jnhansen / nd Goto Github PK

I have installed gls, but the error of not finding gsl-config when installing nd

The metadata of a dataset may get corrupted or not contain the CRS information. For those cases it should be possible to manually specify the source CRS in nd.warp.reproject, e.g. by adding an optional parameter src_crs.

Should probably throw a warning if src_crs differs from the CRS detected from metadata.

At the moment, nd.classify is automatically imported which requires scikit-learn. Should probably make this optional.

It could be useful to add a suite of command line tools:

• nd-align — equivalent to nd.warp.align, align a number of input datasets.
• nd-warp — equivalent to nd.warp.reproject, in analogy to gdalwarp (but working on multi-dimensional NetCDF files)
• nd-filter
• nd-classify

Need to include proper inline documentation for the xarray accesors.
There are two ways to do this:

• manually add docstrings and parameters into the accessor methods,
• or (much preferred) automatically patch docstring and signature, like for wrap_algorithm

Most of the remote sensing data has standardized variable names. The library should be able to infer the data source and type from the variable names, convert between different representations and perform different default actions (such as RGB representation) based on these representations.

This is especially relevant polarimetric SAR data, where data will usually be present in the covariance matrix representation, but may need to be converted to and from sigma0, beta0, gamma0 etc.

Getting the following error/warning when importing module:
proj_create: init=epsg:/init=IGNF: syntax not supported in non-PROJ4 emulation mode

Import fails if scikit-learn is not installed because the vector module is using it.

Possible solutions:

• Use a different implementation than the sklearn.preprocessing.LabelEncoder
• Make import of nd.vector optional
• Make scikit-learn a dependency

Related to #11

Add functionality for GeoTiff output from nd accessor

In nd.algorithm.wrap_algorithm(), need to combine docstrings from Algorithm.__init__ and Algorithm.apply, as follows:

• parameters from apply should be prepended to the parameters of __init__
• the return value should be taken from apply.

Bug description

Steps to reproduce

Expected behavior

System (please complete the following information):

• OS:
• Package version [e.g. 0.1]:

It would be great to have a method to create better visualizations of geographic data, e.g. using cartopy / geoviews / etc.

This would involve displaying the data as an overlay on a map, in an arbitrary projection.

• by default, use the dataset projection but allow custom CRS
• by default, use a buffered view around the data bounds but allow custom extent
• allow clipping to Polygon
• resample to desired resolution

OpenCV is currently needed for lots of things in nd.visualize. Might consider replacing this with a lighter weight dependency or making it an optional dependency only required when calling the corresponding methods (i.e. to_rgb() and write_video()).

test_reproject_with_extra_dims sometimes fails:

___________________________________ test_reproject_with_extra_dims[dims4] ___________________________________

dims = {'band': 5, 'extra': 2, 'time': 10, 'x': 20, ...}

    @pytest.mark.parametrize('dims', [
        {'y': 20, 'x': 20, 'time': 10, 'band': 5},
        {'x': 20, 'y': 20, 'time': 10, 'band': 5},
        {'time': 10, 'band': 5, 'x': 20, 'y': 20},
        {'time': 10, 'x': 20, 'band': 5, 'y': 20},
        {'y': 20, 'x': 20, 'time': 10, 'band': 5, 'extra': 2}
    ])
    def test_reproject_with_extra_dims(dims):
        crs1 = _parse_crs('+init=epsg:4326')
        crs2 = _parse_crs('+init=epsg:3395')
        ds = generate_test_dataset(
            dims=dims, crs=crs1
        )
    
        proj = Reprojection(crs=crs2)
        reprojected = proj.apply(ds)
    
        # Check that a reprojected slice of the dataset is the same as
        # the slice of the reprojection of the entire dataset.
        slices = [
            {'band': 3},
            {'time': slice(1, 3)}
        ]
        for s in slices:
            xr_assert_equal(
                proj.apply(ds.isel(**s)),
>               reprojected.isel(**s)
            )
E           AssertionError: Left and right Dataset objects are not equal
E           
E           
E           Differing data variables:
E           L   C12__im  (time, extra, y, x) float64 -0.4385 0.295 0.5144 ... 0.0594 -0.1833
E           R   C12__im  (time, y, x, extra) float64 -0.4385 0.7736 ... -0.1302 -0.1833
E           L   C12__re  (time, extra, y, x) float64 0.5077 0.1089 0.2545 ... 0.6105 0.2276
E           R   C12__re  (time, y, x, extra) float64 0.5077 0.1572 0.1089 ... -0.51 0.2276
E           L   C22      (time, extra, y, x) float64 -0.1661 -1.155 ... -0.6713 -0.06515
E           R   C22      (time, y, x, extra) float64 -0.1661 -0.3903 ... 0.1816 -0.06515
E           L   C11      (time, extra, y, x) float64 -0.6673 0.6047 -0.325 ... 0.5696 0.1056
E           R   C11      (time, y, x, extra) float64 -0.6673 0.03727 ... 0.01081 0.1056

nd/tests/test_warp.py:535: AssertionError

Is your feature request related to a problem? Please describe.
The following methods are not currently very consistent in what information is used to extract the corresponding parameter:

• nd.warp.get_transform()
• nd.warp.get_bounds()
• nd.warp.get_extent()
• nd.warp.get_resolution()

Describe the solution you'd like
The most reliable information to determine the transform etc. is probably the coordinate arrays and should be used even if there is relevant metadata present.

The methods may optionally raise a warning if the coordinate arrays don't seem to match the values specified in the metadata.

Need to create tests for the nd.vector module, particularly nd.vector.rasterize().

• Create a test shapefile with attributes
• test correct rendering of polygons into raster
• test correct handling of both categorical and numeric labels