def test_cache(dataset_config):
""" Test cache directory for ability to read from or write to
Args:
dataset_config (dict): dictionary of dataset configuration options
Returns:
(read_cache, write_cache): tuple of bools describing ability to read from
and write to cache directory
"""
# Try to find / use cache
read_cache = False
write_cache = False
cache_dir = dataset_config.get('cache_line_dir')
if cache_dir:
# Test existence
if os.path.isdir(cache_dir):
if os.access(cache_dir, os.R_OK):
read_cache = True
if os.access(cache_dir, os.W_OK):
write_cache = True
if read_cache and not write_cache:
logger.warning('Cache directory exists but is not writable')
else:
# If it doesn't already exist, can we create it?
try:
os.makedirs(cache_dir)
except:
logger.warning('Could not create cache directory')
else:
read_cache = True
write_cache = True
logger.debug(
'Attempt reading in from cache directory?: {b}'.format(b=read_cache))
logger.debug(
'Attempt writing to cache directory?: {b}'.format(b=write_cache))
return read_cache, write_cache
def test_cache(dataset_config):
""" Test cache directory for ability to read from or write to
Args:
dataset_config (dict): dictionary of dataset configuration options
Returns:
tuple: tuple of bools describing ability to read from and write to
cache directory
"""
# Try to find / use cache
read_cache = False
write_cache = False
cache_dir = dataset_config.get('cache_line_dir')
if cache_dir:
# Test existence
if os.path.isdir(cache_dir):
if os.access(cache_dir, os.R_OK):
read_cache = True
if os.access(cache_dir, os.W_OK):
write_cache = True
if read_cache and not write_cache:
logger.warning('Cache directory exists but is not writable')
else:
# If it doesn't already exist, can we create it?
try:
os.makedirs(cache_dir)
except:
logger.warning('Could not create cache directory')
else:
read_cache = True
write_cache = True
logger.debug('Attempt reading in from cache directory?: {b}'.format(
b=read_cache))
logger.debug('Attempt writing to cache directory?: {b}'.format(
b=write_cache))
return read_cache, write_cache
def iter_records(records, warn_on_empty=False, yield_filename=False):
""" Iterates over records, returning result NumPy array
Args:
records (list): List containing filenames of results
warn_on_empty (bool, optional): Log warning if result contained no
result records (default: False)
yield_filename (bool, optional): Yield the filename and the record
Yields:
np.ndarray or tuple: Result saved in record and the filename, if desired
"""
n_records = len(records)
for _i, r in enumerate(records):
# Verbose progress
if np.mod(_i, 100) == 0:
logger.debug('{0:.1f}%'.format(_i / n_records * 100))
# Open output
try:
rec = np.load(r)['record']
except (ValueError, AssertionError, IOError) as e:
logger.warning('Error reading a result file (may be corrupted) '
'({}): {}'.format(r, str(e)))
continue
if rec.shape[0] == 0:
# No values in this file
if warn_on_empty:
logger.warning('Could not find results in {f}'.format(f=r))
continue
if yield_filename:
yield rec, r
else:
yield rec
def iter_records(records, warn_on_empty=False, yield_filename=False):
""" Iterates over records, returning result NumPy array
Args:
records (list): List containing filenames of results
warn_on_empty (bool, optional): Log warning if result contained no
result records (default: False)
yield_filename (bool, optional): Yield the filename and the record
Yields:
np.ndarray or tuple: Result saved in record and the filename, if desired
"""
n_records = len(records)
for _i, r in enumerate(records):
# Verbose progress
if np.mod(_i, 100) == 0:
logger.debug('{0:.1f}%'.format(_i / n_records * 100))
# Open output
try:
rec = np.load(r)['record']
except (ValueError, AssertionError, IOError) as e:
logger.warning('Error reading a result file (may be corrupted) '
'({}): {}'.format(r, str(e)))
continue
if rec.shape[0] == 0:
# No values in this file
if warn_on_empty:
logger.warning('Could not find results in {f}'.format(f=r))
continue
if yield_filename:
yield rec, r
else:
yield rec
def write_output(raster, output, image_ds, gdal_frmt, ndv, band_names=None):
""" Write raster to output file """
from osgeo import gdal, gdal_array
logger.debug('Writing output to disk')
driver = gdal.GetDriverByName(str(gdal_frmt))
if len(raster.shape) > 2:
nband = raster.shape[2]
else:
nband = 1
ds = driver.Create(
output,
image_ds.RasterXSize, image_ds.RasterYSize, nband,
gdal_array.NumericTypeCodeToGDALTypeCode(raster.dtype.type)
)
if band_names is not None:
if len(band_names) != nband:
logger.error('Did not get enough names for all bands')
sys.exit(1)
if raster.ndim > 2:
for b in range(nband):
logger.debug(' writing band {b}'.format(b=b + 1))
ds.GetRasterBand(b + 1).WriteArray(raster[:, :, b])
ds.GetRasterBand(b + 1).SetNoDataValue(ndv)
if band_names is not None:
ds.GetRasterBand(b + 1).SetDescription(band_names[b])
ds.GetRasterBand(b + 1).SetMetadata({
'band_{i}'.format(i=b + 1): band_names[b]
})
else:
logger.debug(' writing band')
ds.GetRasterBand(1).WriteArray(raster)
ds.GetRasterBand(1).SetNoDataValue(ndv)
if band_names is not None:
ds.GetRasterBand(1).SetDescription(band_names[0])
ds.GetRasterBand(1).SetMetadata({'band_1': band_names[0]})
ds.SetProjection(image_ds.GetProjection())
ds.SetGeoTransform(image_ds.GetGeoTransform())
ds = None
def write_output(raster, output, image_ds, gdal_frmt, ndv, band_names=None):
""" Write raster to output file """
from osgeo import gdal, gdal_array
logger.debug('Writing output to disk')
driver = gdal.GetDriverByName(str(gdal_frmt))
if len(raster.shape) > 2:
nband = raster.shape[2]
else:
nband = 1
ds = driver.Create(
output, image_ds.RasterXSize, image_ds.RasterYSize, nband,
gdal_array.NumericTypeCodeToGDALTypeCode(raster.dtype.type))
if band_names is not None:
if len(band_names) != nband:
logger.error('Did not get enough names for all bands')
sys.exit(1)
if raster.ndim > 2:
for b in range(nband):
logger.debug(' writing band {b}'.format(b=b + 1))
ds.GetRasterBand(b + 1).WriteArray(raster[:, :, b])
ds.GetRasterBand(b + 1).SetNoDataValue(ndv)
if band_names is not None:
ds.GetRasterBand(b + 1).SetDescription(band_names[b])
ds.GetRasterBand(b + 1).SetMetadata(
{'band_{i}'.format(i=b + 1): band_names[b]})
else:
logger.debug(' writing band')
ds.GetRasterBand(1).WriteArray(raster)
ds.GetRasterBand(1).SetNoDataValue(ndv)
if band_names is not None:
ds.GetRasterBand(1).SetDescription(band_names[0])
ds.GetRasterBand(1).SetMetadata({'band_1': band_names[0]})
ds.SetProjection(image_ds.GetProjection())
ds.SetGeoTransform(image_ds.GetGeoTransform())
ds = None
def update_cache_file(images,
image_IDs,
old_cache_filename,
new_cache_filename,
line,
reader,
reader_kwargs={}):
""" Modify an existing cache file to contain data within `images`
This should be useful for updating a set of cache files to reflect
modifications to the timeseries dataset without completely reading the
data into another cache file.
For example, the cache file could be updated to reflect the deletion of
a misregistered or cloudy image. Another common example would be for
updating cache files to include newly acquired observations.
Note that this updater will not handle updating cache files to include
new bands.
Args:
images (iterable): list of new image filenames
image_IDs (iterable): list of new image identifying strings
old_cache_filename (str): filename of cache file to update
new_cache_filename (str): filename of new cache file which includes
modified data
line (int): the line of data to be updated
reader (callable): GDAL or BIP image reader function from `yatsm.readers`
reader_kwargs (dict): additional keyword arguments for `reader` other
than the filenames to read and the line to read
Raises:
ValueError: Raise error if old cache file does not record `image_IDs`
"""
images = np.asarray(images)
image_IDs = np.asarray(image_IDs)
# Cannot proceed if old cache file doesn't store filenames
old_cache = np.load(old_cache_filename)
if _image_ID_str not in old_cache.files:
raise ValueError('Cannot update cache.'
'Old cache file does not store image IDs.')
old_IDs = old_cache[_image_ID_str]
old_Y = old_cache['Y']
nband, _, ncol = old_Y.shape
# Create new Y and add in values retained from old cache
new_Y = np.zeros((nband, image_IDs.size, ncol), dtype=old_Y.dtype.type)
new_IDs = np.zeros(image_IDs.size, dtype=image_IDs.dtype)
# Check deletions -- find which indices to retain in new cache
retain_old = np.where(np.in1d(old_IDs, image_IDs))[0]
if retain_old.size == 0:
logger.warning('No image IDs in common in old cache file.')
else:
logger.debug(' retaining {r} of {n} images'.format(
r=retain_old.size, n=old_IDs.size))
# Find indices of old data to insert into new data
idx_old_IDs = np.argsort(old_IDs)
sorted_old_IDs = old_IDs[idx_old_IDs]
idx_IDs = np.searchsorted(sorted_old_IDs,
image_IDs[np.in1d(image_IDs, old_IDs)])
retain_old = idx_old_IDs[idx_IDs]
# Indices to insert into new data
retain_new = np.where(np.in1d(image_IDs, old_IDs))[0]
new_Y[:, retain_new, :] = old_Y[:, retain_old, :]
new_IDs[retain_new] = old_IDs[retain_old]
# Check additions -- find which indices we need to insert
insert = np.where(np.in1d(image_IDs, old_IDs) == False)[0]
if retain_old.size == 0 and insert.size == 0:
raise ValueError('Cannot update cache file -- '
'no data retained or added')
# Read in the remaining data from disk
if insert.size > 0:
logger.debug(
'Inserting {n} new images into cache'.format(n=insert.size))
insert_Y = reader(images[insert], line, **reader_kwargs)
new_Y[:, insert, :] = insert_Y
new_IDs[insert] = image_IDs[insert]
np.testing.assert_equal(new_IDs, image_IDs)
# Save
write_cache_file(new_cache_filename, new_Y, image_IDs)
def update_cache_file(images, image_IDs,
old_cache_filename, new_cache_filename,
line, reader):
""" Modify an existing cache file to contain data within `images`
This should be useful for updating a set of cache files to reflect
modifications to the timeseries dataset without completely reading the
data into another cache file.
For example, the cache file could be updated to reflect the deletion of
a misregistered or cloudy image. Another common example would be for
updating cache files to include newly acquired observations.
Note that this updater will not handle updating cache files to include
new bands.
Args:
images (iterable): list of new image filenames
image_IDs (iterable): list of new image identifying strings
old_cache_filename (str): filename of cache file to update
new_cache_filename (str): filename of new cache file which includes
modified data
line (int): the line of data to be updated
reader (callable): GDAL or BIP image reader function from
:mod:`yatsm.io.stack_line_readers`
Raises:
ValueError: Raise error if old cache file does not record ``image_IDs``
"""
images = np.asarray(images)
image_IDs = np.asarray(image_IDs)
# Cannot proceed if old cache file doesn't store filenames
old_cache = np.load(old_cache_filename)
if _image_ID_str not in old_cache.files:
raise ValueError('Cannot update cache.'
'Old cache file does not store image IDs.')
old_IDs = old_cache[_image_ID_str]
old_Y = old_cache['Y']
nband, _, ncol = old_Y.shape
# Create new Y and add in values retained from old cache
new_Y = np.zeros((nband, image_IDs.size, ncol),
dtype=old_Y.dtype.type)
new_IDs = np.zeros(image_IDs.size, dtype=image_IDs.dtype)
# Check deletions -- find which indices to retain in new cache
retain_old = np.where(np.in1d(old_IDs, image_IDs))[0]
if retain_old.size == 0:
logger.warning('No image IDs in common in old cache file.')
else:
logger.debug(' retaining {r} of {n} images'.format(
r=retain_old.size, n=old_IDs.size))
# Find indices of old data to insert into new data
idx_old_IDs = np.argsort(old_IDs)
sorted_old_IDs = old_IDs[idx_old_IDs]
idx_IDs = np.searchsorted(sorted_old_IDs,
image_IDs[np.in1d(image_IDs, old_IDs)])
retain_old = idx_old_IDs[idx_IDs]
# Indices to insert into new data
retain_new = np.where(np.in1d(image_IDs, old_IDs))[0]
new_Y[:, retain_new, :] = old_Y[:, retain_old, :]
new_IDs[retain_new] = old_IDs[retain_old]
# Check additions -- find which indices we need to insert
insert = np.where(np.in1d(image_IDs, old_IDs, invert=True))[0]
if retain_old.size == 0 and insert.size == 0:
raise ValueError('Cannot update cache file -- '
'no data retained or added')
# Read in the remaining data from disk
if insert.size > 0:
logger.debug('Inserting {n} new images into cache'.format(
n=insert.size))
insert_Y = reader.read_row(images[insert], line)
new_Y[:, insert, :] = insert_Y
new_IDs[insert] = image_IDs[insert]
np.testing.assert_equal(new_IDs, image_IDs)
# Save
write_cache_file(new_cache_filename, new_Y, image_IDs)
