def manual_nodev(parcel_rejections, parcels):
df1 = parcels.to_frame(['x', 'y']).dropna(subset=['x', 'y'])
df2 = parcel_rejections.to_frame(['lng', 'lat'])
df2 = df2[parcel_rejections.state == "denied"]
df2 = df2[["lng", "lat"]] # need to change the order
ind = nearest_neighbor(df1, df2)
s = pd.Series(False, parcels.index)
s.loc[ind.flatten()] = True
return s.astype('int')
def manual_nodev(parcel_rejections, parcels):
df1 = parcels.to_frame(['x', 'y']).dropna(subset=['x', 'y'])
df2 = parcel_rejections.to_frame(['lng', 'lat'])
df2 = df2[parcel_rejections.state == "denied"]
df2 = df2[["lng", "lat"]] # need to change the order
ind = nearest_neighbor(df1, df2)
col = pd.Series(False, parcels.index)
col.loc[ind.flatten()] = True
return col.astype('int')
def costar(store, parcels):
df = pd.read_csv(os.path.join(misc.data_dir(), '2015_08_29_costar.csv'))
df["PropertyType"] = df.PropertyType.replace("General Retail", "Retail")
df = df[df.PropertyType.isin(["Office", "Retail", "Industrial"])]
df["costar_rent"] = df["Average Weighted Rent"].astype('float')
df["year_built"] = df["Year Built"].fillna(1980)
df = df.dropna(subset=["costar_rent", "Latitude", "Longitude"])
# now assign parcel id
df["parcel_id"] = nearest_neighbor(
parcels.to_frame(['x', 'y']).dropna(subset=['x', 'y']),
df[['Longitude', 'Latitude']])
return df
def costar(store, parcels):
df = pd.read_csv(os.path.join(misc.data_dir(), '2015_08_29_costar.csv'))
df["PropertyType"] = df.PropertyType.replace("General Retail", "Retail")
df = df[df.PropertyType.isin(["Office", "Retail", "Industrial"])]
df["costar_rent"] = df["Average Weighted Rent"].astype('float')
df["year_built"] = df["Year Built"].fillna(1980)
df = df.dropna(subset=["costar_rent", "Latitude", "Longitude"])
# now assign parcel id
df["parcel_id"] = nearest_neighbor(
parcels.to_frame(['x', 'y']).dropna(subset=['x', 'y']),
df[['Longitude', 'Latitude']]
)
return df
def icp(A, B, max_iterations=10000, tolerance=0.001):
assert A.shape == B.shape
n = A.shape[1]
# 构造齐次坐标
src = np.ones((n+1,A.shape[0]))
dst = np.ones((n+1,B.shape[0]))
src[:n,:] = np.copy(A.T)
dst[:n,:] = np.copy(B.T)
error = 0
for i in range(max_iterations):
distances, indices = nearest_neighbor(src[:n,:].T, dst[:n,:].T)
# print('icp distances', distances.shape[0])
T,_,_ = register_by_SVD(src[:n,:].T, dst[:n,indices].T)
# dst = dst[:,indices]
# print('icp distances', distances.shape[0])
# T,_,_ = register_by_SVD(src[:n,:].T, dst[:n,:].T)
# 变换矩阵作用于齐次坐标
src = np.dot(T, src)
# check error
new_error = np.mean(distances)
if np.abs(error - new_error) < tolerance:
break
error = new_error
# calculate final transformation
T,_,_ = register_by_SVD(A, src[:n,:].T)
return T, distances, i
def test_rough_fit(self):
total_time = 0
for i in range(TESTS_NUM):
# ax2=fig.add_subplot(111,projection='3d')
# ax2.scatter(z2,x2,y2,c='b',marker='.',s=2,linewidth=0,alpha=1,cmap='spectral')
#
# #ax.set_facecolor((0,0,0))
# ax2.axis('scaled')
# # ax.xaxis.set_visible(False)
# # ax.yaxis.set_visible(False)
# ax2.set_xlabel('X Label')
# ax2.set_ylabel('Y Label')
# ax2.set_zlabel('Z Label')
start = time.time()
T, R1, t1 = register_by_SVD(self.B, self.A)
total_time += time.time() - start
# Make C a homogeneous representation of B
C = np.ones((self.A.shape[0], 4))
C[:, 0:3] = self.B
# Transform C
C = np.dot(T, C.T).T
# view_numpy_data(self.A, C)
print(np.mean(nearest_neighbor(self.A.T, C.T))[0])
assert np.allclose(C[:, 0:3], self.A, atol=6 *
NOISE_SIGMA) # T should transform B (or C) to A
assert np.allclose(-t1, self.t, atol=6 *
NOISE_SIGMA) # t and t1 should be inverses
assert np.allclose(R1.T, self.R, atol=6 *
NOISE_SIGMA) # R and R1 should be inverses
print('rough fit time: {:.3}'.format(total_time / TESTS_NUM))
return
def dis_icp(A, B, max_iterations=10000, tolerance=0.001):
assert A.shape == B.shape
n = A.shape[1]
# print(A)
# print(n)
# 构造齐次坐标
src = np.ones((n + 1, A.shape[0]))
dst = np.ones((n + 1, B.shape[0]))
src[:n, :] = np.copy(A.T)
dst[:n, :] = np.copy(B.T)
s = src[:n, :]
d = dst[:n, :]
error = 0
match_l = []
for i in range(max_iterations):
# distances, indices = nearest_neighbor(src[:n,:].T, dst[:n,:].T)
distances, indices = nearest_neighbor(s.T, d.T)
print('indices', indices)
c = Counter(indices)
# print(c)
# print('d', distances.shape[0])
# d = d[:, indices]
max_distance = np.max(distances)
min_distance = np.min(distances)
diff = max_distance - min_distance
threshold = min_distance + diff / 2
# match_l = []
# 剔除过大值
while True:
# print(distances, threshold)
nomatch_indices = np.where(distances > threshold)[0]
print('nomatch_indices', nomatch_indices)
# break
# available_indices = []
# for m in match_indices:
# print(m)
# if Counter(m) == 1:
# available_indices.append(m)
# print('available_indices', available_indices)
# available_indices = np.array(available_indices)
# match_indices = np.array([i[0] for i in np.argwhere(distances<threshold)])
# print('match_indices', match_indices)
# match_l.append(match_indices)
# break
# print(match_indices)
# match_distances = np.minimum(distances, threshold)
if distances.shape[0] * 0.8 < nomatch_indices.shape[0]:
# print(1)
break
else:
threshold += (max_distance - threshold) / 2
# distances = distances[match_indices]
# else:
# threshold = (max_distance+threshold)/2
# match_l.append(match_indices)
# for d in match_distances:
# np.where()
# print('min_distance', min_distance)
# print('max_distance', max_distance)
# print('match_indices', match_indices)
# print('src[:n,:]', src[:n,:].shape[1], src[:n,:])
# print('threshold', threshold)
# s = src[:n, :]
# d = dst[:n, indices]
# print('match_l', len(match_l))
# for m in match_l:
# print(m.shape, m)
# distances = distances[m]
# for m in match_l:
# print(s.shape, m.shape, )
#
# s = s[:, m]
# d = d[:, m]
# s = s[:, match_indices]
# d = d[:, match_indices]
# distances = distances[match_indices]
# print(s.shape[1])
# print('src[:n,:][match_indices]', s.shape[1], s)
# print('dst[:n,indices]', dst[:n,indices])
# print('distances', distances)
T, _, _ = register_by_SVD(s[:, match_indices].T,
d[:, indices][:, match_indices].T)
# T,_,_ = register_by_SVD(src[:n,:].T, dst[:n,indices].T)
# print('T', T)
# 变换矩阵作用于齐次坐标
src = np.dot(T, src)
# print('before', distances)
print('distances', distances.shape[0])
# print('after', distances)
# check error
new_error = np.mean(distances)
if np.abs(error - new_error) < tolerance:
break
error = new_error
# calculate final transformation
T, _, _ = register_by_SVD(A, src[:n, :].T)
return T, distances, i
