Question

我正在尝试使用keras.layer使用变分推理和贝叶斯神经网络，但是当我导入为变分推理编写的类时，会出现值错误。

我可以使用tf.layer.dense而不会出现错误，但是我想使用变分推理。这是带有tf.layer和tun的代码，没有任何错误：

SELECT * from Table
WHERE (`Year` <> "2019" OR `Quarter Name` NOT IN ("QTR3", "QTR4"))

这是我编写的类，在导入它时会导致错误：

h_1 = tf.layers.dense(inputs=inputs,units=128, activation=tf.nn.leaky_relu,kernel_regularizer=regularizer)
h_1_a=tf.layers.dense(inputs=action,units=128, activation=tf.nn.leaky_relu, kernel_regularizer=regularizer)
h_1_concat=tf.concat(axis=1,values=[h_1,h_1_a])
h_2 = tf.layers.dense(inputs=h_1_concat, units=64, activation=tf.nn.leaky_relu, kernel_regularizer=regularizer)
h_3 = tf.layers.dense(inputs=h_2, units=16, activation=tf.nn.leaky_relu, kernel_regularizer=regularizer)
out = tf.layers.dense(inputs=h_3, units=1, kernel_regularizer=regularizer)

这是我使用此类的部分：

def mixture_prior_params(sigma_1, sigma_2, pi, return_sigma=False):
params = K.variable([sigma_1, sigma_2, pi], name='mixture_prior_params')
sigma = np.sqrt(pi * sigma_1 ** 2 + (1 - pi) * sigma_2 ** 2) #VI
return params, sigma
def log_mixture_prior_prob(w):
comp_1_dist = tf.distributions.Normal(0.0, prior_params[0]) 
comp_2_dist = tf.distributions.Normal(0.0, prior_params[1]) 
comp_1_weight = prior_params[2]     
return K.log(comp_1_weight * comp_1_dist.prob(w) + (1 - comp_1_weight) * comp_2_dist.prob(w))
prior_params, prior_sigma = mixture_prior_params(sigma_1=1.5, sigma_2=0.5, pi=0.3)
class DenseVariational(Layer): 
def __init__(self, output_dim, kl_loss_weight, activation=None, **kwargs): 
    self.output_dim = output_dim
    self.kl_loss_weight = kl_loss_weight
    self.activation = activations.get(activation)
    super().__init__(**kwargs)
def build(self, input_shape):  
    self._trainable_weights.append(prior_params) 
    self.kernel_mu = self.add_weight(name='kernel_mu', 
                                     shape=(input_shape[0][1], self.output_dim),
                                   initializer=initializers.normal(stddev=prior_sigma),
                                     trainable=True)
    self.bias_mu = self.add_weight(name='bias_mu', 
                                   shape=(self.output_dim,),
                                   initializer=initializers.normal(stddev=prior_sigma),
                                   trainable=True)
    self.kernel_rho = self.add_weight(name='kernel_rho', 
                                      shape=(input_shape[0][1], self.output_dim),
                                      initializer=initializers.constant(0.0),
                                      trainable=True)
    self.bias_rho = self.add_weight(name='bias_rho', 
                                    shape=(self.output_dim,),
                                    initializer=initializers.constant(0.0),
                                    trainable=True)
super().build(input_shape)

def call(self, x):
    sys.exit()
    kernel_sigma = tf.math.softplus(self.kernel_rho)
    kernel = self.kernel_mu + kernel_sigma * tf.random.normal(self.kernel_mu.shape)

    bias_sigma = tf.math.softplus(self.bias_rho)
    bias = self.bias_mu + bias_sigma * tf.random.normal(self.bias_mu.shape)

    self.add_loss(self.kl_loss(kernel, self.kernel_mu, kernel_sigma) + 
                  self.kl_loss(bias, self.bias_mu, bias_sigma))

    return self.activation(K.dot(x, kernel) + bias)

def compute_output_shape(self, input_shape):
    #sys.exit()
    return (input_shape[0][1], self.output_dim)

def kl_loss(self, w, mu, sigma):
    variational_dist = tf.distributions.Normal(mu, sigma)
    return kl_loss_weight * K.sum(variational_dist.log_prob(w) - log_mixture_prior_prob(w)) `

我输入的形状为[7,128]，但我无法理解错误：

                _inputs=tf.concat(axis=1,values=[inputs, action])
            x_in = Input(shape=(7,))

            x = DenseVariational(128, kl_loss_weight=kl_loss_weight, activation='relu')(x_in)
            x = DenseVariational(64, kl_loss_weight=kl_loss_weight, activation='relu')(x)
            x = DenseVariational(64, kl_loss_weight=kl_loss_weight, activation='relu')(x)
            x = DenseVariational(16, kl_loss_weight=kl_loss_weight, activation='relu')(x)
            x = DenseVariational(3, kl_loss_weight=kl_loss_weight)(x)

            model = Model(x_in,x)

            out = model(x)

ValueError：尺寸必须相等，但输入形状为[3]，[7,128]的“添加”（操作：“添加”）的尺寸应为3和128

0 个答案: