def load_training_from_dataset(self, dataset, training_set, sample=0.2):
"""Retieves training data from the database based on intersection with an xr Dataset
Args:
dataset (xarray.Dataset): Typical Dataset object generated using one of
the datacube load method (GridWorkflow or Datacube clases)
training_set (str): Name of the training set identifier to select the right
training data
sample (float): Float between 0 and 1. Proportion of intersection geometry to sample.
Performs random sampling of the geometries
Return:
list: A feature collection reprojected to the CRS of the xarray Dataset
"""
geobox = dataset.geobox
crs = str(dataset.crs)
poly = Polygon.from_geobox(geobox)
query_set = TrainClassification.objects.filter(
train_object__the_geom__contained=poly,
training_set=training_set).prefetch_related(
'train_object', 'interpret_tag')
if 0 < sample < 1:
nsample = floor(query_set.count() * sample)
query_set = query_set.order_by('?')[:nsample]
fc = (train_object_to_feature(x, crs) for x in query_set)
return fc
def load_segmentation_from_dataset(geoarray, segmentation_name):
"""Retrieve a segmentation intersecting with a geoarray from the database
Args:
geoarray (xarray.Dataset): Typical Dataset object generated using one of
the datacube load method (GridWorkflow or Datacube clases)
segmentation_name (str): Unique segmentation identifier
Return:
list: A geojson like feature collection. See ``madmex.overlay.conversions.predict_object_to_feature``
for more details about the structure of each feature.
"""
# TODO: We'll probably have to introduce a buffer here to account for the curving of reprojected extent
geobox = geoarray.geobox
crs = geoarray.crs._crs.ExportToProj4()
poly = Polygon.from_geobox(geobox)
query_set = PredictObject.objects.filter(
the_geom__contained=poly,
segmentation_information__name=segmentation_name)
fc = [predict_object_to_feature(x, crs) for x in query_set]
return fc
