我正在尝试针对5种不同的拆分数据训练TensorFlow模型。对于每个拆分，我都保存模型，以后再将其还原以进行预测，但是它不会获得应提供的实际预测和准确性（对于以后的还原模型）。 我正在使用一个循环进行训练，每次我从不同的分割中加载TensorFlow并使用不同的名称保存每个模型时。 tf.get_default_graph（）是否有问题？ 我读了文件。提到它不是线程安全的。
这是我的代码：
将my_cnn_tensorflow导入为cnn
for k_splits in range(K, K + 1):
#kf = KFold(k_splits, shuffle=False, random_state=42)
kf = RepeatedKFold(k_splits, n_repeats=1, random_state=42)
for train_index, test_index in kf.split(X, Y_P2):
d_range = int(len(X_train) * TL_percentage)
k_run += 1
Dx_train, Dx_test = Dx[train_index], Dx[test_index]
y_train, y_test = Y_P2[train_index], Y_P2[test_index]
model = cnn.DLSpMVModel(Dx_train, y_train, Dx_test, y_test, classes, TL_percentage,platform1_name, k_run)
model.training()
y_pred = model.testing()
accuracy = accuracy_score(y_test, y_pred) * 100
class DLSpMVModel(object):
def __init__(self, train_data, y_train, test_data, y_test, classes, TL_percentage, platform,
ksplit):
self.RES = 0
self.mean = 0
self.std = 1
self.train = load_data(train_data, y_train, classes)
self.test = load_data(test_data, y_test, classes)
def training(self):
print("Model is in training mode")
assert self.train is not None and self.test is not None, "data not loaded"
with tf.name_scope('upper'):
x, y_, h_fc1_upper = single_net(self.RES, self.classes)
with tf.name_scope('lower'):
x2, y2_, h_fc1_lower = single_net(self.RES, self.classes)
h_fc1 = tf.concat([h_fc1_upper, h_fc1_lower], axis=1)
# [-1, 512 * 2]
with tf.name_scope('dropout'):
keep_prob = tf.placeholder(tf.float32, name='keep_prob')
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)
with tf.name_scope('out'):
W_fc2 = weight_variable([512 * 2, self.classes])
b_fc2 = bias_variable([self.classes])
y_conv = tf.add(tf.matmul(h_fc1_drop, W_fc2), b_fc2, name='y_conv_restore')
with tf.name_scope('cross_entropy'):
cross_entropy = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(labels=y_,
logits=y_conv)) # takes unnormalized output
with tf.name_scope('train'):
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32),
name='acc_to_restore')
tf.summary.scalar('accuracy', accuracy)
merged = tf.summary.merge_all()
saver = tf.train.Saver() # traditional saving api
# train the model
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
for i in range(self.STEPS):
batch = self.train.next_batch(50)
if i % 100 == 0:
train_accuracy = sess.run(accuracy,
feed_dict={
x: batch[0][:, 0, :, :],
y_: batch[1],
x2: batch[0][:, 1, :, :],
y2_: batch[1],
keep_prob: 1.0
})
print('step %d, training accuracy %g' % (i, train_accuracy))
else:
_ = sess.run(train_step,
feed_dict={
x: batch[0][:, 0, :, :],
y_: batch[1],
x2: batch[0][:, 1, :, :],
y2_: batch[1],
keep_prob: 0.5
})
# test
print('test accuracy %g' % accuracy.eval(
feed_dict={
x: self.test.images[:, 0, :, :],
y_: self.test.labels,
x2: self.test.images[:, 1, :, :],
y2_: self.test.labels,
keep_prob: 1.0
}))
# save model and checkpoint
save_path = saver.save(
sess,
os.path.join(
ROOTDIR,
"spmv_models/model-itr-{}-{}-{}.ckpt".format(self.ksplit, self.platform,
self.STEPS)))
print("Model saved in file %s" % save_path)
def testing(self):
""" restore model and checkpoint
[description]
"""
print("Model is in testing mode")
assert self.test is not None, "data not loaded"
tf.reset_default_graph() # the graph is empty now, must build graph before restore value
with tf.Session() as sess:
# retore graph
saver = tf.train.import_meta_graph(
os.path.join(
ROOTDIR,
"spmv_models/model-itr-{}-{}-{}.ckpt.meta".format(self.ksplit, self.platform,
self.STEPS)))
# the current graph can be explored by
graph = tf.get_default_graph()
# restore value
saver.restore(sess, tf.train.latest_checkpoint(os.path.join(ROOTDIR, 'spmv_models')))
print("Model restored")
x = graph.get_tensor_by_name("upper/input/x:0")
y = graph.get_tensor_by_name("upper/input/y:0")
x2 = graph.get_tensor_by_name("lower/input/x:0")
y2_ = graph.get_tensor_by_name("lower/input/y:0")
y_conv = graph.get_tensor_by_name('out/y_conv_restore:0')
keep_prob = graph.get_tensor_by_name("dropout/keep_prob:0")
# for tensor, use get_tensor_by_name()
# for operation, use get_operation_by_name()
# NOTE: Tensor names must be of the form "<op_name>:<output_index>"
acc = graph.get_tensor_by_name('train/acc_to_restore:0')
# test
print("-------------------------------------------------------")
print('Test accuracy %g' % sess.run(acc,
feed_dict={
x: self.test.images[:, 0, :, :],
y: self.test.labels,
x2: self.test.images[:, 1, :, :],
y2_: self.test.labels,
keep_prob: 1.0
}))
pred = sess.run(tf.argmax(y_conv, axis=1),
feed_dict={
x: self.test.images[:, 0, :, :],
x2: self.test.images[:, 1, :, :],
keep_prob: 1.0
})
return pred