data.py 文件源码

def get_input_op(self, fq, *args, **kwargs):
        """
        This is the main method that returns a tensorflow data reading operation.

        This method will get called n_threads * n_attrs times in the method init_ops (see above).
        Specifically, it is called once for each thread id and each attribute group.

        The arguments are:
             fq:  filename queue object.  When run in a tf session, this object will act
                  as a queue of filenames.  When fq.dequeue() is called in a tf.Session, it
                  will produce the next filename to begin reading from.   Note: it only makes
                  sense to dequeue from fq if the current file being read has been completed.
             *args: any position arguments to the reader.  these are specified on a
                  per-attribute-group basis (eg. across thread ids, calls for the same attribute
                  group will get the same args).
             *kwargs: any keyward arguments to the reader.  like for *args, these are specified
                  on a per-attribute-group basis.

        As an example of this method, see the TFRecordParallelByFileProvider.get_input_ops.
        """
        raise NotImplementedError()