def _impute_lat_lon(cov_file, subchunk, config):
cov = geoio.RasterioImageSource(cov_file)
cov_data = features.extract_subchunks(cov, subchunk,
config.n_subchunks,
config.patchsize)
nn_imputer = transforms.NearestNeighboursImputer()
cov_data = nn_imputer(cov_data.reshape(cov_data.shape[0], 1))
return cov_data
def _mask_rows(x, subchunk, config):
mask = config.mask
if mask:
mask_source = geoio.RasterioImageSource(mask)
mask_data = features.extract_subchunks(mask_source, subchunk,
config.n_subchunks,
config.patchsize)
mask_data = mask_data.reshape(mask_data.shape[0], 1)
mask_x = mask_data.data[:, 0] != config.retain
log.info('Areas with mask={} will be predicted'.format(config.retain))
assert x.shape[0] == mask_x.shape[0], 'shape mismatch of ' \
'mask and inputs'
x.mask = np.tile(mask_x, (x.shape[1], 1)).T
return x
def cli(verbosity, geotiff, height, absorption, forward, outputdir, noise,
impute):
uncoverml.mllog.configure(verbosity)
log.info("{} simulating gamma sensor model".format(
"Forward" if forward else "Backward"))
if os.path.isdir(geotiff):
log.info("Globbing directory input for tif files")
geotiff = os.path.join(geotiff, "*.tif")
files = glob.glob(geotiff)
my_files = np.array_split(files, mpiops.chunks)[mpiops.chunk_index]
if len(my_files) == 0:
log.critical("No files found. Exiting")
sys.exit()
for f in my_files:
name = os.path.basename(f).rsplit(".", 1)[0]
log.info("Loading {}".format(name))
image_source = geoio.RasterioImageSource(f)
image = Image(image_source)
data = image.data()
# apply transforms here
log.info("Computing sensor footprint")
img_w, img_h, _ = data.shape
S = filtering.sensor_footprint(img_w, img_h,
image.pixsize_x, image.pixsize_y,
height, absorption)
# Apply and unapply the filter (mirrored boundary conditions)
log.info("Applying transform to array of shape {}".format(data.shape))
if forward:
t_data = filtering.fwd_filter(data, S)
else:
orig_mask = data.mask
if np.ma.count_masked(data) > 0:
data = filtering.kernel_impute(data, S)
t_data = filtering.inv_filter(data, S, noise=noise)
if impute:
orig_mask = np.zeros_like(orig_mask, dtype=bool)
t_data = np.ma.MaskedArray(data=t_data.data, mask=orig_mask)
# Write output:
log.info("Writing output to disk")
write_data(t_data, name, image, outputdir, forward)
log.info("All files transformed successfully")
def band_stats(ds, tif, band_no, partitions=100):
log.info('Calculating band stats for {} band {} in {} partitions'.format(
tif, band_no, partitions))
band = ds.GetRasterBand(band_no)
# For statistics calculation
stats = band.ComputeStatistics(False)
no_data_val = band.GetNoDataValue()
data_type = get_datatype(band)
image_source = geoio.RasterioImageSource(tif)
if data_type is 'Categorical':
no_categories = stats[1] - stats[0] + 1
else:
no_categories = np.nan
l = [
basename(tif), band_no, no_data_val, ds.RasterYSize, ds.RasterXSize,
stats[0], stats[1], stats[2], stats[3], data_type, no_categories,
image_nans(image_source, partitions)
]
ds = None # close dataset
return [str(a) for a in l]
def numpy_band_stats(ds, tif, band_no, partitions=100):
band = ds.GetRasterBand(band_no)
data = band.ReadAsArray()
no_data_val = band.GetNoDataValue()
data_type = get_datatype(band)
mask_data = ma.masked_where(data == no_data_val, data)
if data_type is 'Categorical':
no_categories = np.max(mask_data) - np.min(mask_data) + 1
else:
no_categories = np.nan
image_source = geoio.RasterioImageSource(tif)
l = [
basename(tif), band_no, no_data_val, ds.RasterYSize, ds.RasterXSize,
np.min(mask_data),
np.max(mask_data),
np.mean(mask_data),
np.std(mask_data), data_type,
float(no_categories),
image_nans(image_source, partitions)
]
ds = None
return [str(a) for a in l]
def mask_subchunks(subchunk, config):
image_source = geoio.RasterioImageSource(config.mask)
result = features.extract_subchunks(image_source, subchunk,
config.n_subchunks, config.patchsize)
return result
