def knn_batch(data, k):
batch = []
_, ind = knn(data, k)
for i in range(data.shape[0]):
batch.append(data[ind[i], :])
return batch
def knn_batch(data, k):
batch = []
_, _, ind = knn(data, k)
for i in range(data.shape[0]):
batch.append(data[ind[i], :])
return batch
import numpy as np
import matplotlib.pyplot as plt
from tools.geometric import knn
data = np.loadtxt("swiss.dat")
data = data+abs(np.min(data))
n,dim = data.shape
dimh = 5
id = 2
k = 8
iters = 100000
l_r = 0.001
t = 1.0
dist,neighbors = knn.knn(data,k)
s = np.exp(-dist**2)/t
x = tf.placeholder("float",[n,dim])
nei = tf.placeholder("float",[n,k,dim])
s_w = tf.placeholder("float",[n,k])
W1 = tf.Variable(tf.truncated_normal([dim,dimh]))
b1 = tf.Variable(tf.truncated_normal([dimh]))
Wh = tf.Variable(tf.truncated_normal([dimh,id]))
bh = tf.Variable(tf.truncated_normal([id]))
W2 = tf.Variable(tf.truncated_normal([id,dim]))
b2 = tf.Variable(tf.truncated_normal([dim]))
import matplotlib.pyplot as plt
from sklearn.preprocessing import normalize
from tools.geometric import knn
iters = 100000
l_r = 0.005
dimh = 5
id = 2
k = 8
pro = 27
data = np.loadtxt("swiss.dat")
#data = normalize(data,axis=1, norm='l2')
n,dim = data.shape
_,neighbors = knn.knn(data,k)
x = tf.placeholder("float",[n,dim])
nei = tf.placeholder("float",[n,dim,k])
pW = tf.Variable(tf.truncated_normal([n,k,dim,pro]),"float")
p_r = (tf.batch_matmul(nei,tf.reshape(pW,[n,k,dim*pro])))
pI = tf.reshape(p_r,[n,dim,dim,pro])
x_com = tf.batch_matmul(tf.expand_dims(x,1),tf.reshape(pI,[n,dim,dim*pro]))
x_com = tf.reshape(x_com,[n,dim,pro])
import numpy as np
from tools.geometric import knn
data = np.random.rand(400, 3)
k = 8
n = data.shape[0]
[_, index] = knn.knn(data, k)
traverse = []
start = np.random.randint(n)
temp = index[start, 0]
for i in range(n):
c = 1
neigh_end = False
neighbors = np.random.permutation(k)
while temp in traverse:
temp = index[start, neighbors[c]]
if c == k - 1:
neigh_end = True
break
c = c + 1
if neigh_end:
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
from tools.geometric import knn
data = np.loadtxt("swiss.dat")
n,dim = data.shape
dimh = 5
reg_residual = 0.0125
id = 2
k = 5
iters = 50000
l_r = 0.125
dist,vicini = knn.knn(data,k)
x = tf.placeholder("float",[n,dim])
nei = tf.placeholder("float",[n,dim,k])
neigh_space = tf.batch_matmul(nei,tf.transpose(nei,[0,2,1]))
tan_space = tf.Variable(tf.truncated_normal([n,dim,id]),"float")
proj = (tf.batch_matmul(neigh_space,tan_space))
x_pro = tf.squeeze(tf.batch_matmul(tf.expand_dims(x,1),proj))
W1 = tf.Variable(tf.truncated_normal([id,dimh]))
b1 = tf.Variable(tf.truncated_normal([dimh]))