以下是我的代码以及在colab上使用TPU时的错误。你觉得呢？
我正在尝试在colab上运行VAE模型。有关数据集的更多信息，请访问this页面。这是一个信用卡欺诈数据集。
请让我知道如何解决此问题。
谢谢您的帮助。
'''
#WITH TPU
latent_dim = 2 #number of latent variables to learn
hidden_size = 32
input_dim = x_train.shape[1]
latent_dim = 3 # d, dimensionality of the latent code t.
intermediate_dim = 256 # Size of the hidden layer.
def create_model():
x = Input(shape=(input_dim,))
t = BatchNormalization()(x)
t = Dense(intermediate_dim, activation='relu' , name='encoder_hidden')(t)
t = BatchNormalization()(t)
z_mean = Dense(latent_dim, name='z_mean')(t)
z_log_var = Dense(latent_dim, name='z_log_var')(t)
def sampling(args):
z_mean, z_log_var = args
epsilon = tf.random.normal(shape=tf.shape(z_mean), mean=0., stddev=1.,name="epsilon")
return z_mean + tf.exp(z_log_var / 2) * epsilon
z = Lambda(sampling, name='z_sampled')([z_mean, z_log_var])
t = Dense(intermediate_dim, activation='relu', name='decoder_hidden')(z)
decoded_mean = Dense(input_dim, activation=None, name='decoded_mean')(t)
def kl_loss(y_true, y_pred):
kl_loss_ = - 0.5 * tf.reduce_sum(1 + z_log_var - tf.square(z_mean) - tf.exp(z_log_var), axis=-1)
return kl_loss_
def rec_loss(y_true, y_pred):
rec_loss_ = tf.reduce_sum(tf.square(y_true - y_pred), axis=-1)
return rec_loss_
def vae_loss(x, decoded_mean):
rec_loss_ = tf.reduce_sum(tf.square(x - decoded_mean), axis=-1)
kl_loss_ = - 0.5 * tf.reduce_sum(1 + z_log_var - tf.square(z_mean) - tf.exp(z_log_var), axis=-1)
vae_loss_ = K.mean((rec_loss + kl_loss) / 2)
return vae_loss_
vae = Model(x, decoded_mean)
return vae
with strategy.scope():
vae = create_model()
vae.compile(optimizer=Adam(lr=1e-2), loss=vae_loss, metrics=[rec_loss, kl_loss])
#vae.compile(optimizer=tf.keras.optimizers.Nadam(), loss=negloglik)
vae.summary()
n_epochs = 30
batch_size = 128
early_stopping = EarlyStopping(monitor='loss', patience=10, min_delta=1e-5) #stop training if loss does not decrease with at least 0.00001
reduce_lr = ReduceLROnPlateau(monitor='loss', patience=5, min_delta=1e-5, factor=0.2) #reduce learning rate (divide it by 5 = multiply it by 0.2) if loss does not decrease with at least 0.00001
callbacks = [early_stopping, reduce_lr]
#tf.config.experimental_run_functions_eagerly(True)
tf_train = tf.data.Dataset.from_tensor_slices((x_train, x_train)).batch(batch_size).prefetch(tf.data.experimental.AUTOTUNE).shuffle(int(10e4))
tf_val = tf.data.Dataset.from_tensor_slices((x_val, x_val)).batch(batch_size).prefetch(tf.data.experimental.AUTOTUNE).shuffle(int(10e4))
hist = vae.fit(tf_train,
validation_data=tf_val,
shuffle=True,
verbose=0,
#batch_size=batch_size,
epochs=n_epochs,
callbacks=callbacks)
'''
***这是我得到的错误！ ***
'''
Model: "functional_5"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_3 (InputLayer) [(None, 30)] 0
__________________________________________________________________________________________________
batch_normalization_4 (BatchNor (None, 30) 120 input_3[0][0]
__________________________________________________________________________________________________
encoder_hidden (Dense) (None, 256) 7936 batch_normalization_4[0][0]
__________________________________________________________________________________________________
batch_normalization_5 (BatchNor (None, 256) 1024 encoder_hidden[0][0]
__________________________________________________________________________________________________
z_mean (Dense) (None, 3) 771 batch_normalization_5[0][0]
__________________________________________________________________________________________________
z_log_var (Dense) (None, 3) 771 batch_normalization_5[0][0]
__________________________________________________________________________________________________
z_sampled (Lambda) (None, 3) 0 z_mean[0][0]
z_log_var[0][0]
__________________________________________________________________________________________________
decoder_hidden (Dense) (None, 256) 1024 z_sampled[0][0]
__________________________________________________________________________________________________
decoded_mean (Dense) (None, 30) 7710 decoder_hidden[0][0]
==================================================================================================
Total params: 19,356
Trainable params: 18,784
Non-trainable params: 572
__________________________________________________________________________________________________
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
/usr/local/lib/python3.6/dist-packages/tensorflow/python/eager/execute.py in quick_execute(op_name, num_outputs, inputs, attrs, ctx, name)
59 tensors = pywrap_tfe.TFE_Py_Execute(ctx._handle, device_name, op_name,
---> 60 inputs, attrs, num_outputs)
61 except core._NotOkStatusException as e:
TypeError: An op outside of the function building code is being passed
a "Graph" tensor. It is possible to have Graph tensors
leak out of the function building context by including a
tf.init_scope in your function building code.
For example, the following function will fail:
@tf.function
def has_init_scope():
my_constant = tf.constant(1.)
with tf.init_scope():
added = my_constant * 2
The graph tensor has name: z_log_var/BiasAdd:0
During handling of the above exception, another exception occurred:
_SymbolicException Traceback (most recent call last)
9 frames
/usr/local/lib/python3.6/dist-packages/tensorflow/python/eager/execute.py in quick_execute(op_name, num_outputs, inputs, attrs, ctx, name)
72 raise core._SymbolicException(
73 "Inputs to eager execution function cannot be Keras symbolic "
---> 74 "tensors, but found {}".format(keras_symbolic_tensors))
75 raise e
76 # pylint: enable=protected-access
_SymbolicException: Inputs to eager execution function cannot be Keras symbolic tensors, but found [<tf.Tensor 'z_log_var/BiasAdd:0' shape=(None, 3) dtype=float32>, <tf.Tensor 'z_mean/BiasAdd:0' shape=(None, 3) dtype=float32>]
'''