You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
ENH: Add a function checking if an array is computed or still chunked (dask.is_computed, dask.is_chunked) and its equivalents to assert whether arrays have been computed or not (ci.assert_chunked and ci.assert_computed) (#47)
ENH: Add an environment variable (SERTIT_DEFAULT_CHUNKS) to control default chunks in xarray rasters and a function better managing default chunks (dask.get_default_chunks) (#47)
ENH: Add a classify function to classify a raster according to bins and values (#40)
ENH: Add a performance module (sertit.perf) featuring max core-related functions and environment variables
ENH: Create a wrapper for reprojection, handling different usecases: dask, RPCs, etc. Using odc.geo, rioxarray or rasterio according to the usecase (#41)
ENH: Allow the user to pass environment variables to dask clients (#31)
ENH: Create a os_utils module grouping several functions looking for useful executables, such as QGIS
OPTIM: Use odc.xr.xr_reproject to enable lazy dask computation for rescaling data in rasters.read (in case of upscaling only)
OPTIM: Make rasters.read_bit_array and rasters.read_uint8_array lazy (#27)
OPTIM: Make rasters.rasterize lazy
OPTIM: Some other minor optimizations regarding to dask usage (#27)
FIX: Check if rasters function outputs a lazy or computed array (for now only throw warnings as most functions are not lazy) (#47)
FIX: Fix ci.assert_val for xr.DataArrayCoordinates
FIX: Update also long_name with new_name in rasters.set_metadata
FIX: Fix rasterizing vectors with multiband rasters (#43)
FIX: Don't try to read KML with pyogrio <= 0.11.1
FIX: Fix ci.assert_val with nans
FIX: Better nan management in ci.assert_raster_almost_equal_magnitude
FIX: Make vectorize-based functions to work with xarray.Dataset
FIX: Fix original dtype in read_bit_array
FIX: Better kwargs management in merge_gtiff
FIX: Fix reading a tuple (array, metadata) with rasters.read
CI: Correctly compute numpy case with dask_env decorator
CI: Correctly manage dask in test_rasters.test_set_nodata and test_rasters.test_where (create a dask array to evaluate laziness)
CI: Check correctly the number of attended computations per raster tests (#47)
DOC: Document which function of rasters isn't lazy yet (currently: sieve, rasterize, vectorize and other functions depending on it) (#47)