def do_ping(service):
# get service data (url and param_name)
url,param_name = services[service]
# Prepare input date
sitemap_url = model.get_config().url + 'sitemap.xml'
form_fields = {param_name: sitemap_url}
form_data = urllib.urlencode(form_fields)
# invoke the url fetch
result = urlfetch.fetch(url,payload=form_data,follow_redirects=True)
# return status code
return result.status_code
def twit(key): config = model.get_config() post = model.get_post_by_key(key) if not post.as_draft: auth = tweepy.OAuthHandler(config.consumer_key, config.consumer_secret) auth.set_access_token(config.access_key, config.access_secret) api = tweepy.API(auth) post_url = config.url + post.slug msg = post.title + " " + post_url api.update_status(msg) return msg else: return "draft!"
def main(args=sys.argv[1:]):
parser = argparse.ArgumentParser(description='Studio WebUI server. \
Usage: studio \
<arguments>')
parser.add_argument('--config', help='configuration file', default=None)
# parser.add_argument('--guest',
# help='Guest mode (does not require db credentials)',
# action='store_true')
parser.add_argument('--port',
help='port to run Flask server on',
type=int,
default=5000)
parser.add_argument('--host', help='host name.', default='0.0.0.0')
parser.add_argument('--verbose',
'-v',
help='Verbosity level. Allowed vaules: ' +
'debug, info, warn, error, crit ' +
'or numerical value of logger levels.',
default=None)
args = parser.parse_args(args)
config = model.get_config()
if args.config:
with open(args.config) as f:
config = yaml.load(f)
if args.verbose:
config['verbose'] = args.verbose
# if args.guest:
# config['database']['guest'] = True
global _config
global _db_provider
_config = config
_db_provider = model.get_db_provider(_config, blocking_auth=False)
getlogger().setLevel(model.parse_verbosity(config.get('verbose')))
global _save_auth_cookie
_save_auth_cookie = True
print('Starting Studio UI on port {0}'.format(args.port))
app.run(host=args.host, port=args.port)
def __init__(self, args):
self.config = model.get_config()
if args.config:
if isinstance(args.config, basestring):
with open(args.config) as f:
self.config.update(yaml.load(f))
else:
self.config.update(args.config)
if args.guest:
self.config['database']['guest'] = True
self.db = model.get_db_provider(self.config)
self.logger = logging.getLogger('LocalExecutor')
self.logger.setLevel(model.parse_verbosity(self.config.get('verbose')))
self.logger.debug("Config: ")
self.logger.debug(self.config)
def main(args=sys.argv[1:]):
logger = logs.getLogger('studio-runner')
parser = argparse.ArgumentParser(
description='Studio runner. \
Usage: studio run <runner_arguments> \
script <script_arguments>')
parser.add_argument('--config', help='configuration file', default=None)
parser.add_argument('--project', help='name of the project', default=None)
parser.add_argument(
'--experiment', '-e',
help='Name of the experiment. If none provided, ' +
'random uuid will be generated',
default=None)
parser.add_argument(
'--guest',
help='Guest mode (does not require db credentials)',
action='store_true')
parser.add_argument(
'--force-git',
help='If run in a git directory, force running the experiment ' +
'even if changes are not commited',
action='store_true')
parser.add_argument(
'--gpus',
help='Number of gpus needed to run the experiment',
type=int,
default=None)
parser.add_argument(
'--cpus',
help='Number of cpus needed to run the experiment' +
' (used to configure cloud instance)',
type=int,
default=None)
parser.add_argument(
'--ram',
help='Amount of RAM needed to run the experiment' +
' (used to configure cloud instance), ex: 10G, 10GB',
default=None)
parser.add_argument(
'--gpuMem',
help='Amount of GPU RAM needed to run the experiment',
default=None)
parser.add_argument(
'--hdd',
help='Amount of hard drive space needed to run the experiment' +
' (used to configure cloud instance), ex: 10G, 10GB',
default=None)
parser.add_argument(
'--queue', '-q',
help='Name of the remote execution queue',
default=None)
parser.add_argument(
'--cloud',
help='Cloud execution mode. Could be gcloud, gcspot, ec2 or ec2spot',
default=None)
parser.add_argument(
'--bid',
help='Spot instance price bid, specified in USD or in percentage ' +
'of on-demand instance price. Default is %(default)s',
default='100%')
parser.add_argument(
'--capture-once', '-co',
help='Name of the immutable artifact to be captured. ' +
'It will be captured once before the experiment is run',
default=[], action='append')
parser.add_argument(
'--capture', '-c',
help='Name of the mutable artifact to be captured continuously',
default=[], action='append')
parser.add_argument(
'--reuse', '-r',
help='Name of the artifact from another experiment to use',
default=[], action='append')
parser.add_argument(
'--verbose', '-v',
help='Verbosity level. Allowed values: ' +
'debug, info, warn, error, crit ' +
'or numerical value of logger levels.',
default=None)
parser.add_argument(
'--metric',
help='Metric to show in the summary of the experiment, ' +
'and to base hyperparameter search on. ' +
'Refers a scalar value in tensorboard log ' +
'example: --metric=val_loss[:final | :min | :max] to report ' +
'validation loss in the end of the keras experiment ' +
'(or smallest or largest throughout the experiment for :min ' +
'and :max respectively)',
default=None)
parser.add_argument(
'--hyperparam', '-hp',
help='Try out multiple values of a certain parameter. ' +
'For example, --hyperparam=learning_rate:0.01:0.1:l10 ' +
'will instantiate 10 versions of the script, replace ' +
'learning_rate with a one of the 10 values for learning ' +
'rate that lies on a log grid from 0.01 to 0.1, create '
'experiments and place them in the queue.',
default=[], action='append')
parser.add_argument(
'--num-workers',
help='Number of local or cloud workers to spin up',
type=int,
default=None)
parser.add_argument(
'--python-pkg',
help='Python package not present in the current environment ' +
'that is needed for experiment. Only compatible with ' +
'remote and cloud workers for now',
default=[], action='append')
parser.add_argument(
'--ssh-keypair',
help='Name of the SSH keypair used to access the EC2 ' +
'instances directly',
default=None)
parser.add_argument(
'--optimizer', '-opt',
help='Name of optimizer to use, by default is grid search. ' +
'The name of the optimizer must either be in ' +
'studio/optimizer_plugins ' +
'directory or the path to the optimizer source file ' +
'must be supplied. ',
default='grid')
parser.add_argument(
'--cloud-timeout',
help="Time (in seconds) that cloud workers wait for messages. " +
"If negative, " +
"wait for the first message in the queue indefinitely " +
"and shut down " +
"as soon as no new messages are available. " +
"If zero, don't wait at all." +
"Default value is %(default)d",
type=int,
default=300)
parser.add_argument(
'--user-startup-script',
help='Path of script to run immediately ' +
'before running the remote worker',
default=None)
parser.add_argument(
'--branch',
help='Branch of studioml to use when running remote worker, useful ' +
'for debugging pull requests. Default is current',
default=None)
parser.add_argument(
'--max-duration',
help='Max experiment runtime (i.e. time after which experiment ' +
'should be killed no matter what.). Examples of values ' +
'might include 5h, 48h2m10s',
default=None)
parser.add_argument(
'--lifetime',
help='Max experiment lifetime (i.e. wait time after which ' +
'experiment loses relevance and should not be started)' +
' Examples include 240h30m10s',
default=None)
parser.add_argument(
'--container',
help='Singularity container in which experiment should be run. ' +
'Assumes that container has all dependencies installed',
default=None
)
parser.add_argument(
'--port',
help='Ports to open on a cloud instance',
default=[], action='append'
)
# detect which argument is the script filename
# and attribute all arguments past that index as related to the script
(runner_args, other_args) = parser.parse_known_args(args)
py_suffix_args = [i for i, arg in enumerate(args) if arg.endswith('.py')
or '::' in arg]
rerun = False
if len(py_suffix_args) < 1:
print('None of the arugments end with .py')
if len(other_args) == 0:
print("Trying to run a container job")
assert runner_args.container is not None
exec_filename = None
elif len(other_args) == 1:
print("Treating last argument as experiment key to rerun")
rerun = True
experiment_key = args[-1]
else:
print("Too many extra arguments - should be either none " +
"for container job or one for experiment re-run")
sys.exit(1)
else:
script_index = py_suffix_args[0]
exec_filename, other_args = args[script_index], args[script_index + 1:]
runner_args = parser.parse_args(args[:script_index])
# TODO: Queue the job based on arguments and only then execute.
config = model.get_config(runner_args.config)
if runner_args.verbose:
config['verbose'] = runner_args.verbose
if runner_args.guest:
config['database']['guest'] = True
if runner_args.container:
runner_args.capture_once.append(
runner_args.container + ':_singularity')
verbose = model.parse_verbosity(config['verbose'])
logger.setLevel(verbose)
if git_util.is_git() and not git_util.is_clean() and not rerun:
logger.warn('Running from dirty git repo')
if not runner_args.force_git:
logger.error(
'Specify --force-git to run experiment from dirty git repo')
sys.exit(1)
resources_needed = parse_hardware(runner_args, config['resources_needed'])
logger.debug('resources requested: ')
logger.debug(str(resources_needed))
artifacts = {}
artifacts.update(parse_artifacts(runner_args.capture, mutable=True))
artifacts.update(parse_artifacts(runner_args.capture_once, mutable=False))
with model.get_db_provider(config) as db:
artifacts.update(parse_external_artifacts(runner_args.reuse, db))
if runner_args.branch:
config['cloud']['branch'] = runner_args.branch
if runner_args.user_startup_script:
config['cloud']['user_startup_script'] = \
runner_args.user_startup_script
if runner_args.lifetime:
config['experimentLifetime'] = runner_args.lifetime
if any(runner_args.hyperparam):
if runner_args.optimizer is "grid":
experiments = add_hyperparam_experiments(
exec_filename,
other_args,
runner_args,
artifacts,
resources_needed,
logger)
queue_name = submit_experiments(
experiments,
config=config,
logger=logger,
queue_name=runner_args.queue,
cloud=runner_args.cloud)
spin_up_workers(
runner_args,
config,
resources_needed,
queue_name=queue_name,
verbose=verbose)
else:
opt_modulepath = os.path.join(
os.path.dirname(os.path.abspath(__file__)),
"optimizer_plugins",
runner_args.optimizer + ".py")
if not os.path.exists(opt_modulepath):
opt_modulepath = os.path.abspath(
os.path.expanduser(runner_args.optimizer))
logger.info('optimizer path: %s' % opt_modulepath)
assert os.path.exists(opt_modulepath)
sys.path.append(os.path.dirname(opt_modulepath))
opt_module = importlib.import_module(
os.path.basename(opt_modulepath.replace(".py", '')))
h = HyperparameterParser(runner_args, logger)
hyperparams = h.parse()
optimizer = getattr(
opt_module,
"Optimizer")(
hyperparams,
config['optimizer'],
logger)
workers_started = False
queue_name = runner_args.queue
while not optimizer.stop():
hyperparam_pop = optimizer.ask()
hyperparam_tuples = h.convert_to_tuples(hyperparam_pop)
experiments = add_hyperparam_experiments(
exec_filename,
other_args,
runner_args,
artifacts,
resources_needed,
logger,
optimizer=optimizer,
hyperparam_tuples=hyperparam_tuples)
queue_name = submit_experiments(
experiments,
config=config,
logger=logger,
cloud=runner_args.cloud,
queue_name=queue_name)
if not workers_started:
spin_up_workers(
runner_args,
config,
resources_needed,
queue_name=queue_name,
verbose=verbose)
workers_started = True
fitnesses, behaviors = get_experiment_fitnesses(
experiments, optimizer, config, logger)
# for i, hh in enumerate(hyperparam_pop):
# print fitnesses[i]
# for hhh in hh:
# print hhh
try:
optimizer.tell(hyperparam_pop, fitnesses, behaviors)
except BaseException:
optimizer.tell(hyperparam_pop, fitnesses)
try:
optimizer.disp()
except BaseException:
logger.warn('Optimizer has no disp() method')
else:
if rerun:
with model.get_db_provider(config) as db:
experiment = db.get_experiment(experiment_key)
new_key = runner_args.experiment if runner_args.experiment \
else experiment_key + '_rerun' + str(uuid.uuid4())
experiment.key = new_key
for _, art in six.iteritems(experiment.artifacts):
art['mutable'] = False
experiments = [experiment]
else:
experiments = [create_experiment(
filename=exec_filename,
args=other_args,
experiment_name=runner_args.experiment,
project=runner_args.project,
artifacts=artifacts,
resources_needed=resources_needed,
metric=runner_args.metric,
max_duration=runner_args.max_duration,
)]
queue_name = submit_experiments(
experiments,
config=config,
logger=logger,
cloud=runner_args.cloud,
queue_name=runner_args.queue)
spin_up_workers(
runner_args,
config,
resources_needed,
queue_name=queue_name,
verbose=verbose)
return
def main(args=sys.argv):
logger = logging.getLogger('studio-runner')
parser = argparse.ArgumentParser(
description='Studio runner. \
Usage: studio run <runner_arguments> \
script <script_arguments>')
parser.add_argument('--config', help='configuration file', default=None)
parser.add_argument('--project', help='name of the project', default=None)
parser.add_argument(
'--experiment', '-e',
help='Name of the experiment. If none provided, ' +
'random uuid will be generated',
default=None)
parser.add_argument(
'--guest',
help='Guest mode (does not require db credentials)',
action='store_true')
parser.add_argument(
'--force-git',
help='If run in a git directory, force running the experiment ' +
'even if changes are not commited',
action='store_true')
parser.add_argument(
'--gpus',
help='Number of gpus needed to run the experiment',
default=None)
parser.add_argument(
'--cpus',
help='Number of cpus needed to run the experiment' +
' (used to configure cloud instance)',
default=None)
parser.add_argument(
'--ram',
help='Amount of RAM needed to run the experiment' +
' (used to configure cloud instance)',
default=None)
parser.add_argument(
'--hdd',
help='Amount of hard drive space needed to run the experiment' +
' (used to configure cloud instance)',
default=None)
parser.add_argument(
'--queue', '-q',
help='Name of the remote execution queue',
default=None)
parser.add_argument(
'--cloud',
help='Cloud execution mode. Could be gcloud, ec2 or ec2spot',
default=None)
parser.add_argument(
'--bid',
help='Spot instance price bid, specified in USD or in percentage ' +
'of on-demand instance price. Default is %(default)s',
default='100%')
parser.add_argument(
'--capture-once', '-co',
help='Name of the immutable artifact to be captured. ' +
'It will be captured once before the experiment is run',
default=[], action='append')
parser.add_argument(
'--capture', '-c',
help='Name of the mutable artifact to be captured continuously',
default=[], action='append')
parser.add_argument(
'--reuse', '-r',
help='Name of the artifact from another experiment to use',
default=[], action='append')
parser.add_argument(
'--verbose', '-v',
help='Verbosity level. Allowed values: ' +
'debug, info, warn, error, crit ' +
'or numerical value of logger levels.',
default=None)
parser.add_argument(
'--metric', '-m',
help='Metric to show in the summary of the experiment, ' +
'and to base hyperparameter search on. ' +
'Refers a scalar value in tensorboard log ' +
'example: --metric=val_loss[:final | :min | :max] to report ' +
'validation loss in the end of the keras experiment ' +
'(or smallest or largest throughout the experiment for :min ' +
'and :max respectively)',
default=None)
parser.add_argument(
'--hyperparam', '-hp',
help='Try out multiple values of a certain parameter. ' +
'For example, --hyperparam=learning_rate:0.01:0.1:l10 ' +
'will instantiate 10 versions of the script, replace ' +
'learning_rate with a one of the 10 values for learning ' +
'rate that lies on a log grid from 0.01 to 0.1, create '
'experiments and place them in the queue.',
default=[], action='append')
parser.add_argument(
'--num-workers',
help='Number of local or cloud workers to spin up',
default=None)
parser.add_argument(
'--python-pkg',
help='Python package not present in the current environment ' +
'that is needed for experiment. Only compatible with ' +
'remote and cloud workers for now',
default=[], action='append')
parser.add_argument(
'--ssh-keypair',
help='Name of the SSH keypair used to access the EC2 ' +
'instances directly',
default=None)
parser.add_argument(
'--optimizer', '-opt',
help='Name of optimizer to use, by default is grid search. ' +
'The name of the optimizer must either be in ' +
'studio/optimizer_plugins ' +
'directory or the path to the optimizer source file ' +
'must be supplied. ',
default='grid')
parser.add_argument(
'--cloud-timeout',
help="Time (in seconds) that cloud workers wait for messages. " +
"If negative, " +
"wait for the first message in the queue indefinitely " +
"and shut down " +
"as soon as no new messages are available. " +
"If zero, don't wait at all." +
"Default value is %(default)",
type=int,
default=300)
# detect which argument is the script filename
# and attribute all arguments past that index as related to the script
py_suffix_args = [i for i, arg in enumerate(args) if arg.endswith('.py')]
if len(py_suffix_args) < 1:
print('At least one argument should be a python script ' +
'(end with *.py)')
parser.print_help()
exit()
script_index = py_suffix_args[0]
runner_args = parser.parse_args(args[1:script_index])
exec_filename, other_args = args[script_index], args[script_index + 1:]
# TODO: Queue the job based on arguments and only then execute.
config = model.get_config(runner_args.config)
if runner_args.verbose:
config['verbose'] = runner_args.verbose
verbose = model.parse_verbosity(config['verbose'])
logger.setLevel(verbose)
db = model.get_db_provider(config)
if git_util.is_git() and not git_util.is_clean():
logger.warn('Running from dirty git repo')
if not runner_args.force_git:
logger.error(
'Specify --force-git to run experiment from dirty git repo')
sys.exit(1)
resources_needed = parse_hardware(runner_args, config['cloud'])
logger.debug('resources requested: ')
logger.debug(str(resources_needed))
artifacts = {}
artifacts.update(parse_artifacts(runner_args.capture, mutable=True))
artifacts.update(parse_artifacts(runner_args.capture_once, mutable=False))
artifacts.update(parse_external_artifacts(runner_args.reuse, db))
if any(runner_args.hyperparam):
if runner_args.optimizer is "grid":
experiments = add_hyperparam_experiments(
exec_filename,
other_args,
runner_args,
artifacts,
resources_needed)
submit_experiments(
experiments,
config,
runner_args,
logger,
resources_needed)
else:
opt_modulepath = os.path.join(
os.path.dirname(os.path.abspath(__file__)),
"optimizer_plugins",
runner_args.optimizer + ".py")
# logger.info('optimizer path: %s' % opt_modulepath)
if not os.path.exists(opt_modulepath):
opt_modulepath = os.path.abspath(
os.path.expanduser(runner_args.optimizer))
logger.info('optimizer path: %s' % opt_modulepath)
assert os.path.exists(opt_modulepath)
sys.path.append(os.path.dirname(opt_modulepath))
opt_module = importlib.import_module(
os.path.basename(opt_modulepath.replace(".py", '')))
hyperparam_values, log_scale_dict = get_hyperparam_values(
runner_args)
optimizer = getattr(opt_module, "Optimizer")(hyperparam_values,
log_scale_dict)
while not optimizer.stop():
hyperparam_tuples = optimizer.ask()
experiments = add_hyperparam_experiments(
exec_filename,
other_args,
runner_args,
artifacts,
resources_needed,
optimizer=optimizer,
hyperparam_tuples=hyperparam_tuples)
submit_experiments(
experiments,
config,
runner_args,
logger,
resources_needed)
fitnesses = get_experiment_fitnesses(experiments,
optimizer, config, logger)
optimizer.tell(hyperparam_tuples, fitnesses)
# if config['verbose'] == "info" or config['verbose'] ==
# "debug":
try:
optimizer.disp()
except BaseException:
logger.warn('Optimizer has no disp() method')
else:
experiments = [model.create_experiment(
filename=exec_filename,
args=other_args,
experiment_name=runner_args.experiment,
project=runner_args.project,
artifacts=artifacts,
resources_needed=resources_needed,
metric=runner_args.metric)]
submit_experiments(
experiments,
config,
runner_args,
logger,
resources_needed)
db = None
return
def render(template_name, **kwargs): user = users.get_current_user() config = model.get_config() return render_template(template_name, config=config, user=user, **kwargs)
import traceback
import six
import google.oauth2.id_token
import google.auth.transport.requests
from experiment import experiment_from_dict
import logs
app = Flask(__name__)
DB_PROVIDER_EXPIRATION = 1800
_db_provider_timestamp = None
_db_provider = None
_config = model.get_config()
_tensorboard_dirs = {}
_grequest = google.auth.transport.requests.Request()
_save_auth_cookie = False
logger = None
@app.route('/')
def dashboard():
return _render('dashboard.html')
@app.route('/projects')
def projects():