def test_nested_contained(self): ''' Test geometry surrounded by other geometries then contained by a larger geometry''' direc_path = os.getcwd() + '/testing/test_data/merge_fully_contained/' file_path = direc_path + '/nested.shp' df = fm.load_shapefile(file_path) df = merge_fully_contained(df) assert len(df) == 1
def test_regular_contained(self): ''' Test donut hole type case. Checks correct merge''' direc_path = os.getcwd() + '/testing/test_data/merge_fully_contained/' file_path = direc_path + '/regular.shp' df = fm.load_shapefile(file_path) # merge df = merge_fully_contained(df) assert len(df) == 1
def test_cols_to_add(self): ''' Check that columns sum correctly when in cols to add''' direc_path = os.getcwd() + '/testing/test_data/merge_fully_contained/' file_path = direc_path + '/regular.shp' df = fm.load_shapefile(file_path) df['value'] = df['value'].astype(float) # merge df = merge_fully_contained(df, cols_to_add=['value']) assert df.at[0, 'value'] == 2
def image_classification(shp_path,
img_path,
num_regions,
num_colors=False,
out_path=False):
'''Generate a dissolved boundary file of larger geometries according to
how the geometry is colored in a corresponding image of the same
geographic region.
The image should be georeferened to the boundary.
Also, the image should be cropped to the extents of the geometry. It is
usually better to do this by hand because the autocropping algorithm
sometimes stops because of a single pixel difference
If there exists a one noncontiguous larger shape, then the number of
regions should be one greater becausue the algorithm will split
noncontiguous regions.
We limit the number of samples to be 500 per geometry for speed purposes
Arguments:
shp_path:
path to the shapefile the algorithm will be performed on
img_path:
path to the image we use for the classification. This should
already be cropped to the boundaries of the geometry. This can
be performed with the function cropped_border_image
num_regions:
the number of regions that should remain at the end of the
algorithm.
num_colors:
The number of colors to reduce the image to. Sometimes helps if
classification regions in the images are differently shades of the
same color within a region. Defaul is no reduction
out_path:
path to save final dataframe if applicable. Default will not save
Output:
df_classified:
dataframe with geometries classified into regions
df:
the original dataframe with color and region assignments
'''
# Load image and shapefile
img = Image.open(img_path)
if num_colors:
img = si.reduce_colors(img, num_colors)
img_arr = np.asarray(img)
df = fm.load_shapefile(shp_path)
# create a color series and region series in the dataframe
df['color'] = pd.Series(dtype=object)
df['region'] = pd.Series(dtype=object)
# Get the boundaries of the geodataframe
bounds = shp.ops.cascaded_union(list(df['geometry'])).bounds
shp_xlen = bounds[2] - bounds[0]
shp_ylen = bounds[3] - bounds[1]
shp_xmin = bounds[0]
shp_ymin = bounds[1]
# Assign each polygon and assign its most common color
for ix, row in df.iterrows():
poly = row['geometry']
df.at[ix, 'color'] = si.most_common_color(poly, img_arr, shp_xmin,
shp_xlen, shp_ymin, shp_ylen,
500)
# Assign each polygon with a certain color a region index
for ix, color in enumerate(df['color'].unique()):
df.loc[df['color'] == color, 'region'] = ix
# Get different region ids
regions = list(df['region'].unique())
# Create the classification dataframe
df_classified = pd.DataFrame(columns=['region', 'geometry'])
# Create classification geoemtries for each region
for ix, region in enumerate(regions):
df_region = df[df['region'] == region]
polys = list(df_region['geometry'])
df_classified.at[ix, 'geometry'] = shp.ops.cascaded_union(polys)
df_classified.at[ix, 'region'] = region
# Convert clasified dataframe into a geodataframe
df_classified = gpd.GeoDataFrame(df_classified, geometry='geometry')
# # Split noncontiguous regions and merge fully contained regions
df_classified = sm.split_noncontiguous(df_classified)
df_classified = sm.merge_fully_contained(df_classified)
# # Merge regions until we have the correct number
df_classified = sm.merge_to_right_number(df_classified, num_regions)
# save file if necessary
if out_path:
fm.save_shapefile(df_classified, out_path)
return df_classified, df
def clean_manual_classification(in_path, classification_col, out_path=False):
'''Generate a dissolved boundary file of larger geometries after
being given a geodataframe with smaller geometries assigned to a value
designated by the classification column.
Will auto-assign unassigned geometries using the greedy shared perimeters
method.
Will also split non-contiguous geometries and merge fully contained
geometries
Usually used when a user has manually classified census blocks into
precincts and needs to clean up their work
Arguments:
in_path:
path dataframe containing smaller geometries
classification_col:
name of colum in df that identifies which
larger "group" each smaller geometry belongs to.
out_path:
path to save final dataframe file if applicable. Default
is false and will not save
'''
df = fm.load_shapefile(in_path)
# obtain unique values in classification column
class_ids = list(df[classification_col].unique())
# determine the number of larger "groups"
num_classes = len(class_ids)
# Check if there are any unassigned census blocks
if None in class_ids:
# decrement number of regions because nan is not an actual region
num_classes -= 1
# Assign unassigned blocks a unique dummy name
for i, _ in df[df[classification_col].isnull()].iterrows():
df.at[i, classification_col] = 'foobar' + str(i)
# Update the classes to include the dummy groups
class_ids = list(df[classification_col].unique())
# Dissolve the boundaries given the group assignments for each small geom
df = sm.dissolve(df, classification_col)
# Split noncontiguous geometries after the dissolve
df = sm.split_noncontiguous(df)
# Merge geometries fully contained in other geometries
df = sm.merge_fully_contained(df)
# Get the correct number of regions
df_nan = df[df[classification_col].str.slice(0, 6) == 'foobar']
ixs_to_merge = df_nan.index.to_list()
df = sm.merge_geometries(df, ixs_to_merge)
# drop neighbor column and reset the indexes
df = df.drop(columns=['neighbors'])
df = df.reset_index(drop=True)
# save file if necessary
if out_path:
fm.save_shapefile(df, out_path)
return df