def _accuracy(self):
if self.model_config['Model'] in [11, 13, 14, 15]:
if self.model_config['Model'] in [
14, 15
] or self.model_config['Model11'] == 'weighted':
sample2label = self.placeholders['sample2label']
unsoftmaxed = tf.matmul(
self.outputs, sample2label) / tf.reduce_sum(
sample2label, axis=0, keep_dims=True)
self.accuracy = masked_accuracy(
unsoftmaxed, self.placeholders['labels'],
self.placeholders['labels_mask'])
elif self.model_config['Model11'] == 'nearest':
sample2label = self.placeholders['sample2label']
outputs = tf.one_hot(
tf.argmax(self.outputs, axis=1),
depth=self.placeholders['label_per_sample'].get_shape(
).as_list()[1])
unsoftmaxed = tf.matmul(outputs, sample2label)
self.accuracy = masked_accuracy(
unsoftmaxed, self.placeholders['labels'],
self.placeholders['labels_mask'])
else:
raise ValueError(
"model_config['Model11'] should be either 'weighted' or 'nearest'"
)
else:
self.accuracy = masked_accuracy(self.outputs,
self.placeholders['labels'],
self.placeholders['labels_mask'])
def _accuracy(self):
if self.is_gcn:
#self.merged = tf.math.add(tf.math.scalar_mul(self.outputs_weight, self.outputs),
#tf.math.scalar_mul(self.outs_weight, self.outs))
self.outs_for_graph = tf.nn.softmax(self.outs)
self.accuracy = masked_accuracy(tf.nn.softmax(self.outs),
self.placeholders['labels'],
self.placeholders['labels_mask'])
self.mo_accuarcy = masked_accuracy(
self.outputs_weight * tf.nn.softmax(self.outs) +
tf.nn.softmax(self.outputs), self.placeholders['labels'],
self.placeholders['labels_mask'])
else:
self.accuracy = masked_accuracy(self.outputs,
self.placeholders['labels'],
self.placeholders['labels_mask'])
def _accuracy_of_class(self):
self.accuracy_of_class = [
masked_accuracy(
self.outputs, self.placeholders['labels'],
self.placeholders['labels_mask'] *
self.placeholders['labels'][:, i])
for i in range(self.placeholders['labels'].shape[1])
]
def _accuracy_of_class(self):
if self.model_config['Model'] in [11, 13, 14, 15]:
sample2label = self.placeholders['sample2label']
unsoftmaxed = tf.matmul(self.outputs,
sample2label) / tf.reduce_sum(
sample2label, axis=0, keep_dims=True)
self.accuracy_of_class = [
masked_accuracy(
unsoftmaxed, self.placeholders['labels'],
self.placeholders['labels_mask'] *
self.placeholders['labels'][:, i])
for i in range(self.placeholders['labels'].shape[1])
]
else:
self.accuracy_of_class = [
masked_accuracy(
self.outputs, self.placeholders['labels'],
self.placeholders['labels_mask'] *
self.placeholders['labels'][:, i])
for i in range(self.placeholders['labels'].shape[1])
]
def _accuracy_of_class(self):
# if self.is_gcn:
#self.merged = tf.math.add(tf.math.scalar_mul(self.outputs_weight, self.outputs),
# tf.math.scalar_mul(self.outs_weight, self.outs))
# self.accuracy_of_class = [masked_accuracy(self.merged, self.placeholders['labels'],
# self.placeholders['labels_mask'] * self.placeholders['labels'][:, i])
# for i in range(self.placeholders['labels'].shape[1])]
#else:
self.accuracy_of_class = [
masked_accuracy(
self.outputs, self.placeholders['labels'],
self.placeholders['labels_mask'] *
self.placeholders['labels'][:, i])
for i in range(self.placeholders['labels'].shape[1])
]
def baseline_predict(y_train,
y_test,
y_val,
train_mask,
test_mask,
val_mask,
label_samples=None):
"""Get baseline random predictions. If 'label_samples' is used then the predictor will take into account the frequencies of the labels in the file."""
sess = tf.Session()
acc_vector = []
for t in [(y_train, train_mask), (y_val, val_mask), (y_test, test_mask)]:
if label_samples == None:
y_rnd = random_pred(t[0])
else:
y_rnd = weighted_random_pred(label_samples, len(t[0]))
acc = sess.run(masked_accuracy(y_rnd, t[0], t[1]))
acc_vector.append(acc)
return acc_vector
print("Base model optimization Finished!")
feed_dict = construct_feed_dict(features, support, y_train, train_mask,
placeholders)
preds = sess.run(model.outputs, feed_dict=feed_dict)
print("Calculating neighorhood distance matrix...")
d2 = neighborhood_distance_matrix(adj, preds, FLAGS.dataset)
print("Calculating weighted distance matrix...")
d = weighted_distance_matrix(d1, d2)
print("Calculating the MAP estimate...")
map_support = map_estimate(adj, d, FLAGS.dataset)
map_support = [preprocess_adj(map_support)]
print("Sampling weights by running a GCN with MC dropout...")
final_model = model_func(placeholders, input_dim=features[2][1], logging=True)
sess.run(tf.global_variables_initializer())
final_pred_soft, final_pred = train_model(FLAGS, sess, final_model, features,
map_support, y_train, y_val,
train_mask, val_mask, placeholders)
test_acc = sess.run(masked_accuracy(final_pred_soft, y_test, test_mask))
print("Test accuracy: ", "{:.5f}".format(test_acc))
gcn_test_acc = sess.run(masked_accuracy(gcn_pred_soft, y_test, test_mask))
print("GCN Test accuracy: ", "{:.5f}".format(gcn_test_acc))
def _accuracy(self):
self.accuracy = masked_accuracy(self.outputs,
self.placeholders['labels'],
self.placeholders['labels_mask'])