess, config['training']['checkpoint_path'])
print 'Model weights restored from checkpoint file: %s' % config['training']['checkpoint_path']
num_batches = {
'train': dataset.get_num_batches('train', config['training']['batch_size']),
'val': dataset.get_num_batches('val', config['training']['batch_size'])
}
ops = {
'train': [loss_op, merged_summary_op, optimization_op],
'val': [loss_op, merged_summary_op, confusion_matrix_update_op]
}
# Start loading samples into FIFO queue
coord, loader_thread = start_data_loader(
sess, enqueue_op, queue_placeholders, model, dataset, config)
# Save configuration file (with derived parameters) to session directory
util.save_config(os.path.join(config['training']['session_dir'], 'config.json'), config)
# Start training
sample_i = 0
for epoch_i in range(config['training']['max_epochs']):
print '\nEpoch: %d' % epoch_i
for s in ['train', 'val']:
is_training = (s == 'train')
if s == 'train':
is_training = True
print 'Training set\nBatch/Total Batches | Loss | Items in Queue'
else:
print 'Validation set\nBatch/Total Batches | Loss | Items in Queue'
for epoch_batch_i in range(num_batches[s]):
loss, summary, _ = sess.run(
ops[s], feed_dict={is_training_pl: is_training})
# Log statistics
if epoch_batch_i % config['training']['log_every_n_batches'] == 0:
summary_writers[s].add_summary(summary, sample_i)
summary_writers[s].flush()
print '%i/%i | %f | %d' % (epoch_batch_i + 1, num_batches[s], loss, get_queue_size_op.eval())
# Only do when in training phase
if s == 'train':
sample_i += config['training']['batch_size']
# Save snapshot of model
if epoch_batch_i % config['training']['save_every_n_batches'] == 0:
save_path = saver.save(sess, os.path.join(
config['training']['session_dir'], "model.ckpt"), global_step=epoch_i)
print 'Checkpoint saved at batch %d to %s' % (
epoch_batch_i, save_path)
# Only do at the end of the validation phase
if s == 'train':
save_path = saver.save(sess, os.path.join(
config['training']['session_dir'], "model.ckpt"), global_step=epoch_i)
print 'Checkpoint saved at batch %d to %s' % (epoch_batch_i, save_path)
elif s == 'val':
confusion_matrix = confusion_matrix_op.eval()
# Compute and print per-class statistics
true_positives, false_negatives, false_positives, ious = util.compute_per_class_statistics(confusion_matrix[:dataset.get_empty_class(),:dataset.get_empty_class()])
util.pretty_print_confusion_matrix(confusion_matrix, dataset.get_learnable_classes_strings())
util.pretty_print_per_class_statistics(dataset.get_learnable_classes_strings()[:dataset.get_empty_class()], true_positives, false_negatives, false_positives, ious)
avg_iou = np.mean(ious)
summary = tf.Summary()
summary.value.add(
tag='avg_iou', simple_value=avg_iou)
# Add per-class IoUs to summary to be viewable in Tensorboard
for class_i, class_label in enumerate(dataset.get_learnable_classes_strings()[:dataset.get_empty_class()]):
summary.value.add(
tag=class_label + '_iou', simple_value=ious[class_i])
summary_writers[s].add_summary(summary, sample_i)
summary_writers[s].flush()
confusion_matrix_clear_op.eval()
coord.request_stop()
coord.join([loader_thread])
print 'Training complete.'