用matalb、python画聚类结果图
1.utils.py
import numpy as np
import pickle as pkl
import networkx as nx
import scipy.sparse as sp
from scipy.sparse.linalg.eigen.arpack import eigsh
import sys
from scipy.sparse.linalg import norm as sparsenorm
from scipy.linalg import qr
import math def parse_index_file(filename):
"""Parse index file."""
index = []
for line in open(filename):
index.append(int(line.strip()))
return index def sample_mask(idx, l):
"""Create mask."""
mask = np.zeros(l)
mask[idx] = 1
return np.array(mask, dtype=np.bool) def load_data(dataset_str):
"""Load data."""
names = ['x', 'y', 'tx', 'ty', 'allx', 'ally', 'graph']
objects = []
for i in range(len(names)):
with open("data/ind.{}.{}".format(dataset_str, names[i]), 'rb') as f:
if sys.version_info > (3, 0):
objects.append(pkl.load(f, encoding='latin1'))
else:
objects.append(pkl.load(f)) x, y, tx, ty, allx, ally, graph = tuple(objects)
test_idx_reorder = parse_index_file(
"data/ind.{}.test.index".format(dataset_str))
test_idx_range = np.sort(test_idx_reorder) if dataset_str == 'citeseer':
# Fix citeseer dataset (there are some isolated nodes in the graph)
# Find isolated nodes, add them as zero-vecs into the right position
test_idx_range_full = range(
min(test_idx_reorder), max(test_idx_reorder)+1)
tx_extended = sp.lil_matrix((len(test_idx_range_full), x.shape[1]))
tx_extended[test_idx_range-min(test_idx_range), :] = tx
tx = tx_extended
ty_extended = np.zeros((len(test_idx_range_full), y.shape[1]))
ty_extended[test_idx_range-min(test_idx_range), :] = ty
ty = ty_extended features = sp.vstack((allx, tx)).tolil()
features[test_idx_reorder, :] = features[test_idx_range, :]
adj = nx.adjacency_matrix(nx.from_dict_of_lists(graph)) labels = np.vstack((ally, ty))
labels[test_idx_reorder, :] = labels[test_idx_range, :] idx_test = test_idx_range.tolist()
idx_train = range(len(ally)-500)
idx_val = range(len(ally)-500, len(ally)) train_mask = sample_mask(idx_train, labels.shape[0])
val_mask = sample_mask(idx_val, labels.shape[0])
test_mask = sample_mask(idx_test, labels.shape[0]) y_train = np.zeros(labels.shape)
y_val = np.zeros(labels.shape)
y_test = np.zeros(labels.shape)
y_train[train_mask, :] = labels[train_mask, :]
y_val[val_mask, :] = labels[val_mask, :]
y_test[test_mask, :] = labels[test_mask, :] return adj, features, y_train, y_val, y_test, train_mask, val_mask, test_mask def load_data_original(dataset_str):
"""Load data."""
names = ['x', 'y', 'tx', 'ty', 'allx', 'ally', 'graph']
objects = []
for i in range(len(names)):
with open("data/ind.{}.{}".format(dataset_str, names[i]), 'rb') as f:
if sys.version_info > (3, 0):
objects.append(pkl.load(f, encoding='latin1'))
else:
objects.append(pkl.load(f)) x, y, tx, ty, allx, ally, graph = tuple(objects)
test_idx_reorder = parse_index_file(
"data/ind.{}.test.index".format(dataset_str))
test_idx_range = np.sort(test_idx_reorder) if dataset_str == 'citeseer':
# Fix citeseer dataset (there are some isolated nodes in the graph)
# Find isolated nodes, add them as zero-vecs into the right position
test_idx_range_full = range(
min(test_idx_reorder), max(test_idx_reorder)+1)
tx_extended = sp.lil_matrix((len(test_idx_range_full), x.shape[1]))
tx_extended[test_idx_range-min(test_idx_range), :] = tx
tx = tx_extended
ty_extended = np.zeros((len(test_idx_range_full), y.shape[1]))
ty_extended[test_idx_range-min(test_idx_range), :] = ty
ty = ty_extended # features (2708,1433) labels (2708,7)
features = sp.vstack((allx, tx)).tolil()
features[test_idx_reorder, :] = features[test_idx_range, :]
adj = nx.adjacency_matrix(nx.from_dict_of_lists(graph)) labels = np.vstack((ally, ty))
labels[test_idx_reorder, :] = labels[test_idx_range, :] idx_test = test_idx_range.tolist()
idx_train = range(len(y))
idx_val = range(len(y), len(y)+500) train_mask = sample_mask(idx_train, labels.shape[0])
val_mask = sample_mask(idx_val, labels.shape[0])
test_mask = sample_mask(idx_test, labels.shape[0]) y_train = np.zeros(labels.shape)
y_val = np.zeros(labels.shape)
y_test = np.zeros(labels.shape)
y_train[train_mask, :] = labels[train_mask, :]
y_val[val_mask, :] = labels[val_mask, :]
y_test[test_mask, :] = labels[test_mask, :] return adj, features, y_train, y_val, y_test, train_mask, val_mask, test_mask def sparse_to_tuple(sparse_mx):
"""Convert sparse matrix to tuple representation."""
def to_tuple(mx):
if not sp.isspmatrix_coo(mx):
mx = mx.tocoo()
coords = np.vstack((mx.row, mx.col)).transpose()
values = mx.data
shape = mx.shape
return coords, values, shape if isinstance(sparse_mx, list):
for i in range(len(sparse_mx)):
sparse_mx[i] = to_tuple(sparse_mx[i])
else:
sparse_mx = to_tuple(sparse_mx) return sparse_mx def nontuple_preprocess_features(features):
"""Row-normalize feature matrix and convert to tuple representation"""
rowsum = np.array(features.sum(1))
r_inv = np.power(rowsum, -1).flatten()
r_inv[np.isinf(r_inv)] = 0.
r_mat_inv = sp.diags(r_inv)
features = r_mat_inv.dot(features)
return features def preprocess_features(features):
"""Row-normalize feature matrix and convert to tuple representation"""
rowsum = np.array(features.sum(1))
r_inv = np.power(rowsum, -1).flatten()
r_inv[np.isinf(r_inv)] = 0.
r_mat_inv = sp.diags(r_inv)
features = r_mat_inv.dot(features)
return sparse_to_tuple(features) def normalize_adj(adj):
"""Symmetrically normalize adjacency matrix."""
adj = sp.coo_matrix(adj)
rowsum = np.array(adj.sum(1))
d_inv_sqrt = np.power(rowsum, -0.5).flatten()
d_inv_sqrt[np.isinf(d_inv_sqrt)] = 0.
d_mat_inv_sqrt = sp.diags(d_inv_sqrt)
return adj.dot(d_mat_inv_sqrt).transpose().dot(d_mat_inv_sqrt).tocoo() def nontuple_preprocess_adj(adj):
""" 返回对称归一化的邻接矩阵 type:csr """
adj_normalized = normalize_adj(sp.eye(adj.shape[0]) + adj)
return adj_normalized.tocsr() def column_prop(adj):
""" detail reference:
https://docs.scipy.org/doc/scipy/reference/generated/scipy.sparse.linalg.norm.html#scipy.sparse.linalg.norm
等价于array形式:
arr = np.array([[1, 4, 7],
[2, 5, 8],
[3, 6, 9]])
column_norm = np.linalg.norm(arr, axis=0) 对每一列求二范数
print column_norm
>>> [ 3.74165739 8.77496439 13.92838828] => [sqrt(1^2+2^2+3^2) sqrt(4^2+5^2+6^2) sqrt(7^2+8^2+9^2)]
norm_sum = sum(column_norm) # 归一化
print(column_norm/norm_sum)
>>> [0.14148822 0.33181929 0.52669249]
"""
column_norm = sparsenorm(adj, axis=0)
# column_norm = pow(sparsenorm(adj, axis=0),2)
norm_sum = sum(column_norm)
return column_norm/norm_sum def mix_prop(adj, features, sparseinputs=False):
adj_column_norm = sparsenorm(adj, axis=0)
if sparseinputs:
features_row_norm = sparsenorm(features, axis=1)
else:
features_row_norm = np.linalg.norm(features, axis=1)
mix_norm = adj_column_norm*features_row_norm norm_sum = sum(mix_norm)
return mix_norm / norm_sum def preprocess_adj(adj):
"""Preprocessing of adjacency matrix for simple GCN model and conversion to tuple representation."""
adj_normalized = normalize_adj(sp.eye(adj.shape[0]) + adj)
return sparse_to_tuple(adj_normalized) def dense_lanczos(A, K):
q = np.random.randn(A.shape[0], )
Q, sigma = lanczos(A, K, q)
A2 = np.dot(Q[:, :K], np.dot(sigma[:K, :K], Q[:, :K].T))
return sp.csr_matrix(A2) def sparse_lanczos(A, k):
q = sp.random(A.shape[0], 1)
n = A.shape[0]
Q = sp.lil_matrix(np.zeros((n, k+1)))
A = sp.lil_matrix(A) Q[:, 0] = q/sparsenorm(q) alpha = 0
beta = 0 for i in range(k):
if i == 0:
q = A*Q[:, i]
else:
q = A*Q[:, i] - beta*Q[:, i-1]
alpha = q.T*Q[:, i]
q = q - Q[:, i]*alpha
q = q - Q[:, :i]*Q[:, :i].T*q # full reorthogonalization
beta = sparsenorm(q)
Q[:, i+1] = q/beta
print(i) Q = Q[:, :k] Sigma = Q.T*A*Q
A2 = Q[:, :k]*Sigma[:k, :k]*Q[:, :k].T
return A2
# return Q, Sigma def dense_RandomSVD(A, K):
G = np.random.randn(A.shape[0], K)
B = np.dot(A, G)
Q, R = qr(B, mode='economic')
M = np.dot(Q, np.dot(Q.T, A))
return sp.csr_matrix(M) def construct_feed_dict(features, support, labels, labels_mask, placeholders):
"""Construct feed dictionary."""
feed_dict = dict()
feed_dict.update({placeholders['labels']: labels})
feed_dict.update({placeholders['labels_mask']: labels_mask})
feed_dict.update({placeholders['features']: features})
feed_dict.update({placeholders['support'][i]: support[i]
for i in range(len(support))})
feed_dict.update({placeholders['num_features_nonzero']: features[1].shape})
return feed_dict def chebyshev_polynomials(adj, k):
"""Calculate Chebyshev polynomials up to order k. Return a list of sparse matrices (tuple representation)."""
print("Calculating Chebyshev polynomials up to order {}...".format(k)) adj_normalized = normalize_adj(adj)
laplacian = sp.eye(adj.shape[0]) - adj_normalized
largest_eigval, _ = eigsh(laplacian, 1, which='LM')
scaled_laplacian = (
2. / largest_eigval[0]) * laplacian - sp.eye(adj.shape[0]) t_k = list()
t_k.append(sp.eye(adj.shape[0]))
t_k.append(scaled_laplacian) def chebyshev_recurrence(t_k_minus_one, t_k_minus_two, scaled_lap):
s_lap = sp.csr_matrix(scaled_lap, copy=True)
return 2 * s_lap.dot(t_k_minus_one) - t_k_minus_two for i in range(2, k+1):
t_k.append(chebyshev_recurrence(t_k[-1], t_k[-2], scaled_laplacian)) return sparse_to_tuple(t_k) def view_bar(message, num, total, loss, train_acc, val_acc, test_acc, times):
rate = num / total
rate_num = int(rate * 40)
rate_nums = math.ceil(rate * 100)
r = '\r%s:[%s%s]%d%%\t%d/%d - loss:%.3f - train_acc:%.3f - val_acc:%.3f - test_acc:%.3f - time:%.2fs' % (message,
"=" * rate_num,
" " *
(40 - rate_num),
rate_nums,
num,
total,
loss,
train_acc,
val_acc,
test_acc,
times)
sys.stdout.write(r)
sys.stdout.flush()
2.layers.py
import tensorflow as tf
from model.inits import *
from model.utils import * def sparse_dropout(x, keep_prob, noise_shape):
"""Dropout for sparse tensors."""
random_tensor = keep_prob
random_tensor += tf.random.uniform(noise_shape)
dropout_mask = tf.cast(tf.floor(random_tensor), dtype=tf.bool)
pre_out = tf.sparse.retain(x, dropout_mask)
return pre_out * (1./keep_prob) def dot(x, y, sparse=False):
"""Wrapper for tf.matmul (sparse vs dense)."""
if sparse:
res = tf.sparse.sparse_dense_matmul(x, y)
else:
res = tf.matmul(x, y)
return res class Layer(object):
def __init__(self, **kwargs):
self.vars = {} def _call(self, params):
"""implement the layer operation """
return params def __call__(self, params):
outputs = self._call(params)
return outputs class Dense(Layer):
def __init__(self, name, input_dim, output_dim, dropout=0.0,
sparse_inputs=False, act=tf.nn.relu, bias=False, **kwargs):
super(Dense, self).__init__(**kwargs)
self.name = name
self.dropout = dropout
self.act = act
self.out_dim = output_dim # bool
self.bias = bias
self.sparse_inputs = sparse_inputs # params['features'] sparse or not
# define params
self.vars['weight'] = glorot([input_dim, output_dim],
name=self.name+'weight')
if self.bias:
self.vars['bias'] = zeros([output_dim],
name=self.name+'bias') def _call(self, params):
"""params: features support """
x = params['features'] # dropout x
if self.sparse_inputs:
x = sparse_dropout(x, 1-self.dropout,
params['num_features_nonzero'])
else:
x = tf.nn.dropout(x, 1-self.dropout) output = dot(x, self.vars['weight'], sparse=self.sparse_inputs) if self.bias:
output += self.vars['bias']
return self.act(output) class GraphConvolution(Layer):
def __init__(self, name, input_dim, output_dim, dropout=0.0,
sparse_inputs=False, act=tf.nn.relu, bias=False, **kwargs):
super(GraphConvolution, self).__init__(**kwargs)
self.name = name
self.dropout = dropout
self.act = act
self.out_dim = output_dim # bool
self.bias = bias
self.sparse_inputs = sparse_inputs # params['features'] sparse or not # define params
self.vars['weight'] = glorot([input_dim, output_dim],
name=self.name+'weight')
if self.bias:
self.vars['bias'] = zeros([output_dim],
name=self.name+'bias') def _call(self, params):
x = params['features']
support = params['support']
# dropout x
if self.sparse_inputs:
x = sparse_dropout(x, 1-self.dropout,
params['num_features_nonzero'])
else:
x = tf.nn.dropout(x, 1-self.dropout) # x is sparse
pre_sup = dot(x, self.vars['weight'],
sparse=self.sparse_inputs) # x.dot(w) # support is sparse
output = dot(support, pre_sup, sparse=True) # Axw if self.bias:
output += self.vars['bias']
return self.act(output)
3.models.py
from model.layers import *
from model.metrics import *
import numpy as np class Model(object):
def __init__(self):
self.vars = []
self.layers = [] def forward(self):
raise NotImplementedError def _update(self):
raise NotImplementedError def _loss(self):
raise NotImplementedError class GCN(Model):
""" kipf & welling """ def __init__(self, placeholders, sparse_inputs=False):
super(GCN, self).__init__()
self.input_dim = placeholders['in_dim']
self.hid_dim = placeholders['hid_dim']
self.output_dim = placeholders['out_dim']
self.weight_decay = placeholders['weight_decay']
self.dropout = placeholders['dropout']
self.lr = placeholders['lr']
self.sparse_inputs = sparse_inputs # params['features'] sparse or not
self.build() def build(self):
"""构建2 layer GCN, 并保存参数到vars中
第一层需要sparse_inputs 如果features是sparse,则features x W 要sparsedot
第二层不需要sparse_inputs 因为H0是dense的""" self.layers.append(GraphConvolution(name='GCN_0',
input_dim=self.input_dim,
output_dim=self.hid_dim,
dropout=self.dropout,
act=tf.nn.relu,
sparse_inputs=self.sparse_inputs))
self.layers.append(GraphConvolution(name='GCN_1',
input_dim=self.hid_dim,
output_dim=self.output_dim,
dropout=self.dropout,
act=lambda x: x,
sparse_inputs=False)) for layer in self.layers:
for var in layer.vars.values():
self.vars.append(var) self.op = tf.optimizers.Adam(self.lr) def forward(self, params):
# params: features support
for layer in self.layers:
hidden = layer(params)
params.update({'features': hidden})
return hidden def _update(self, tape, loss):
gradients = tape.gradient(target=loss, sources=self.vars)
self.op.apply_gradients(zip(gradients, self.vars)) def _loss(self, outputs, labels, labels_mask):
loss = masked_softmax_cross_entropy(outputs,
labels,
labels_mask)
for var in self.vars:
loss += self.weight_decay*tf.nn.l2_loss(var)
# Cross entropy error
return loss class FASTGCN(Model): def __init__(self, placeholders, sparse_inputs=False):
super(FASTGCN, self).__init__()
self.input_dim = placeholders['in_dim']
self.hid_dim = placeholders['hid_dim']
self.output_dim = placeholders['out_dim']
self.weight_decay = placeholders['weight_decay']
self.dropout = placeholders['dropout']
self.lr = placeholders['lr']
self.sparse_inputs = sparse_inputs # params['features'] sparse or not
self.build() def build(self):
"""构建2 layer GCN, 并保存参数到vars中
第一层需要sparse_inputs 如果features是sparse,则features x W 要sparsedot
第二层不需要sparse_inputs 因为H0是dense的""" self.layers.append(Dense(name='Dense_0',
input_dim=self.input_dim,
output_dim=self.hid_dim,
dropout=self.dropout,
act=tf.nn.relu,
sparse_inputs=self.sparse_inputs)) self.layers.append(GraphConvolution(name='GCN_1',
input_dim=self.hid_dim,
output_dim=self.output_dim,
dropout=self.dropout,
act=lambda x: x,
sparse_inputs=False)) for layer in self.layers:
for var in layer.vars.values():
self.vars.append(var) self.op = tf.optimizers.Adam(self.lr) def forward(self, params):
# params: features support
for layer in self.layers:
hidden = layer(params)
params.update({'features': hidden})
return hidden def _update(self, tape, loss):
gradients = tape.gradient(target=loss, sources=self.vars)
self.op.apply_gradients(zip(gradients, self.vars)) def _loss(self, outputs, labels):
# batch outputs
loss = softmax_cross_entropy(outputs,
labels)
for var in self.vars:
loss += self.weight_decay*tf.nn.l2_loss(var)
# Cross entropy error
return loss
4.metrics.py
import tensorflow as tf def masked_softmax_cross_entropy(preds, labels, mask):
"""Softmax cross-entropy loss with masking."""
loss = tf.nn.softmax_cross_entropy_with_logits(logits=preds, labels=labels)
mask = tf.cast(mask, dtype=tf.float32)
mask /= tf.reduce_mean(mask)
loss *= mask # element-wise 把其它节点遮掉,只用train nodes来训练
return tf.reduce_mean(loss) def masked_accuracy(preds, labels, mask):
"""Accuracy with masking."""
correct_prediction = tf.equal(tf.argmax(preds, 1), tf.argmax(labels, 1))
accuracy_all = tf.cast(correct_prediction, tf.float32)
mask = tf.cast(mask, dtype=tf.float32)
mask /= tf.reduce_mean(mask)
accuracy_all *= mask
return tf.reduce_mean(accuracy_all) def softmax_cross_entropy(preds, labels):
loss = tf.nn.softmax_cross_entropy_with_logits(logits=preds, labels=labels)
return tf.reduce_mean(loss) def accuracy(preds, labels):
correct_prediction = tf.equal(tf.argmax(preds, 1), tf.argmax(labels, 1))
accuracy_all = tf.cast(correct_prediction, tf.float32)
return tf.reduce_mean(accuracy_all)
5.inits.py
import tensorflow as tf
import numpy as np def uniform(shape, scale=0.05, name=None):
"""Uniform init."""
initial = tf.random.uniform(
shape, minval=-scale, maxval=scale, dtype=tf.float32)
return tf.Variable(initial, name=name) def glorot(shape, name=None):
"""Glorot & Bengio (AISTATS 2010) init."""
init_range = np.sqrt(6.0/(shape[0]+shape[1]))
initial = tf.random.uniform(
shape, minval=-init_range, maxval=init_range, dtype=tf.float32)
return tf.Variable(initial, name=name) def zeros(shape, name=None):
"""All zeros."""
initial = tf.zeros(shape, dtype=tf.float32)
return tf.Variable(initial, name=name) def ones(shape, name=None):
"""All ones."""
initial = tf.ones(shape, dtype=tf.float32)
return tf.Variable(initial, name=name)
6.main.py
from model.utils import *
from model.metrics import *
from model.models import FASTGCN
import tensorflow as tf
from scipy.sparse import csr_matrix
import time def iterate_minibatches_listinputs(inputs, batchsize, shuffle=False):
""" 对inputs: [normADJ_train, y_train]进行切片"""
assert inputs is not None
numSamples = inputs[0].shape[0] # 训练节点个数
if shuffle:
indices = np.arange(numSamples)
np.random.shuffle(indices)
""" 步长为batchsize,如果需要shuffle 则对indices进行切片.否则直接按顺序切片. """
for start_idx in range(0, numSamples - batchsize + 1, batchsize):
if shuffle:
excerpt = indices[start_idx:start_idx + batchsize]
else:
"""slice(start, stop, step)
=> start -- 起始位置 stop -- 结束位置 step -- 间距 """
excerpt = slice(start_idx, start_idx + batchsize)
""" print(len(excerpt))
>>> 250 (batch_size)
[input[excerpt] for input in inputs] =>
inputs由normADJ_train和y_train组成,
input相当于normADJ_train或y_train,即分别对二者切片 """
yield [input[excerpt] for input in inputs] def construct_params(features, support):
params = dict()
params.update({'support': tf.cast(tf.SparseTensor(
support[0], support[1], support[2]), tf.float32)})
params.update({'features': tf.cast(tf.SparseTensor(
features[0], features[1], features[2]), tf.float32)})
params.update({'num_features_nonzero': features[1].shape})
return params if __name__ == "__main__": (adj, features,
y_train, y_val, y_test,
train_mask, val_mask, test_mask) = load_data('cora')
"""np.where 找出mask中为true的下标 """
train_index = np.where(train_mask)[0]
y_train = y_train[train_index]
val_index = np.where(val_mask)[0]
y_val = y_val[val_index]
test_index = np.where(test_mask)[0]
y_test = y_test[test_index] """print(adj_train.shape, y_train.shape, y_test.shape, y_val.shape)
>>> (1208, 1208) (1208, 7) (1000, 7) (500, 7) """ train_val_index = np.concatenate([train_index, val_index], axis=0)
train_test_idnex = np.concatenate([train_index, test_index], axis=0) """preprocessing csr adj && features:
print(csr_normADJ_train.shape, csr_normADJ_val.shape, csr_normADJ_test.shape)
print(csr_features_train.shape,csr_features_val.shape, csr_features_test.shape)
>>> (1208, 1208) (1708, 1708) (2208, 2208)
>>> (1208, 1433) (1708, 1433) (2208, 1433)"""
csr_normADJ_train = nontuple_preprocess_adj(
adj[train_index, :][:, train_index]) # (1208, 1208)
csr_normADJ_val = nontuple_preprocess_adj(
adj[train_val_index, :][:, train_val_index]) # (1708, 1708)
csr_normADJ_test = nontuple_preprocess_adj(
adj[train_test_idnex, :][:, train_test_idnex]) # (2208, 2208) csr_features_train = nontuple_preprocess_features(
features[train_index])
csr_features_val = nontuple_preprocess_features(
features[train_val_index])
csr_features_test = nontuple_preprocess_features(
features[train_test_idnex]) y_val = np.vstack((y_train, y_val))
y_test = np.vstack((y_train, y_test))
""" 计算每个节点的概率: q(u) = ||A(: , u)||^2 / sum(||A(: , v)||^2) """
p0 = column_prop(csr_normADJ_train) epochs = 200
samplesize = 50 placeholders = {'in_dim': 1433,
'hid_dim': 32,
'out_dim': 7,
'weight_decay': 5e-4,
'dropout': 0.5,
'lr': 0.01} dense_AXfeatures_train = csr_normADJ_train.dot(
csr_features_train.todense())
dense_AXfeatures_val = csr_normADJ_val.dot(
csr_features_val.todense())
dense_AXfeatures_test = csr_normADJ_test.dot(
csr_features_test.todense()) """print(dense_AXfeatures_train.shape, dense_AXfeatures_val.shape, dense_AXfeatures_test.shape)
>>> (1208, 1433) (1708, 1433) (2208, 1433)""" # transform into tuple
tuple_AXfeatures_train = sparse_to_tuple(
csr_matrix(dense_AXfeatures_train))
tuple_AXfeatures_val = sparse_to_tuple(
csr_matrix(dense_AXfeatures_val))
tuple_AXfeatures_test = sparse_to_tuple(
csr_matrix(dense_AXfeatures_test)) model = FASTGCN(placeholders, sparse_inputs=True)
cost_val = []
t = time.time()
for epoch in range(epochs):
for batch in iterate_minibatches_listinputs([csr_normADJ_train, y_train], batchsize=1024, shuffle=True):
[normADJ_batch, y_train_batch] = batch """get support_batch(tuple), features_inputs(tuple). """
if samplesize == -1:
support_batch = sparse_to_tuple(normADJ_batch)
features_inputs = sparse_to_tuple(
csr_matrix(dense_AXfeatures_train))
else:
distr = np.nonzero(np.sum(normADJ_batch, axis=0))[1]
if samplesize > len(distr):
q1 = distr
else:
q1 = np.random.choice(
distr, samplesize, replace=False, p=p0[distr]/sum(p0[distr])) # 根据概率p0选出rank1个顶点
support_batch = sparse_to_tuple(normADJ_batch[:, q1].dot(
sp.diags(1.0 / (p0[q1] * samplesize))))
if len(support_batch[1]) == 0:
continue
features_inputs = sparse_to_tuple(
csr_matrix(dense_AXfeatures_train[q1, :])) """print(support_batch[2], features_inputs[2])
>>> (200, 50) (50, 1433)""" # support_batch used at 2nd layer
params = construct_params(
features_inputs, support_batch) with tf.GradientTape() as tape:
logits = model.forward(params)
loss = model._loss(logits, y_train_batch)
model._update(tape, loss) train_logits = model.forward(construct_params(
tuple_AXfeatures_train, sparse_to_tuple(csr_normADJ_train)))
train_acc = accuracy(train_logits, y_train) val_logits = model.forward(construct_params(
tuple_AXfeatures_val, sparse_to_tuple(csr_normADJ_val)))
val_acc = accuracy(val_logits, y_val) test_logits = model.forward(construct_params(
tuple_AXfeatures_test, sparse_to_tuple(csr_normADJ_test)))
test_acc = accuracy(test_logits, y_test) view_bar('epoch', epoch+1, epochs, loss, train_acc,
val_acc, test_acc, time.time()-t)
用matalb、python画聚类结果图的更多相关文章
- python 画广东省等压线图
最近开发时要实现一个业务逻辑: 调用中国气象数据网API接口获取广东省实时气象数据 根据数据,基于广东省地图渲染等压线图 最终效果图是这样的: 首先是获取实时气压数据,由于中国气象数据网每次只能获得3 ...
- python画箱线图
# -*- coding: utf-8 -*- """ Created on Wed Jun 14 13:00:11 2017 @author: Miao "& ...
- Python matplot画散列图
同matlab一样,matplot也可画散列图scatter. import numpy as np import matplotlib.pyplot as plt #fig = plt.figure ...
- 利用Tkinter和matplotlib两种方式画饼状图
当我们学习python的时候,总会用到一些常用的模块,接下来我就详细讲解下利用两种不同的方式画饼状图.首先利用[Tkinter]中的canvas画布来画饼状图: from tkinter import ...
- 沉淀再出发:用python画各种图表
沉淀再出发:用python画各种图表 一.前言 最近需要用python来做一些统计和画图,因此做一些笔记. 二.python画各种图表 2.1.使用turtle来画图 import turtle as ...
- 利用python画出动态高优先权优先调度
之前写过一个文章. 利用python画出SJF调度图 动态高度优先权优先调度 动态优先权调度算法,以就绪队列中各个进程的优先权作为进程调度的依据.各个进程的优先权在创建进程时所赋予,随着进程的推进或其 ...
- 用Python画如此漂亮的专业插图 ？简直So easy！
本文整理自知乎问答,仅用于学术分享,著作权归作者所有.如有侵权,请联系我删文处理.多多转发,多多学习! 方法一 强烈推荐 Python 的绘图模块 matplotlib: python plottin ...
- 用python画xy散点图
import matplotlib.pyplot as plt plt.plot([1,2,3],[4,5,6],'ro') plt.show()#这个智障的编辑器 这样的话,就可以画一个散点图,图中 ...
- 四步轻松实现用Visio画UML类图
本节和大家一起学习一下用Visio画UML类图的方法,主要有四个步骤,这里和大家分享一下,相信通过本节的学习,你对Visio画UML类图的步骤一定会有所了解. 用Visio画UML类图 对于画类图的工 ...
随机推荐
- Navicat连接Mysql报错：Client does not support authentication protocol requested by server；
Navicat连接Mysql报错:Client does not support authentication protocol requested by server: 刚安装Mysql,想用Nav ...
- C16记技术服务支持
1.首页笔记:可以通过点击加号添加笔记的分类,还可以查看最近的10条笔记 2.图形:可以通过查看笔记的比例来分析自己最近的状况 3.快速添加笔记:点击添加笔记能够直接进行笔记 4.时间轴:能够通过文字 ...
- hive防止数据被误删除
1.HDFS层面开启trash功能(fs.trash.interval) 被删除的数据在HDFS中的/user/$USER/.Trash目录中,开启这个功能秩序要将配置属性fs.trash.inter ...
- js循环出相同name，不同id的按钮，对其进行点击回复操作
function getseat(){ var option= "<button class='btn'style='margin:5px;' onclick='onclickSeat ...
- java 对同一个文件进行读写操作
同一个文件是不可以进行同时的读写的,因为我们写入文件会覆盖原文件的,如果这样,对于同一文件来来说,文件发生覆盖,无法进行下次读取 当然,对于两个不同的文件,可以一边读一边写的操作 题目:一个文本中存储 ...
- A The Empire Age
1月28日 Description 帝国时代3是一款十分刺激的RTS游戏.你需要控制自己的一块殖民地,发展殖民地的经济和军事,最终打败其他殖民地.小L是这个游戏的狂热爱好者.一次小L打算打AI试试身手 ...
- 2018-No.7-SicnuCtf
5月份的比赛现在才有时间在博客贴出来,这是我第二次出题了,第一次是上届的初赛,这次是决赛的题. 签到(base_or_base) 解压得到两个文件小明.zip和......txt,根据文件名提示猜测是 ...
- 常见Soc平台图形内存管理学习笔记
硬件编解码.硬件图像scale等过程,是在专有的硬件单元里进行,其使用的内存也是专有的内存,这种内存多是SoC中图形内存.如此方便与硬件加速图形渲染.图像显示.硬件图像加速处理等功能相交互. 上述过程 ...
- linux压缩、解压缩和归档工具
linux基础之压缩.解压缩和归档工具 1.压缩工具 基本介绍 为了减少文件的原来的文件大小而过多的浪费磁盘的存储空间,我们使用压缩后多文件进行存储 压缩工具的介绍 compress:把文件压缩成以. ...
- TensorFlow，Keras限制GPU显存
运行TensorFlow程序会占用过多的显卡比例,多人共同使用GPU的时候,会造成后面的人无法运行程序. 一.TensorFlow 1.预加载比例限制 tf_config = tensorflow.C ...