cdd8a8d686272217f7d9a7a0de34262192696b28,batchflow/models/tf/encoder_decoder.py,VariationalAutoEncoder,embedding,#Any#Any#Any#,367

Before Change

        base_block = kwargs.get("base")

        mean = base_block(inputs, name="mean", **kwargs)
        std = base_block(inputs, name="std", **kwargs)
        eps = tf.random.normal(shape=tf.shape(mean))
        return mean + eps*std

After Change

        -------
        tf.Tensor
        
        steps = kwargs.get("num_stages", 1)
        base_block = kwargs.get("base")
        x = inputs

        with tf.variable_scope(name):
            for i in range(steps):
                args = {**kwargs, **unpack_args(kwargs, i, steps)} // enforce priority of keys
                mean = base_block(x, name="mean-"+str(i), **args)
                std = base_block(x, name="std-"+str(i), **args)
                eps = tf.random.normal(shape=tf.shape(mean), name="eps-"+str(i))
                x = mean + eps*std
        return x

Instances