def train_raw(raster_file, train_geo):
training_vectors = gpd.read_file(train_geo)
class_dict = crop_classes(train_geo)
print(class_dict)
# set up our training data lists
# this larger cell reads data from a raster file for each training vector
x_raw = []
y_raw = []
with rasterio.open(raster_file, 'r') as src:
for (label, geom) in zip(training_vectors.name,
training_vectors.geometry):
# read the raster data matching the geometry bounds
window = bounds_window(geom.bounds, src.transform)
# store our window information
window_affine = src.window_transform(window)
fsrc = src.read(window=window)
# rasterize the geometry into the larger shape and affine
mask = rasterize([(geom, 1)],
out_shape=fsrc.shape[1:],
transform=window_affine,
fill=0,
dtype='uint8',
all_touched=True).astype(bool)
# for each label pixel (places where the mask is true)...
label_pixels = np.argwhere(mask)
for (row, col) in label_pixels:
# add a pixel of data to X
data = fsrc[:, row, col]
one_x = np.nan_to_num(data, nan=1e-3)
x_raw.append(one_x)
y_raw.append(class_dict[label])
return x_raw, y_raw
def get_zonal_stat(geojson: Feature, raster: str) -> Tuple[float, float]:
"""Return zonal statistics."""
geom = shape(geojson.geometry.dict())
with rasterio.open(raster) as src:
# read the raster data matching the geometry bounds
window = bounds_window(geom.bounds, src.transform)
# store our window information & read
window_affine = src.window_transform(window)
data = src.read(window=window)
# calculate the coverage of pixels for weighting
pctcover = rasterize_pctcover(geom, atrans=window_affine, shape=data.shape[1:])
return (
np.average(data[0], weights=pctcover),
np.nanmedian(data),
)
def test_bounds_window():
with rasterio.open(raster) as src:
assert bounds_window(src.bounds, src.transform) == \
((0, src.shape[0]), (0, src.shape[1]))
def test_bounds_window():
with rasterio.open(raster) as src:
assert bounds_window(src.bounds, src.transform) == \
((0, src.shape[0]), (0, src.shape[1]))