各位读者大家好，
Keras 2.10.0
我正在尝试在训练完后保存在 Keras 和 Tensorflow 中构建的变分自动编码器。 我无法这样做。我该怎么做才能保存模型？
附件是我的跑步基础模型的 Colab 链接。
它创建的错误消息在这里
ValueError: Model <__main__.VAE object at 0x00000278F0A96E60> cannot be saved either because the input shape is not available or because the forward pass of the model is not defined.To define a forward pass, please override `Model.call()`. To specify an input shape, either call `build(input_shape)` directly, or call the model on actual data using `Model()`, `Model.fit()`, or `Model.predict()`. If you have a custom training step, please make sure to invoke the forward pass in train step through `Model.__call__`, i.e. `model(inputs)`, as opposed to `model.call()`.
此外，如果您不想打开 Colab，我会在此处添加主要代码行。
任何帮助将不胜感激。
import numpy as np
from tensorflow import keras
from keras.layers import Input, Dense, Lambda
import tensorflow as tf
import pandas as pd
import matplotlib.pyplot as plt
from sklearn.datasets import make_classification
from keras import backend as K
import matplotlib.pyplot as plt
import numpy as np
# Creating fake data to replicate
# Define the desired number of samples for each class
class_samples = [2000, 3000, 2500, 1500] # Adjust these numbers as needed
# Calculate weights based on the desired number of samples
class_weights = [num_samples / sum(class_samples) for num_samples in class_samples]
# Generate a synthetic dataset with different numbers of samples for each class
X, y = make_classification(
n_samples=sum(class_samples),
n_features=4,
n_informative=4,
n_redundant=0,
n_classes=4,
weights=class_weights,
random_state=42,
)
columns = ["Feature_1", "Feature_2", "Feature_3", "Feature_4"]
synthetic_df = pd.DataFrame(data=X, columns=columns)
for column in synthetic_df:
std = np.std(synthetic_df[column])
mean = np.mean(synthetic_df[column])
synthetic_df[column] = synthetic_df[column]-mean
synthetic_df[column] = synthetic_df[column]/std
synthetic_df["target"] = y
synthetic_array =synthetic_df.values
# Defining the sampling layer that is also the call
class Sampling(keras.layers.Layer):
def call(self, inputs):
z_mean, z_log_var = inputs
batch = tf.shape(z_mean)[0]
dim = tf.shape(z_mean)[1]
epsilon = tf.keras.backend.random_normal(shape=(batch, dim))
return z_mean + tf.exp(0.5 * z_log_var) * epsilon
# Encoder
latent_dim = 10
encoder_inputs = Input(shape=(5), name="input_layer")
n_x = 50
x = Dense(n_x, activation="relu", name="h1")(encoder_inputs)
# Split x into two halves
half_size = n_x // 2
x3_first_half = Lambda(lambda x: x[:, :half_size], name="select_z_mean")(x)
x3_second_half = Lambda(lambda x: x[:, half_size:], name="select_z_var")(x)
z_mean = Dense(latent_dim, name="z_mean")(x3_first_half)
z_log_var = Dense(latent_dim, name="z_log_var")(x3_second_half)
z = Sampling(name="Sampling")([z_mean, z_log_var])
encoder = keras.Model(encoder_inputs, [z_mean, z_log_var, z], name="encoder")
# Decoder
# Decoder
latent_inputs = keras.Input(shape=(latent_dim,))
n_x = 30
x = Dense(n_x, activation="relu", name="h4")(latent_inputs)
cont_decoder_outputs = Dense(4, activation="linear", name="cont_decoder_output")(x)
class_decoder_output = Dense(4, activation="softmax", name="classification_output")(x)
decoder = keras.Model(latent_inputs, [cont_decoder_outputs, class_decoder_output], name="decoder")
decoder.summary()
# Custom VAE
class VAE(keras.Model):
def __init__(self, encoder, decoder, **kwargs):
super().__init__(**kwargs)
self.encoder = encoder
self.decoder = decoder
self.total_loss_tracker = keras.metrics.Mean(name="total_loss")
@property
def metrics(self):
return [
self.total_loss_tracker,
]
def train_step(self, data):
with tf.GradientTape() as tape:
z_mean, z_log_var, z = self.encoder(data)
reconstruction_cont, reconstruction_class = self.decoder(z)
data_cont = data[
:, :4
] # Assuming the first 4 columns are for continuous variables
data_class = data[:, 4:] # Assuming the last column is for classification
# Reconstruction loss for continuous outputs
reconstruction_loss_cont = keras.losses.mean_squared_error(
data_cont, reconstruction_cont
)
# Reconstruction loss for classification output
reconstruction_loss_class = keras.losses.sparse_categorical_crossentropy(
data_class, reconstruction_class
)
kl_loss = -0.5 * (1 + z_log_var - tf.square(z_mean) - tf.exp(z_log_var))
kl_loss = tf.reduce_mean(kl_loss, axis=1)
# Combine losses
total_loss = (
0.8*reconstruction_loss_cont + 0.2*reconstruction_loss_class + kl_loss
)
grads = tape.gradient(total_loss, self.trainable_weights)
self.optimizer.apply_gradients(zip(grads, self.trainable_weights))
self.total_loss_tracker.update_state(total_loss)
return {
"loss": self.total_loss_tracker.result(),
}
def get_config(self):
config = super(VAE, self).get_config()
config.update(
{
"encoder": self.encoder,
"decoder": self.decoder,
}
)
return config
# Compile and Train
vae = VAE(encoder, decoder)
vae.compile(optimizer=keras.optimizers.Adam())
vae.fit(
synthetic_array,
epochs=1,
batch_size=16,
)
# Save in TensorFlow SavedModel format
vae.save("path_to_save_model_tf")
非常感谢！
错误消息告诉您该怎么做，尽管在这种情况下有点尴尬。
首先，您必须为模型定义一个
callVAEdef call(self, inputs):
return self.decoder(self.encoder(inputs)[2])
这使得 Keras 能够弄清楚这个模型实际上应该做什么。仅仅调用
fittrain_step因此我们仍然需要实际调用模型一次，因此它具有定义的输入形状。在定义模型之后、尝试保存模型之前，将此行撒在任意位置：
vae(synthetic_array[:1])
这只是在单个数据点上调用模型来定义输入/输出形状。之后，保存并加载保存的模型都可以正常工作。请注意，我在 Colab 上对此进行了测试，它使用更新的 2.15 版本。我认为他们同时对保存模型进行了一些更改，因此这可能无法在旧版本上正常工作。