def __init__(self, lower, upper):
"""
Initialise the Sampler class.
.. note::
Currently only supports rectangular type restrictions on the
parameter space
Parameters
----------
lower : array_like
Lower or minimum bounds for the parameter space
upper : array_like
Upper or maximum bounds for the parameter space
"""
self.lower = np.array(lower)
self.upper = np.array(upper)
self.dims = self.upper.shape[0]
assert (self.lower.ndim == 1) and (self.upper.ndim == 1)
assert self.lower.shape[0] == self.dims
self.X = ArrayBuffer()
self.y = ArrayBuffer()
self.virtual_flag = ArrayBuffer()
self.pending_results = {}
self.n_tasks = None
def __init__(self, lower, upper):
"""
Initialise the Sampler class.
.. note::
Currently only supports rectangular type restrictions on the
parameter space
Parameters
----------
lower : array_like
Lower or minimum bounds for the parameter space
upper : array_like
Upper or maximum bounds for the parameter space
"""
self.lower = np.array(lower)
self.upper = np.array(upper)
self.dims = self.upper.shape[0]
assert (self.lower.ndim == 1) and (self.upper.ndim == 1)
assert self.lower.shape[0] == self.dims
self.X = ArrayBuffer()
self.y = ArrayBuffer()
self.virtual_flag = ArrayBuffer()
self.pending_results = {}
self.n_tasks = None
def main():
d = 20
n_stack = 10000
buf = ArrayBuffer()
st = time.time()
for i in range(n_stack):
# buf.append(np.random.random())
buf.append(np.random.random(d)) # NOQA we dont do anything with a
ft = time.time()
print('Efficient buffer took {0:.5f} seconds'.format(ft - st))
a = buf()
import IPython
IPython.embed()
import sys
sys.exit()
print(a)
exit()
st = time.time()
b = np.zeros((0, d))
for i in range(n_stack):
b = np.vstack((b, np.random.random(d)))
ft = time.time()
print('Repeated Vstack took {0:.5f} seconds'.format(ft - st))
st = time.time()
b_buf = []
for i in range(n_stack):
b_buf.append(np.random.random(d))
b = np.array(b_buf)
ft = time.time()
print('List buffering and casting took {0:.5f} seconds'.format(ft - st))
def _update(self, uid, y_true):
"""
Update a job with its observed value.
Parameters
----------
uid : str
A hexadecimal ID that identifies the job to be updated
y_true : float
The observed value corresponding to the job identified by 'uid'
Returns
-------
int
Index location in the data lists 'Sampler.X' and
'Sampler.y' corresponding to the job being updated
"""
# Make sure the job uid given is valid
if uid not in self.pending_results:
warnings.warn('Result was not pending!')
assert uid in self.pending_results
# Kill the job and update collected data with true observation
ind = self.pending_results.pop(uid)
# If the user has been pushing Nones until now, we will init properly
if self.n_tasks is None:
self.n_tasks = len(np.atleast_1d(y_true))
pending_count = len(self.y)
self.y = ArrayBuffer()
for _ in range(pending_count):
self.y.append(np.zeros(self.n_tasks))
self.y()[ind] = y_true
self.virtual_flag()[ind] = False
return ind
def _update(self, uid, y_true):
"""
Update a job with its observed value.
Parameters
----------
uid : str
A hexadecimal ID that identifies the job to be updated
y_true : float
The observed value corresponding to the job identified by 'uid'
Returns
-------
int
Index location in the data lists 'Sampler.X' and
'Sampler.y' corresponding to the job being updated
"""
# Make sure the job uid given is valid
if uid not in self.pending_results:
warnings.warn('Result was not pending!')
assert uid in self.pending_results
# Kill the job and update collected data with true observation
ind = self.pending_results.pop(uid)
# If the user has been pushing Nones until now, we will init properly
if self.n_tasks is None:
self.n_tasks = len(np.atleast_1d(y_true))
pending_count = len(self.y)
self.y = ArrayBuffer()
for _ in range(pending_count):
self.y.append(np.zeros(self.n_tasks))
self.y()[ind] = y_true
self.virtual_flag()[ind] = False
return ind
class Sampler:
"""
Provide a basic template and interface to specific Sampler subclasses.
Attributes
----------
lower : numpy.ndarray
Lower bounds for each parameter in the parameter space
upper : numpy.ndarray
Upper bounds for each parameter in the parameter space
dims : int
Dimension of the parameter space (number of parameters)
X : ArrayBuffer
Contiguous Buffer of feature vectors representing observed locations
in the parameter space
y : ArrayBuffer
Contiguous Buffer of target outputs or expected (virtual) target
outputs corresponding to the feature vectors 'X'
virtual_flag : ArrayBuffer
A contiguous array of boolean flags indicating virtual elements of 'y'
True: Corresponding target output is virtual
False: Corresponding target output is observed
pending_results : dict
A dictionary that maps the job ID to the corresponding index in both
the 'X' and 'y' buffers.
"""
def __init__(self, lower, upper):
"""
Initialise the Sampler class.
.. note::
Currently only supports rectangular type restrictions on the
parameter space
Parameters
----------
lower : array_like
Lower or minimum bounds for the parameter space
upper : array_like
Upper or maximum bounds for the parameter space
"""
self.lower = np.array(lower)
self.upper = np.array(upper)
self.dims = self.upper.shape[0]
assert (self.lower.ndim == 1) and (self.upper.ndim == 1)
assert self.lower.shape[0] == self.dims
self.X = ArrayBuffer()
self.y = ArrayBuffer()
self.virtual_flag = ArrayBuffer()
self.pending_results = {}
self.n_tasks = None
def pick(self):
"""
Pick the next feature location for the next observation to be taken.
.. note::
Currently only supports rectangular type restrictions on the
parameter space
Returns
-------
numpy.ndarray
Location in the parameter space for the next observation to be
taken
str
A random hexadecimal ID to identify the corresponding job
Raises
------
AssertionError
Under all circumstances. See note above.
"""
assert False
def update(self, uid, y_true):
"""
Update a job with its observed value.
.. note::
Currently a dummy function whose functionality will be
filled by subclasses of the Sampler class
Parameters
----------
uid : str
A hexadecimal ID that identifies the job to be updated
y_true : float
The observed value corresponding to the job identified by 'uid'
Returns
-------
int
Index location in the data lists 'Sampler.X' and
'Sampler.y' corresponding to the job being updated
Raises
------
AssertionError
Under all circumstances. See note above.
"""
assert False
def _assign(self, xq, yq_exp):
"""
Assign a pair (location in parameter space, virtual target) a job ID.
Parameters
----------
xq : numpy.ndarray
Location in the parameter space for the next observation to be
taken
yq_exp : float
The virtual target output at that parameter location
Returns
-------
str
A random hexadecimal ID to identify the corresponding job
"""
# Place a virtual observation onto the collected data
n = len(self.X)
self.X.append(xq)
self.virtual_flag.append(True)
# If we get a None, insert zeros instead
if yq_exp is None and self.n_tasks is not None:
self.y.append(np.zeros(self.n_tasks))
else:
self.y.append(yq_exp) # then add the real one
# Create an uid for this observation
# m = hashlib.md5()
# m.update(np.array(np.random.random()))
# uid = m.hexdigest()
uid = uuid.uuid4().hex # "%032x" % random.getrandbits(128)
# Note the index of corresponding to this picked location
self.pending_results[uid] = n
return uid
def _update(self, uid, y_true):
"""
Update a job with its observed value.
Parameters
----------
uid : str
A hexadecimal ID that identifies the job to be updated
y_true : float
The observed value corresponding to the job identified by 'uid'
Returns
-------
int
Index location in the data lists 'Sampler.X' and
'Sampler.y' corresponding to the job being updated
"""
# Make sure the job uid given is valid
if uid not in self.pending_results:
warnings.warn('Result was not pending!')
assert uid in self.pending_results
# Kill the job and update collected data with true observation
ind = self.pending_results.pop(uid)
# If the user has been pushing Nones until now, we will init properly
if self.n_tasks is None:
self.n_tasks = len(np.atleast_1d(y_true))
pending_count = len(self.y)
self.y = ArrayBuffer()
for _ in range(pending_count):
self.y.append(np.zeros(self.n_tasks))
self.y()[ind] = y_true
self.virtual_flag()[ind] = False
return ind
class Sampler:
"""
Provide a basic template and interface to specific Sampler subclasses.
Attributes
----------
lower : numpy.ndarray
Lower bounds for each parameter in the parameter space
upper : numpy.ndarray
Upper bounds for each parameter in the parameter space
dims : int
Dimension of the parameter space (number of parameters)
X : ArrayBuffer
Contiguous Buffer of feature vectors representing observed locations
in the parameter space
y : ArrayBuffer
Contiguous Buffer of target outputs or expected (virtual) target
outputs corresponding to the feature vectors 'X'
virtual_flag : ArrayBuffer
A contiguous array of boolean flags indicating virtual elements of 'y'
True: Corresponding target output is virtual
False: Corresponding target output is observed
pending_results : dict
A dictionary that maps the job ID to the corresponding index in both
the 'X' and 'y' buffers.
"""
def __init__(self, lower, upper):
"""
Initialise the Sampler class.
.. note::
Currently only supports rectangular type restrictions on the
parameter space
Parameters
----------
lower : array_like
Lower or minimum bounds for the parameter space
upper : array_like
Upper or maximum bounds for the parameter space
"""
self.lower = np.array(lower)
self.upper = np.array(upper)
self.dims = self.upper.shape[0]
assert (self.lower.ndim == 1) and (self.upper.ndim == 1)
assert self.lower.shape[0] == self.dims
self.X = ArrayBuffer()
self.y = ArrayBuffer()
self.virtual_flag = ArrayBuffer()
self.pending_results = {}
self.n_tasks = None
def pick(self):
"""
Pick the next feature location for the next observation to be taken.
.. note::
Currently only supports rectangular type restrictions on the
parameter space
Returns
-------
numpy.ndarray
Location in the parameter space for the next observation to be
taken
str
A random hexadecimal ID to identify the corresponding job
Raises
------
AssertionError
Under all circumstances. See note above.
"""
assert False
def update(self, uid, y_true):
"""
Update a job with its observed value.
.. note::
Currently a dummy function whose functionality will be
filled by subclasses of the Sampler class
Parameters
----------
uid : str
A hexadecimal ID that identifies the job to be updated
y_true : float
The observed value corresponding to the job identified by 'uid'
Returns
-------
int
Index location in the data lists 'Sampler.X' and
'Sampler.y' corresponding to the job being updated
Raises
------
AssertionError
Under all circumstances. See note above.
"""
assert False
def _assign(self, xq, yq_exp):
"""
Assign a pair (location in parameter space, virtual target) a job ID.
Parameters
----------
xq : numpy.ndarray
Location in the parameter space for the next observation to be
taken
yq_exp : float
The virtual target output at that parameter location
Returns
-------
str
A random hexadecimal ID to identify the corresponding job
"""
# Place a virtual observation onto the collected data
n = len(self.X)
self.X.append(xq)
self.virtual_flag.append(True)
# If we get a None, insert zeros instead
if yq_exp is None and self.n_tasks is not None:
self.y.append(np.zeros(self.n_tasks))
else:
self.y.append(yq_exp) # then add the real one
# Create an uid for this observation
# m = hashlib.md5()
# m.update(np.array(np.random.random()))
# uid = m.hexdigest()
uid = uuid.uuid4().hex # "%032x" % random.getrandbits(128)
# Note the index of corresponding to this picked location
self.pending_results[uid] = n
return uid
def _update(self, uid, y_true):
"""
Update a job with its observed value.
Parameters
----------
uid : str
A hexadecimal ID that identifies the job to be updated
y_true : float
The observed value corresponding to the job identified by 'uid'
Returns
-------
int
Index location in the data lists 'Sampler.X' and
'Sampler.y' corresponding to the job being updated
"""
# Make sure the job uid given is valid
if uid not in self.pending_results:
warnings.warn('Result was not pending!')
assert uid in self.pending_results
# Kill the job and update collected data with true observation
ind = self.pending_results.pop(uid)
# If the user has been pushing Nones until now, we will init properly
if self.n_tasks is None:
self.n_tasks = len(np.atleast_1d(y_true))
pending_count = len(self.y)
self.y = ArrayBuffer()
for _ in range(pending_count):
self.y.append(np.zeros(self.n_tasks))
self.y()[ind] = y_true
self.virtual_flag()[ind] = False
return ind
