def test(self, override=False):
"""
Applies randomised VD to the dataset. This function does NOT write results to the hdf5 file. Call compute() to
write to the file. Handles complex, compound datasets such that the V matrix is of the same data-type as the
input matrix.
Parameters
----------
override : bool, optional. default = False
Set to true to recompute results if prior results are available. Else, returns existing results
Returns
-------
U : :class:`numpy.ndarray`
Abundance matrix
S : :class:`numpy.ndarray`
variance vector
V : :class:`numpy.ndarray`
eigenvector matrix
"""
'''
Check if a number of compnents has been set and ensure that the number is less than
the minimum axis length of the data. If both conditions are met, use fsvd. If not
use the regular svd.
C.Smith -- We might need to put a lower limit on num_comps in the future. I don't
know enough about svd to be sure.
'''
if not override:
if isinstance(self.duplicate_h5_groups, list) and len(self.duplicate_h5_groups) > 0:
self.h5_results_grp = self.duplicate_h5_groups[-1]
print('Returning previously computed results from: {}'.format(self.h5_results_grp.name))
print('set the "override" flag to True to recompute results')
return reshape_to_n_dims(self.h5_results_grp['U'])[0], self.h5_results_grp['S'][()], \
reshape_to_n_dims(self.h5_results_grp['V'])[0]
self.h5_results_grp = None
t1 = time.time()
self.__u, self.__s, self.__v = randomized_svd(self.data_transform_func(self.h5_main), self.num_components,
n_iter=3)
self.__v = stack_real_to_target_dtype(self.__v, self.h5_main.dtype)
print('Took {} to compute randomized SVD'.format(format_time(time.time() - t1)))
u_mat, success = reshape_to_n_dims(self.__u, h5_pos=self.h5_main.h5_pos_inds,
h5_spec=np.expand_dims(np.arange(self.__u.shape[1]), axis=0))
if not success:
raise ValueError('Could not reshape U to N-Dimensional dataset! Error:' + success)
v_mat, success = reshape_to_n_dims(self.__v, h5_pos=np.expand_dims(np.arange(self.__u.shape[1]), axis=1),
h5_spec=self.h5_main.h5_spec_inds)
if not success:
raise ValueError('Could not reshape V to N-Dimensional dataset! Error:' + success)
return u_mat, self.__s, v_mat
def format_time(time_in_seconds, decimals=2):
"""
Formats the provided time in seconds to seconds, minutes, or hours
Parameters
----------
time_in_seconds : number
Time in seconds
decimals : uint, optional. default = 2
Number of decimal places to which the time needs to be formatted
Returns
-------
str
String with time formatted correctly
"""
warn(
'pyUSID.io.io_utils.format_time has been moved to '
'sidpy.base.string_utils.format_time. This copy in pyUSID will'
'be removed in future release. Please update your import statements',
FutureWarning)
return sut.format_time(time_in_seconds, decimals=decimals)
def compute(self, override=False, *args, **kwargs):
"""
Creates placeholders for the results, applies the :meth:`~pyUSID.processing.process.Process._unit_computation`
to chunks of the dataset
Parameters
----------
override : bool, optional. default = False
By default, compute will simply return duplicate results to avoid recomputing or resume computation on a
group with partial results. Set to True to force fresh computation.
args : list
arguments to the mapped function in the correct order
kwargs : dict
keyword arguments to the mapped function
Returns
-------
h5_results_grp : :class:`h5py.Group`
Group containing all the results
"""
class SimpleFIFO(object):
"""
Simple class that maintains a moving average of some numbers.
"""
def __init__(self, length=5):
"""
Create a SimpleFIFO object
Parameters
----------
length : unsigned integer
Number of values that need to be maintained for the moving average
"""
self.__queue = list()
if not isinstance(length, int):
raise TypeError('length must be a positive integer')
if length <= 0:
raise ValueError('length must be a positive integer')
self.__max_length = length
self.__count = 0
def put(self, item):
"""
Adds the item to the internal queue. If the size of the queue exceeds its capacity, the oldest
item is removed.
Parameters
----------
item : float or int
Any real valued number
"""
if (not isinstance(item, Number)) or isinstance(item, complex):
raise TypeError(
'Provided item: {} is not a Number'.format(item))
self.__queue.append(item)
self.__count += 1
if len(self.__queue) > self.__max_length:
_ = self.__queue.pop(0)
def get_mean(self):
"""
Returns the average of the elements within the queue
Returns
-------
avg : number.Number
Mean of all elements within the queue
"""
return np.mean(self.__queue)
def get_cycles(self):
"""
Returns the number of items that have been added to the queue in total
Returns
-------
count : int
number of items that have been added to the queue in total
"""
return self.__count
if not override:
if len(self.duplicate_h5_groups) > 0:
if self.mpi_rank == 0:
print('Returned previously computed results at ' +
self.duplicate_h5_groups[-1].name)
self.h5_results_grp = self.duplicate_h5_groups[-1]
return self.duplicate_h5_groups[-1]
elif len(self.partial_h5_groups
) > 0 and self.h5_results_grp is None:
if self.mpi_rank == 0:
print('Resuming computation in group: ' +
self.partial_h5_groups[-1].name)
self.use_partial_computation()
resuming = False
if self.h5_results_grp is None:
# starting fresh
if self.verbose and self.mpi_rank == 0:
print('Creating HDF5 group and datasets to hold results')
self._create_results_datasets()
self._write_source_dset_provenance()
else:
# resuming from previous checkpoint
resuming = True
self._get_existing_datasets()
self.__create_compute_status_dataset()
if resuming and self.mpi_rank == 0:
percent_complete = int(
100 * len(np.where(self._h5_status_dset[()] == 1)[0]) /
self._h5_status_dset.shape[0])
print('Resuming computation. {}% completed already'.format(
percent_complete))
self.__assign_job_indices()
# Not sure if this is necessary but I don't think it would hurt either
if self.mpi_comm is not None:
self.mpi_comm.barrier()
compute_times = SimpleFIFO(5)
write_times = SimpleFIFO(5)
orig_rank_start = self.__start_pos
if self.mpi_rank == 0 and self.mpi_size == 1:
if self.__resume_implemented:
print(
'\tThis class (likely) supports interruption and resuming of computations!\n'
'\tIf you are operating in a python console, press Ctrl+C or Cmd+C to abort\n'
'\tIf you are in a Jupyter notebook, click on "Kernel">>"Interrupt"\n'
'\tIf you are operating on a cluster and your job gets killed, re-run the job to resume\n'
)
else:
print(
'\tThis class does NOT support interruption and resuming of computations.\n'
'\tIn order to enable this feature, simply implement the _get_existing_datasets() function'
)
if self.verbose and self.mpi_rank == self.__socket_master_rank:
print('Rank: {} - with nothing loaded has {} free memory'
''.format(self.mpi_rank,
format_size(get_available_memory())))
self._read_data_chunk()
if self.mpi_comm is not None:
self.mpi_comm.barrier()
if self.verbose and self.mpi_rank == self.__socket_master_rank:
print('Rank: {} - with only raw data loaded has {} free memory'
''.format(self.mpi_rank,
format_size(get_available_memory())))
while self.data is not None:
num_jobs_in_batch = self.__end_pos - self.__start_pos
t_start_1 = tm.time()
self._unit_computation(*args, **kwargs)
comp_time = np.round(tm.time() - t_start_1,
decimals=2) # in seconds
time_per_pix = comp_time / num_jobs_in_batch
compute_times.put(time_per_pix)
if self.verbose:
print(
'Rank {} - computed chunk in {} or {} per pixel. Average: {} per pixel'
'.'.format(self.mpi_rank, format_time(comp_time),
format_time(time_per_pix),
format_time(compute_times.get_mean())))
# Ranks can become memory starved. Check memory usage - raw data + results in memory at this point
if self.verbose and self.mpi_rank == self.__socket_master_rank:
print(
'Rank: {} - now holding onto raw data + results has {} free memory'
''.format(self.mpi_rank,
format_size(get_available_memory())))
t_start_2 = tm.time()
self._write_results_chunk()
# NOW, update the positions. Users are NOT allowed to touch start and end pos
self.__start_pos = self.__end_pos
# Leaving in this provision that will allow restarting of processes
if self.mpi_size == 1:
self.h5_results_grp.attrs['last_pixel'] = self.__end_pos
# Child classes don't even have to worry about flushing. Process will do it.
self.h5_main.file.flush()
dump_time = np.round(tm.time() - t_start_2, decimals=2)
write_times.put(dump_time / num_jobs_in_batch)
if self.verbose:
print('Rank {} - wrote its {} pixel chunk in {}'.format(
self.mpi_rank, num_jobs_in_batch, format_time(dump_time)))
time_remaining = (self.__rank_end_pos - self.__end_pos) * \
(compute_times.get_mean() + write_times.get_mean())
if self.verbose or self.mpi_rank == 0:
percent_complete = int(100 *
(self.__end_pos - orig_rank_start) /
(self.__rank_end_pos - orig_rank_start))
print('Rank {} - {}% complete. Time remaining: {}'.format(
self.mpi_rank, percent_complete,
format_time(time_remaining)))
# All ranks should mark the pixels for this batch as completed. 'last_pixel' attribute will be updated later
# Setting each section to 1 independently
for curr_slice in integers_to_slices(self.__pixels_in_batch):
self._h5_status_dset[curr_slice] = 1
self._read_data_chunk()
if self.verbose:
print('Rank {} - Finished computing all jobs!'.format(
self.mpi_rank))
if self.mpi_comm is not None:
self.mpi_comm.barrier()
if self.mpi_rank == 0:
print('Finished processing the entire dataset!')
# Update the legacy 'last_pixel' attribute here:
if self.mpi_rank == 0:
self.h5_results_grp.attrs['last_pixel'] = self.h5_main.shape[0]
return self.h5_results_grp
def test(self, rearrange_clusters=True, override=False):
"""
Clusters the hdf5 dataset and calculates mean response for each cluster. This function does NOT write results to
the hdf5 file. Call :meth:`~pycroscopy.processing.Cluster.compute()` to write to the file.
Handles complex, compound datasets such that the
mean response vector for each cluster matrix is of the same data-type as the input matrix.
Parameters
----------
rearrange_clusters : bool, optional. Default = True
Whether or not the clusters should be re-ordered by relative distances between the mean response
override : bool, optional. default = False
Set to true to recompute results if prior results are available. Else, returns existing results
Returns
-------
labels : :class:`numpy.ndarray`
1D unsigned integer array containing the cluster labels as obtained from the fit
mean_response : :class:`numpy.ndarray`
2D array containing the mean response for each cluster arranged as [cluster number, response]
"""
if not override:
if isinstance(self.duplicate_h5_groups,
list) and len(self.duplicate_h5_groups) > 0:
self.h5_results_grp = self.duplicate_h5_groups[-1]
print('Returning previously computed results from: {}'.format(
self.h5_results_grp.name))
print('set the "override" flag to True to recompute results')
return np.squeeze(reshape_to_n_dims(self.h5_results_grp['Labels'])[0]), \
reshape_to_n_dims(self.h5_results_grp['Mean_Response'])[0]
self.h5_results_grp = None
t1 = time.time()
print('Performing clustering on {}.'.format(self.h5_main.name))
# perform fit on the real dataset
results = self.estimator.fit(
self.data_transform_func(self.h5_main[self.data_slice]))
print('Took {} to compute {}'.format(format_time(time.time() - t1),
self.method_name))
t1 = time.time()
self.__mean_resp = self._get_mean_response(results.labels_)
print('Took {} to calculate mean response per cluster'.format(
format_time(time.time() - t1)))
self.__labels = results.labels_
if rearrange_clusters:
self.__labels, self.__mean_resp = reorder_clusters(
results.labels_, self.__mean_resp, self.data_transform_func)
# TODO: What if test() is called repeatedly?
labels_mat, success = reshape_to_n_dims(
np.expand_dims(np.squeeze(self.__labels), axis=1),
h5_pos=self.h5_main.h5_pos_inds,
h5_spec=np.expand_dims([0], axis=0))
if not success:
raise ValueError(
'Could not reshape labels to N-Dimensional dataset! Error:' +
success)
centroid_mat, success = reshape_to_n_dims(
self.__mean_resp,
h5_spec=self.h5_main.h5_spec_inds[:, :self.num_comps],
h5_pos=np.expand_dims(np.arange(self.__mean_resp.shape[0]),
axis=1))
if not success:
raise ValueError(
'Could not reshape mean response to N-Dimensional dataset! Error:'
+ success)
return np.squeeze(labels_mat), centroid_mat
def test(self, override=False):
"""
Decomposes the hdf5 dataset to calculate the components and projection. This function does NOT write results to
the hdf5 file. Call :meth:`~pycroscopy.processing.Decomposition.compute()` to write to the file. Handles
complex, compound datasets such that the
components are of the same data-type as the input matrix.
Parameters
----------
override : bool, optional. default = False
Set to true to recompute results if prior results are available. Else, returns existing results
Returns
-------
components : :class:`numpy.ndarray`
Components
projections : :class:`numpy.ndarray`
Projections
"""
if not override:
if isinstance(self.duplicate_h5_groups,
list) and len(self.duplicate_h5_groups) > 0:
self.h5_results_grp = self.duplicate_h5_groups[-1]
print('Returning previously computed results from: {}'.format(
self.h5_results_grp.name))
print('set the "override" flag to True to recompute results')
return USIDataset(self.h5_results_grp['Components']).get_n_dim_form(), \
USIDataset(self.h5_results_grp['Projection']).get_n_dim_form()
self.h5_results_grp = None
print('Performing Decomposition on {}.'.format(self.h5_main.name))
t0 = time.time()
self._fit()
self._transform()
print('Took {} to compute {}'.format(format_time(time.time() - t0),
self.method_name))
self.__components = stack_real_to_target_dtype(
self.estimator.components_, self.h5_main.dtype)
projection_mat, success = reshape_to_n_dims(
self.__projection,
h5_pos=self.h5_main.h5_pos_inds,
h5_spec=np.expand_dims(np.arange(self.__projection.shape[1]),
axis=0))
if not success:
raise ValueError(
'Could not reshape projections to N-Dimensional dataset! Error:'
+ success)
components_mat, success = reshape_to_n_dims(
self.__components,
h5_spec=self.h5_main.h5_spec_inds,
h5_pos=np.expand_dims(np.arange(self.__components.shape[0]),
axis=1))
if not success:
raise ValueError(
'Could not reshape components to N-Dimensional dataset! Error:'
+ success)
return components_mat, projection_mat
