def _setup_monitor(self):
"""
Set up monitor to model the objective value, learning rate,
momentum (if applicable), and extra channels defined by
the cost.
This method must be called after `learning_rule.get_updates`,
since it may have an effect on `learning_rule.add_channels_to_monitor`
(that is currently the case for `learning_rule.RMSProp`).
"""
if bool(self.monitoring_dataset):
if self.monitoring_batch_size is None and self.monitoring_batches is None:
self.monitoring_batch_size = self.batch_size
self.monitoring_batches = self.batches_per_iter
self.monitor.setup(
dataset=self.monitoring_dataset,
cost=self.cost,
batch_size=self.monitoring_batch_size,
num_batches=self.monitoring_batches,
extra_costs=self.monitoring_costs,
mode=self.monitor_iteration_mode,
)
dataset_name = first_key(self.monitoring_dataset)
monitoring_dataset = self.monitoring_dataset[dataset_name]
# TODO: have Monitor support non-data-dependent channels
self.monitor.add_channel(
name="learning_rate",
ipt=None,
val=self.learning_rate,
data_specs=(NullSpace(), ""),
dataset=monitoring_dataset,
)
if self.learning_rule:
self.learning_rule.add_channels_to_monitor(self.monitor, monitoring_dataset)
def _setup_monitor(self):
"""
Set up monitor to model the objective value, learning rate,
momentum (if applicable), and extra channels defined by
the cost.
This method must be called after `learning_rule.get_updates`,
since it may have an effect on `learning_rule.add_channels_to_monitor`
(that is currently the case for `learning_rule.RMSProp`).
"""
if bool(self.monitoring_dataset):
if (self.monitoring_batch_size is None
and self.monitoring_batches is None):
self.monitoring_batch_size = self.batch_size
self.monitoring_batches = self.batches_per_iter
self.monitor.setup(dataset=self.monitoring_dataset,
cost=self.cost,
batch_size=self.monitoring_batch_size,
num_batches=self.monitoring_batches,
extra_costs=self.monitoring_costs,
mode=self.monitor_iteration_mode)
dataset_name = first_key(self.monitoring_dataset)
monitoring_dataset = self.monitoring_dataset[dataset_name]
# TODO: have Monitor support non-data-dependent channels
self.monitor.add_channel(name='learning_rate',
ipt=None,
val=self.learning_rate,
data_specs=(NullSpace(), ''),
dataset=monitoring_dataset)
if self.learning_rule:
self.learning_rule.add_channels_to_monitor(
self.monitor, monitoring_dataset)
def test_unpickle_key():
fd, fname = tempfile.mkstemp()
with os.fdopen(fd, 'wb') as f:
d = ('a', 1)
cPickle.dump(d, f)
loaded = load("{!pkl: '%s': 50}" % fname)
assert_(first_key(loaded) == d)
assert_(first_value(loaded) == 50)
os.remove(fname)
def test_unpickle_key():
fd, fname = tempfile.mkstemp()
with os.fdopen(fd, 'wb') as f:
d = ('a', 1)
cPickle.dump(d, f)
loaded = load("{!pkl: '%s': 50}" % fname)
assert_(first_key(loaded) == d)
assert_(first_value(loaded) == 50)
os.remove(fname)
def summarize(path):
"""
Summarize the model
Parameters
----------
path : str
The path to the pickled model to summarize
"""
model = serial.load(path)
for param in model.get_params():
name = param.name
if name is None:
name = '<anon>'
v = param.get_value()
print(name + ': ' + str((v.min(), v.mean(), v.max())), end='')
print(str(v.shape))
if np.sign(v.min()) != np.sign(v.max()):
v = np.abs(v)
print('abs(' + name + '): ' + str((v.min(), v.mean(), v.max())))
if v.ndim == 2:
row_norms = np.sqrt(np.square(v).sum(axis=1))
print(name + " row norms:", end='')
print((row_norms.min(), row_norms.mean(), row_norms.max()))
col_norms = np.sqrt(np.square(v).sum(axis=0))
print(name + " col norms:", end='')
print((col_norms.min(), col_norms.mean(), col_norms.max()))
if hasattr(model, 'monitor'):
print('trained on', model.monitor.get_examples_seen(), 'examples')
print('which corresponds to ', end='')
print(model.monitor.get_batches_seen(), 'batches')
key = first_key(model.monitor.channels)
hour = float(model.monitor.channels[key].time_record[-1]) / 3600.
print('Trained for {0} hours'.format(hour))
try:
print(model.monitor.get_epochs_seen(), 'epochs')
except Exception:
pass
if hasattr(model.monitor, 'training_succeeded'):
if model.monitor.training_succeeded:
print('Training succeeded')
else:
print('Training was not yet completed ' +
'at the time of this save.')
else:
print('This pickle file is damaged, or was made before the ' +
'Monitor tracked whether training completed.')
def summarize(path):
"""
Summarize the model
Parameters
----------
path : str
The path to the pickled model to summarize
"""
model = serial.load(path)
for param in model.get_params():
name = param.name
if name is None:
name = '<anon>'
v = param.get_value()
print(name + ': ' + str((v.min(), v.mean(), v.max())), end='')
print(str(v.shape))
if np.sign(v.min()) != np.sign(v.max()):
v = np.abs(v)
print('abs(' + name + '): ' + str((v.min(), v.mean(), v.max())))
if v.ndim == 2:
row_norms = np.sqrt(np.square(v).sum(axis=1))
print(name + " row norms:", end='')
print((row_norms.min(), row_norms.mean(), row_norms.max()))
col_norms = np.sqrt(np.square(v).sum(axis=0))
print(name + " col norms:", end='')
print((col_norms.min(), col_norms.mean(), col_norms.max()))
if hasattr(model, 'monitor'):
print('trained on', model.monitor.get_examples_seen(), 'examples')
print('which corresponds to ', end='')
print(model.monitor.get_batches_seen(), 'batches')
key = first_key(model.monitor.channels)
hour = float(model.monitor.channels[key].time_record[-1]) / 3600.
print('Trained for {0} hours'.format(hour))
try:
print(model.monitor.get_epochs_seen(), 'epochs')
except Exception:
pass
if hasattr(model.monitor, 'training_succeeded'):
if model.monitor.training_succeeded:
print('Training succeeded')
else:
print('Training was not yet completed ' +
'at the time of this save.')
else:
print('This pickle file is damaged, or was made before the ' +
'Monitor tracked whether training completed.')
