Update utils to do linear transformation of polygons

Brings in https://github.com/alphagov/notifications-utils/pull/889/files

At the moment, we are not doing any transformation of features before
applying geometric algorithms to them. This is, in effect, assuming that
the earth is flat.

This new version of utils implements the transformation of our polygons
to a Cartesian plane. In other words, it converts them from being
defined in spherical degrees to metres.

For the admin app this means we need to convert places where the code
expects things to be measured in degrees to work in metres instead.
This commit is contained in:
Chris Hill-Scott
2021-08-03 18:46:30 +01:00
parent 6b52735dac
commit 6cb326f153
16 changed files with 99 additions and 80 deletions

View File

@@ -76,6 +76,7 @@ from app.formatters import (
recipient_count, recipient_count,
recipient_count_label, recipient_count_label,
round_to_significant_figures, round_to_significant_figures,
square_metres_to_square_miles,
valid_phone_number, valid_phone_number,
) )
from app.models.organisation import Organisation from app.models.organisation import Organisation
@@ -572,6 +573,7 @@ def add_template_filters(application):
message_count_noun, message_count_noun,
format_mobile_network, format_mobile_network,
format_yes_no, format_yes_no,
square_metres_to_square_miles,
]: ]:
application.add_template_filter(fn) application.add_template_filter(fn)

View File

@@ -44,14 +44,30 @@ def simplify_geometry(feature):
def clean_up_invalid_polygons(polygons, indent=" "): def clean_up_invalid_polygons(polygons, indent=" "):
"""
This function expects a list of lists of coordinates defined in degrees
"""
for index, polygon in enumerate(polygons): for index, polygon in enumerate(polygons):
shapely_polygon = Polygon(polygon) shapely_polygon = Polygon(polygon)
if shapely_polygon.is_valid: # Some of our data has points which are incredibly close
# together. In some cases they are close enough to be duplicates
# at a given precision, which makes an invalid topology. In
# other cases they are close enough that, when converting from
# one coordinate system to another, they shift about enough to
# create self-intersection. The fix in both cases is to reduce
# the precision of the coordinates and then apply simplification
# with a tolerance of 0.
simplified_polygon = wkt.loads(wkt.dumps(
shapely_polygon,
rounding_precision=Polygons.output_precision_in_decimal_places - 1
)).simplify(0)
if simplified_polygon.is_valid:
print( # noqa: T001 print( # noqa: T001
f"{indent}Polygon {index + 1}/{len(polygons)} is valid" f"{indent}Polygon {index + 1}/{len(polygons)} is valid"
) )
yield polygon yield simplified_polygon
else: else:
invalid_polygons.append(shapely_polygon) invalid_polygons.append(shapely_polygon)
@@ -59,7 +75,7 @@ def clean_up_invalid_polygons(polygons, indent=" "):
# Weve found polygons where all the points line up, so they # Weve found polygons where all the points line up, so they
# dont have an area. They wouldnt contribute to a broadcast # dont have an area. They wouldnt contribute to a broadcast
# so we can ignore them. # so we can ignore them.
if shapely_polygon.area == 0: if simplified_polygon.area == 0:
print( # noqa: T001 print( # noqa: T001
f"{indent}Polygon {index + 1}/{len(polygons)} has 0 area, skipping" f"{indent}Polygon {index + 1}/{len(polygons)} has 0 area, skipping"
) )
@@ -69,18 +85,6 @@ def clean_up_invalid_polygons(polygons, indent=" "):
f"{indent}Polygon {index + 1}/{len(polygons)} needs fixing..." f"{indent}Polygon {index + 1}/{len(polygons)} needs fixing..."
) )
# The simplest kind of invalid polygon is one that has two
# duplicate points in a row at a given precision, so the
# first thing to try is removing those points using
# simplification with a tolerance of 0
polygon_with_duplicate_points_removed = wkt.loads(
wkt.dumps(shapely_polygon, rounding_precision=5)
).simplify(0)
if polygon_with_duplicate_points_removed.is_valid:
yield polygon_with_duplicate_points_removed
continue
# Buffering with a size of 0 is a trick to make valid # Buffering with a size of 0 is a trick to make valid
# geometries from polygons that self intersect # geometries from polygons that self intersect
buffered = shapely_polygon.buffer(0) buffered = shapely_polygon.buffer(0)
@@ -101,7 +105,7 @@ def clean_up_invalid_polygons(polygons, indent=" "):
# Make sure the polygon is now valid, and that we havent # Make sure the polygon is now valid, and that we havent
# drastically transformed the polygon by fixing it # drastically transformed the polygon by fixing it
assert fixed_polygon.is_valid assert fixed_polygon.is_valid
assert isclose(fixed_polygon.area, polygon.area, rel_tol=0.001) assert isclose(fixed_polygon.area, shapely_polygon.area, rel_tol=0.001)
print( # noqa: T001 print( # noqa: T001
f"{indent}Polygon {index + 1}/{len(polygons)} fixed!" f"{indent}Polygon {index + 1}/{len(polygons)} fixed!"
@@ -115,14 +119,22 @@ def polygons_and_simplified_polygons(feature):
# cheat and shortcut out # cheat and shortcut out
return [], [] return [], []
polygons = Polygons(simplify_geometry(feature)) raw_polygons = simplify_geometry(feature)
polygons = list(clean_up_invalid_polygons(polygons)) clean_raw_polygons = [
polygons = Polygons(polygons) [[x, y] for x, y in polygon.exterior.coords]
for polygon in clean_up_invalid_polygons(raw_polygons)
]
polygons = Polygons(clean_raw_polygons)
full_resolution = polygons.remove_too_small full_resolution = polygons.remove_too_small
smoothed = full_resolution.smooth smoothed = full_resolution.smooth
simplified = smoothed.simplify simplified = smoothed.simplify
if not (len(full_resolution) or len(simplified)):
raise RuntimeError(
'Polygon of 0 size found'
)
print( # noqa: T001 print( # noqa: T001
f' Original:{full_resolution.point_count: >5} points' f' Original:{full_resolution.point_count: >5} points'
f' Smoothed:{smoothed.point_count: >5} points' f' Smoothed:{smoothed.point_count: >5} points'

View File

@@ -7,6 +7,8 @@ from notifications_utils.serialised_model import SerialisedModelCollection
from rtreelib import Rect from rtreelib import Rect
from werkzeug.utils import cached_property from werkzeug.utils import cached_property
from app.formatters import square_metres_to_square_miles
from .populations import CITY_OF_LONDON from .populations import CITY_OF_LONDON
from .repo import BroadcastAreasRepository, rtree_index from .repo import BroadcastAreasRepository, rtree_index
@@ -55,13 +57,13 @@ class BaseBroadcastArea(ABC):
@cached_property @cached_property
def simple_polygons_with_bleed(self): def simple_polygons_with_bleed(self):
return self.simple_polygons.bleed_by(self.estimated_bleed_in_degrees) return self.simple_polygons.bleed_by(self.estimated_bleed_in_m)
@cached_property @cached_property
def phone_density(self): def phone_density(self):
if not self.polygons.estimated_area: if not self.polygons.estimated_area:
return 0 return 0
return self.count_of_phones / self.polygons.estimated_area return self.count_of_phones / square_metres_to_square_miles(self.polygons.estimated_area)
@property @property
def estimated_bleed_in_m(self): def estimated_bleed_in_m(self):
@@ -72,14 +74,10 @@ class BaseBroadcastArea(ABC):
range masts, so the typical bleed will be high (up to 5,000m). range masts, so the typical bleed will be high (up to 5,000m).
''' '''
if self.phone_density < 1: if self.phone_density < 1:
return Polygons.approx_bleed_in_degrees * Polygons.approx_metres_to_degree return Polygons.approx_bleed_in_m
estimated_bleed = 5_900 - (math.log(self.phone_density, 10) * 1_250) estimated_bleed = 5_900 - (math.log(self.phone_density, 10) * 1_250)
return max(500, min(estimated_bleed, 5000)) return max(500, min(estimated_bleed, 5000))
@property
def estimated_bleed_in_degrees(self):
return self.estimated_bleed_in_m / Polygons.approx_metres_to_degree
class BroadcastArea(BaseBroadcastArea, SortableMixin): class BroadcastArea(BaseBroadcastArea, SortableMixin):
@@ -123,7 +121,7 @@ class BroadcastArea(BaseBroadcastArea, SortableMixin):
def count_of_phones(self): def count_of_phones(self):
if self.id.endswith(CITY_OF_LONDON.WARDS): if self.id.endswith(CITY_OF_LONDON.WARDS):
return CITY_OF_LONDON.DAYTIME_POPULATION * ( return CITY_OF_LONDON.DAYTIME_POPULATION * (
self.polygons.estimated_area / CITY_OF_LONDON.AREA_SQUARE_MILES self.polygons.estimated_area / CITY_OF_LONDON.AREA_SQUARE_METRES
) )
if self.sub_areas: if self.sub_areas:
return sum(area.count_of_phones for area in self.sub_areas) return sum(area.count_of_phones for area in self.sub_areas)
@@ -169,7 +167,10 @@ class CustomBroadcastArea(BaseBroadcastArea):
return Polygons( return Polygons(
# Polygons in the DB are stored with the coordinate pair # Polygons in the DB are stored with the coordinate pair
# order flipped this flips them back again # order flipped this flips them back again
Polygons(self._polygons).as_coordinate_pairs_lat_long [
[[lat, long] for long, lat in polygon]
for polygon in self._polygons
]
) )
simple_polygons = polygons simple_polygons = polygons

View File

@@ -31,8 +31,9 @@ class CITY_OF_LONDON:
) )
# https://data.london.gov.uk/blog/daytime-population-of-london-2014/ # https://data.london.gov.uk/blog/daytime-population-of-london-2014/
DAYTIME_POPULATION = 553_000 DAYTIME_POPULATION = 553_000
# Approx area of the polygons were storing, not the actual area # Exact area of the polygons were storing, which matches the 2.9km²
AREA_SQUARE_MILES = 1.78 # given by https://en.wikipedia.org/wiki/City_of_London
AREA_SQUARE_METRES = 2_885_598
class BRYHER: class BRYHER:

Binary file not shown.

View File

@@ -524,3 +524,7 @@ def format_yes_no(value, yes='Yes', no='No', none='No'):
if value is None: if value is None:
return none return none
return yes if value else no return yes if value else no
def square_metres_to_square_miles(area):
return area * 3.86e-7

View File

@@ -221,7 +221,8 @@ class BroadcastMessage(JSONModel):
polygons = Polygons( polygons = Polygons(
list(itertools.chain(*( list(itertools.chain(*(
getattr(area, area_attribute) for area in self.areas getattr(area, area_attribute) for area in self.areas
))) ))),
utm_crs=self.areas[0].polygons.utm_polygons.utm_crs,
) )
if area_attribute != 'polygons' and len(self.areas) > 1: if area_attribute != 'polygons' and len(self.areas) > 1:
# Were combining simplified polygons from multiple areas so we # Were combining simplified polygons from multiple areas so we

View File

@@ -6,7 +6,7 @@
<polygon points="25 5, 45 25, 25 45, 5 25" stroke="#0B0B0C" stroke-width="2" fill="#96C6E2" /> <polygon points="25 5, 45 25, 25 45, 5 25" stroke="#0B0B0C" stroke-width="2" fill="#96C6E2" />
</svg> </svg>
<span class="govuk-visually-hidden"> <span class="govuk-visually-hidden">
An area of {{ (broadcast_message.simple_polygons.estimated_area)|round_to_significant_figures(1)|format_thousands }} square miles&nbsp; An area of {{ (broadcast_message.simple_polygons.estimated_area)|square_metres_to_square_miles|round_to_significant_figures(1)|format_thousands }} square miles&nbsp;
</span> </span>
Will get Will get
<span class="govuk-visually-hidden"> <span class="govuk-visually-hidden">
@@ -20,7 +20,7 @@
<polygon points="25 5, 45 25, 25 45, 5 25" stroke="#005ea5" stroke-opacity="1" stroke-width="2" stroke-linecap="square" stroke-linejoin="round" stroke-dasharray="4,7.5,5,7.5,8,8,5,8,7.5,8,5,8,7,8,5,8,4" fill="#2B8CC4" fill-opacity="0.15" /> <polygon points="25 5, 45 25, 25 45, 5 25" stroke="#005ea5" stroke-opacity="1" stroke-width="2" stroke-linecap="square" stroke-linejoin="round" stroke-dasharray="4,7.5,5,7.5,8,8,5,8,7.5,8,5,8,7,8,5,8,4" fill="#2B8CC4" fill-opacity="0.15" />
</svg> </svg>
<span class="govuk-visually-hidden"> <span class="govuk-visually-hidden">
An extra area of {{ (broadcast_message.simple_polygons_with_bleed.estimated_area - broadcast_message.simple_polygons.estimated_area)|round_to_significant_figures(1)|format_thousands }} square miles is&nbsp; An extra area of {{ (broadcast_message.simple_polygons_with_bleed.estimated_area - broadcast_message.simple_polygons.estimated_area)|square_metres_to_square_miles|round_to_significant_figures(1)|format_thousands }} square miles is&nbsp;
</span> </span>
Likely to get Likely to get
<span class="govuk-visually-hidden"> <span class="govuk-visually-hidden">

View File

@@ -31,7 +31,7 @@ pyproj==3.2.1
awscli-cwlogs>=1.4,<1.5 awscli-cwlogs>=1.4,<1.5
itsdangerous==1.1.0 # pyup: <2 itsdangerous==1.1.0 # pyup: <2
notifications-utils @ git+https://github.com/alphagov/notifications-utils.git@48.0.0 notifications-utils @ git+https://github.com/alphagov/notifications-utils.git@49.0.0
govuk-frontend-jinja @ git+https://github.com/alphagov/govuk-frontend-jinja.git@v0.5.8-alpha govuk-frontend-jinja @ git+https://github.com/alphagov/govuk-frontend-jinja.git@v0.5.8-alpha
# cryptography 3.4+ incorporates Rust code, which isn't supported on PaaS # cryptography 3.4+ incorporates Rust code, which isn't supported on PaaS

View File

@@ -123,7 +123,7 @@ mistune==0.8.4
# via notifications-utils # via notifications-utils
notifications-python-client==6.3.0 notifications-python-client==6.3.0
# via -r requirements.in # via -r requirements.in
notifications-utils @ git+https://github.com/alphagov/notifications-utils.git@48.0.0 notifications-utils @ git+https://github.com/alphagov/notifications-utils.git@49.0.0
# via -r requirements.in # via -r requirements.in
openpyxl==3.0.7 openpyxl==3.0.7
# via pyexcel-xlsx # via pyexcel-xlsx
@@ -165,7 +165,9 @@ pyparsing==2.4.7
pypdf2==1.26.0 pypdf2==1.26.0
# via notifications-utils # via notifications-utils
pyproj==3.2.1 pyproj==3.2.1
# via -r requirements.in # via
# -r requirements.in
# notifications-utils
python-dateutil==2.8.1 python-dateutil==2.8.1
# via # via
# awscli-cwlogs # awscli-cwlogs

View File

@@ -15,11 +15,11 @@ SKYE = [
] ]
SANTA_A = [ SANTA_A = [
[25.8890, 66.5500], [66.5500, 25.8890],
[25.8890, 66.551], [66.551, 25.8890],
[25.8910, 66.551], [66.551, 25.8910],
[25.8910, 66.5500], [66.5500, 25.8910],
[25.889, 66.55000], [66.55000, 25.889],
] ]
BURFORD = [ BURFORD = [

View File

@@ -101,7 +101,7 @@ def test_has_polygons():
assert len(scotland.polygons) == 195 assert len(scotland.polygons) == 195
assert england.polygons.as_coordinate_pairs_lat_long[0][0] == [ assert england.polygons.as_coordinate_pairs_lat_long[0][0] == [
55.811085, -2.034358 # https://goo.gl/maps/wsf2LUWzYinwydMk8 55.81108, -2.03436 # https://goo.gl/maps/HMFHGogohXdh9ggo8
] ]
@@ -276,24 +276,24 @@ def test_estimate_number_of_smartphones_for_population(
@pytest.mark.parametrize('area, expected_phones_per_square_mile', ( @pytest.mark.parametrize('area, expected_phones_per_square_mile', (
( (
# Islington (most dense in UK) # Islington (most dense in UK)
'lad20-E09000019', 21_348 'lad20-E09000019', 34_281
), ),
( (
# Cordwainer Ward (City of London) # Cordwainer Ward (City of London)
# This is higher than Islington because we inflate the # This is higher than Islington because we inflate the
# popualtion to account for daytime workers # popualtion to account for daytime workers
'wd20-E05009300', 310_674 'wd20-E05009300', 496_480
), ),
( (
# Crewe East # Crewe East
'wd20-E05008621', 2_078), 'wd20-E05008621', 3_460),
( (
# Eden (Cumbria, least dense in England) # Eden (Cumbria, least dense in England)
'lad20-E07000030', 25.57 'lad20-E07000030', 44.12
), ),
( (
# Highland (least dense in UK) # Highland (least dense in UK)
'lad20-S12000017', 4.40 'lad20-S12000017', 8.18
), ),
)) ))
def test_phone_density( def test_phone_density(
@@ -305,44 +305,40 @@ def test_phone_density(
) )
@pytest.mark.parametrize('area, expected_bleed_in_m, expected_bleed_in_degrees', ( @pytest.mark.parametrize('area, expected_bleed_in_m', (
( (
# Islington (most dense in UK) # Islington (most dense in UK)
'lad20-E09000019', 500, 0.00449 'lad20-E09000019', 500
), ),
( (
# Cordwainer Ward (City of London) # Cordwainer Ward (City of London)
# Special case because of inflated daytime population # Special case because of inflated daytime population
'wd20-E05009300', 500, 0.00449 'wd20-E05009300', 500
), ),
( (
# Crewe East # Crewe East
'wd20-E05008621', 1_752, 0.01574 'wd20-E05008621', 1_476
), ),
( (
# Eden (Cumbria, least dense in England) # Eden (Cumbria, least dense in England)
'lad20-E07000030', 4_140, 0.0372 'lad20-E07000030', 3_844
), ),
( (
# Highland (least dense in UK) # Highland (least dense in UK)
'lad20-S12000017', 5_000, 0.0449 'lad20-S12000017', 4_759
), ),
( (
# No population data available # No population data available
'test-santa-claus-village-rovaniemi-a', 1_500, 0.01347 'test-santa-claus-village-rovaniemi-a', 1_500
) )
)) ))
def test_estimated_bleed( def test_estimated_bleed(
area, expected_bleed_in_m, expected_bleed_in_degrees, area, expected_bleed_in_m
): ):
assert close_enough( assert close_enough(
broadcast_area_libraries.get_areas([area])[0].estimated_bleed_in_m, broadcast_area_libraries.get_areas([area])[0].estimated_bleed_in_m,
expected_bleed_in_m, expected_bleed_in_m,
) )
assert close_enough(
broadcast_area_libraries.get_areas([area])[0].estimated_bleed_in_degrees,
expected_bleed_in_degrees,
)
@pytest.mark.parametrize('polygon, expected_possible_overlaps, expected_count_of_phones', ( @pytest.mark.parametrize('polygon, expected_possible_overlaps, expected_count_of_phones', (
@@ -362,7 +358,7 @@ def test_estimated_bleed(
'Stoke Bishop', 'Stoke Bishop',
'Windmill Hill', 'Windmill Hill',
], ],
73_119, 72_817,
), ),
( (
SKYE, SKYE,
@@ -372,7 +368,7 @@ def test_estimated_bleed(
'Na Hearadh agus Ceann a Deas nan Loch', 'Na Hearadh agus Ceann a Deas nan Loch',
'Wester Ross, Strathpeffer and Lochalsh', 'Wester Ross, Strathpeffer and Lochalsh',
], ],
3_534, 3_413,
), ),
)) ))
def test_count_of_phones_for_custom_area( def test_count_of_phones_for_custom_area(

View File

@@ -5,7 +5,7 @@ from tests.app.broadcast_areas.custom_polygons import BRISTOL, SANTA_A, SKYE
@pytest.mark.parametrize(('simple_polygon', 'expected_wards_length'), [ @pytest.mark.parametrize(('simple_polygon', 'expected_wards_length'), [
(SKYE, 2), (SKYE, 1),
(BRISTOL, 12), (BRISTOL, 12),
(SANTA_A, 0) # does not overlap with UK (SANTA_A, 0) # does not overlap with UK
]) ])

View File

@@ -837,8 +837,8 @@ def test_broadcast_page(
'England Remove England', 'England Remove England',
'Scotland Remove Scotland', 'Scotland Remove Scotland',
], [ ], [
'An area of 200,000 square miles Will get the alert', 'An area of 100,000 square miles Will get the alert',
'An extra area of 8,000 square miles is Likely to get the alert', 'An extra area of 6,000 square miles is Likely to get the alert',
'40,000,000 phones estimated', '40,000,000 phones estimated',
]), ]),
([ ([
@@ -854,8 +854,8 @@ def test_broadcast_page(
'Penrith South Remove Penrith South', 'Penrith South Remove Penrith South',
'Penrith West Remove Penrith West', 'Penrith West Remove Penrith West',
], [ ], [
'An area of 6 square miles Will get the alert', 'An area of 4 square miles Will get the alert',
'An extra area of 20 square miles is Likely to get the alert', 'An extra area of 10 square miles is Likely to get the alert',
'9,000 to 10,000 phones', '9,000 to 10,000 phones',
]), ]),
([ ([
@@ -863,17 +863,17 @@ def test_broadcast_page(
], [ ], [
'Islington Remove Islington', 'Islington Remove Islington',
], [ ], [
'An area of 10 square miles Will get the alert', 'An area of 6 square miles Will get the alert',
'An extra area of 5 square miles is Likely to get the alert', 'An extra area of 4 square miles is Likely to get the alert',
'200,000 to 500,000 phones', '200,000 to 600,000 phones',
]), ]),
([ ([
'ctyua19-E10000019', 'ctyua19-E10000019',
], [ ], [
'Lincolnshire Remove Lincolnshire', 'Lincolnshire Remove Lincolnshire',
], [ ], [
'An area of 4,000 square miles Will get the alert', 'An area of 2,000 square miles Will get the alert',
'An extra area of 700 square miles is Likely to get the alert', 'An extra area of 500 square miles is Likely to get the alert',
'500,000 to 600,000 phones', '500,000 to 600,000 phones',
]), ]),
([ ([
@@ -882,8 +882,8 @@ def test_broadcast_page(
], [ ], [
'Lincolnshire Remove Lincolnshire', 'North Yorkshire Remove North Yorkshire', 'Lincolnshire Remove Lincolnshire', 'North Yorkshire Remove North Yorkshire',
], [ ], [
'An area of 10,000 square miles Will get the alert', 'An area of 6,000 square miles Will get the alert',
'An extra area of 2,000 square miles is Likely to get the alert', 'An extra area of 1,000 square miles is Likely to get the alert',
'1,000,000 phones estimated', '1,000,000 phones estimated',
]), ]),
)) ))
@@ -936,7 +936,7 @@ def test_preview_broadcast_areas_page(
[[7, 8], [9, 10], [11, 12]], [[7, 8], [9, 10], [11, 12]],
], ],
[ [
'An area of 700 square miles Will get the alert', 'An area of 1,000 square miles Will get the alert',
'An extra area of 1,000 square miles is Likely to get the alert', 'An extra area of 1,000 square miles is Likely to get the alert',
'Unknown number of phones', 'Unknown number of phones',
] ]
@@ -944,17 +944,17 @@ def test_preview_broadcast_areas_page(
( (
[BRISTOL], [BRISTOL],
[ [
'An area of 7 square miles Will get the alert', 'An area of 4 square miles Will get the alert',
'An extra area of 6 square miles is Likely to get the alert', 'An extra area of 3 square miles is Likely to get the alert',
'70,000 to 100,000 phones', '70,000 to 100,000 phones',
] ]
), ),
( (
[SKYE], [SKYE],
[ [
'An area of 3,000 square miles Will get the alert', 'An area of 2,000 square miles Will get the alert',
'An extra area of 800 square miles is Likely to get the alert', 'An extra area of 600 square miles is Likely to get the alert',
'4,000 phones estimated', '3,000 to 4,000 phones',
] ]
), ),
)) ))

View File

@@ -45,7 +45,7 @@ def test_simple_polygons():
# Because the areas are close to each other, the simplification # Because the areas are close to each other, the simplification
# and unioning process results in a single polygon with fewer # and unioning process results in a single polygon with fewer
# total coordinates # total coordinates
[55], [57],
] ]