def __init__(self,
file_writer: LogdirWriter,
model: Model,
minibatch_size: Optional[int] = 100,
display_progress: Optional[bool] = True) -> None:
"""
:param model: Model tensor
:param file_writer: Event file writer object.
:param minibatch_size: Number of points per minibatch
:param display_progress: if True the task displays the progress of calculating LML.
"""
super().__init__(file_writer, model)
self._minibatch_size = minibatch_size
self._full_lml = tf.placeholder(settings.float_type, shape=())
self._summary = tf.summary.scalar(model.name + '/full_lml',
self._full_lml)
self.wrapper = None # type: Callable[[Iterator], Iterator]
if display_progress: # pragma: no cover
try:
import tqdm
self.wrapper = tqdm.tqdm
except ImportError:
logger = settings.logger()
if logger.isEnabledFor(logging.WARNING):
logger.warning(
"LML monitor task: to display progress install `tqdm`."
)
if self.wrapper is None:
self.wrapper = lambda x: x
def __init__(self, file_writer: LogdirWriter, model: Model, minibatch_size: Optional[int] = 100,
display_progress: Optional[bool] = True) -> None:
"""
:param model: Model tensor
:param file_writer: Event file writer object.
:param minibatch_size: Number of points per minibatch
:param display_progress: if True the task displays the progress of calculating LML.
"""
super().__init__(file_writer, model)
self._minibatch_size = minibatch_size
self._full_lml = tf.placeholder(settings.tf_float, shape=())
self._summary = tf.summary.scalar(model.name + '/full_lml', self._full_lml)
self.wrapper = None # type: Callable[[Iterator], Iterator]
if display_progress: # pragma: no cover
try:
import tqdm
self.wrapper = tqdm.tqdm
except ImportError:
logger = settings.logger()
if logger.isEnabledFor(logging.WARNING):
logger.warning("LML monitor task: to display progress install `tqdm`.")
if self.wrapper is None:
self.wrapper = lambda x: x
def restore_session(session, checkpoint_dir):
"""
Restores Tensorflow session from the latest checkpoint.
:param session: The TF session
:param checkpoint_dir: checkpoint files directory.
"""
checkpoint_path = tf.train.latest_checkpoint(checkpoint_dir)
logger = settings.logger()
if logger.isEnabledFor(logging.INFO):
logger.info("Restoring session from `%s`.", checkpoint_path)
saver = tf.train.Saver(max_to_keep=1)
saver.restore(session, checkpoint_path)
def restore_session(session: tf.Session, checkpoint_dir: str,
saver: Optional[tf.train.Saver] = None) -> None:
"""
Restores Tensorflow session from the latest checkpoint.
:param session: The TF session
:param checkpoint_dir: checkpoint files directory.
:param saver: The saver object, if not provided a default saver object will be created.
"""
checkpoint_path = tf.train.latest_checkpoint(checkpoint_dir)
logger = settings.logger()
if logger.isEnabledFor(logging.INFO):
logger.info("Restoring session from `%s`.", checkpoint_path)
saver = saver or get_default_saver()
saver.restore(session, checkpoint_path)
def restore_session(session: tf.Session, checkpoint_dir: str,
saver: Optional[tf.train.Saver] = None) -> None:
"""
Restores Tensorflow session from the latest checkpoint.
:param session: The TF session
:param checkpoint_dir: checkpoint files directory.
:param saver: The saver object, if not provided a default saver object will be created.
"""
checkpoint_path = tf.train.latest_checkpoint(checkpoint_dir)
logger = settings.logger()
if logger.isEnabledFor(logging.INFO):
logger.info("Restoring session from `%s`.", checkpoint_path)
saver = saver or get_default_saver()
saver.restore(session, checkpoint_path)
import tensorflow as tf
from gpflow import settings
logger = settings.logger()
def conditional(Kmn, Kmm, Knn, f, *, full_cov=False, q_sqrt=None, white=False):
"""
Given a g1 and g2, and distribution p and q such that
p(g2) = N(g2;0,Kmm)
p(g1) = N(g1;0,Knn)
p(g1|g2) = N(g1;0,Knm)
And
q(g2) = N(g2;f,q_sqrt*q_sqrt^T)
This method computes the mean and (co)variance of
q(g1) = \int q(g2) p(g1|g2)
:param Kmn: P x M x N
:param Kmm: M x M
:param Knn: P x N x N or P x N
:param f: M x R
:param full_cov: bool
:param q_sqrt: R x M x M (lower triangular)
:param white: bool
:return: N x R or R x N x N
"""
logger.debug("base conditional")
# compute kernel stuff
num_func = tf.shape(f)[1] # R
Lm = tf.cholesky(Kmm)
def solve_A(MN_Kmn):