def voronoiPolynomials(points,boundingPolygon ):
'''constructs a proper Voronoi diagram (out to FAR) for the points'''
assert(len(points)>2)
voronoiDiagram = scipy.spatial.Voronoi(points)
regions,vertices = voronoi_finite_polygons_2d(voronoiDiagram,radius=FAR) # read the main function in the file colorized_voronoi for an idea of what's going on here
fullPolygons = [ Polygon.Polygon([ vertices[point] for point in region]) for region in regions] # okay, nested listcomp. So what happens here is each region returned from voronoi_finite_polygons_2d becomes a Polygon library polygon. This polygon is constructed from the appropriate vertices of the voronoi diagram for said region. (A region is represented as something like [2,0,5], meaning it is a polynomial constructed from voronoi vertices 2, 0, and 5, in that order)
return [Polygon.Polygon(item & boundingPolygon) for item in fullPolygons]
def getVoronoi(shape, samples, vec):
h, w = shape
mask = np.where(vec != 0)
samples = samples[mask]
vec = vec[mask]
he = np.arange(0, h)
wi = np.arange(0, w)
Yq, Zq = np.meshgrid(wi, he)
Y_sample = Yq.flatten()[samples]
Z_sample = Zq.flatten()[samples]
points = np.column_stack((Y_sample, Z_sample))
voronoi = Voronoi(points)
regions, vertices = voronoi_finite_polygons_2d(voronoi)
reconstructed = np.zeros((w, h))
b = Polygon([(0, 0), (w - 1, 0), (w - 1, h - 1), (0, h - 1)])
reconstructed = np.zeros((w, h))
for i, region in enumerate(regions):
polygon = vertices[region]
shape = Polygon(polygon)
if not b.contains(shape):
shape = shape.intersection(b)
x, y = shape.exterior.coords.xy
row_indices = np.array(x, dtype=np.int16)
column_indices = np.array(y, dtype=np.int16)
row_min = np.amin(row_indices)
row_max = np.amax(row_indices) + 1
column_min = np.amin(column_indices)
column_max = np.amax(column_indices) + 1
enclose = np.zeros((row_max - row_min, column_max - column_min))
enclose_rows = row_indices - row_min
enclose_columns = column_indices - column_min
enclose[enclose_rows, enclose_columns] = 1
points = np.indices(enclose.shape).reshape(2, -1).T
path = Path(zip(x - row_min, y - column_min))
mask = path.contains_points(points, radius=-1e-9)
mask = mask.reshape(enclose.shape)
reconstructed_rows, reconstructed_columns = np.where(mask)
reconstructed[row_min + reconstructed_rows,
column_min + reconstructed_columns] = vec[i]
return reconstructed.T
def voronoiSegmentation(polygon, npoints):
""" returns a list of polygons that segment polygon into npoints regions. npoints must be an integer greater than two."""
assert npoints > 2
trulyRandomPoints = [randomPointWithin(polygon) for _ in range(npoints)]
points = lloydRelaxation(trulyRandomPoints, 2, boundingPolygon=polygon) # the ones we actually use, less random
voronoiDiagram = scipy.spatial.Voronoi(points)
regions, vertices = voronoi_finite_polygons_2d(
voronoiDiagram, radius=FAR
) # read the main function in the file colorized_voronoi for an idea of what's going on here
bigPolygons = [
DiplomacyPolygon([vertices[point] for point in region]) for region in regions
] # okay, nested listcomp. So what happens here is each region returned from voronoi_finite_polygons_2d becomes a Polygon library polygon. This polygon is constructed from the appropriate vertices of the voronoi diagram for said region. (A region is represented as something like [2,0,5], meaning it is a polynomial constructed from voronoi vertices 2, 0, and 5, in that order)
return [
DiplomacyPolygon(polygon & bigPolygon) for bigPolygon in bigPolygons
] # iterate over the big polygons and take the intersection (&) of each big polygon and the polgyon to be segmented. We have to cast the result to DiplomacyPolygon because the '&' operator returns cPolygon.Polygon. (I have no idea why that cast operation works. Python, man.)
def getVoronoi(shape, samples, vec):
'''
Constructs new depth image by creating Voronoi regions.
Args:
shape: Shape of the depth matrix
samples: List of flattened indices of non-NaN values
in depth matrix
vec: List of depth values at the indices
given by the previous list
* NOTE: samples and vec must be obtained from the function
create_samples.createSamples()
Returns:
matrix: New depth matrix
'''
h, w = shape
he = np.arange(0, h)
wi = np.arange(0, w)
Yq, Zq = np.meshgrid(wi, he)
Y_sample = Yq.flatten()[samples]
Z_sample = Zq.flatten()[samples]
points = np.column_stack((Y_sample, Z_sample))
voronoi = Voronoi(points)
regions, vertices = voronoi_finite_polygons_2d(voronoi)
reconstructed = np.zeros((w, h))
b = Polygon([(0, 0), (w - 1, 0), (w - 1, h - 1), (0, h - 1)])
reconstructed = np.zeros((w, h))
for i, region in enumerate(regions):
polygon = vertices[region]
shape = Polygon(polygon)
if not b.contains(shape):
shape = shape.intersection(b)
x, y = shape.exterior.coords.xy
row_indices = np.array(x, dtype=np.int16)
column_indices = np.array(y, dtype=np.int16)
row_min = np.amin(row_indices)
row_max = np.amax(row_indices) + 1
column_min = np.amin(column_indices)
column_max = np.amax(column_indices) + 1
enclose = np.zeros((row_max - row_min, column_max - column_min))
enclose_rows = row_indices - row_min
enclose_columns = column_indices - column_min
enclose[enclose_rows, enclose_columns] = 1
points = np.indices(enclose.shape).reshape(2, -1).T
path = Path(zip(x - row_min, y - column_min))
mask = path.contains_points(points, radius=-1e-9)
mask = mask.reshape(enclose.shape)
reconstructed_rows, reconstructed_columns = np.where(mask)
reconstructed[row_min + reconstructed_rows,
column_min + reconstructed_columns] = vec[i]
return reconstructed.T
df_vor["update"] = update_Z
df_vor["vor_label"] = vor_cell
df_vor.to_csv(
"{}TrainFrac{}_PairNegWgt{}_Delta{}_GroupEvt{}_df_vor.csv".format(
outputPath, tf, negative_weight, delta, group_events))
print("finish merge algorithm")
file_log.write(
"finish merge algorithm df_vor is saved into {}PairNegWgt{}_Delta{}_GroupEvt{}.csv \n"
.format(outputPath, negative_weight, delta, group_events))
# dictionary to save {label:[p1,p2,..., pN]}
label_points = df_vor.groupby('vor_label')['vor_point'].apply(list).to_dict()
regions, vertices = voronoi_finite_polygons_2d(Vor.points, Vor.vertices,
Vor.regions, Vor.ridge_vertices,
Vor.ridge_points,
Vor.point_region)
# save map
#print(data)
df_map, drop_x, drop_y = tool.save_vor_map(data.ix[(data.target.values == 1)],
regions, vertices, label_points)
df_map = df_map.reset_index(drop=True)
df_map.to_csv(
"{}TrainFrac{}_PairNegWgt{}_Delta{}_GroupEvt{}_map.csv".format(
outputPath, tf, negative_weight, delta, group_events),
columns=["mvaOutput_2lss_ttV", "mvaOutput_2lss_ttbar", "vor_label"])
# ksTest
# create train df
df_vor_train = df_vor.groupby('vor_label').sum()[[
