从kafka读数据训练怎么指定topic

关于从kafka读数据进行训练，文档上是这么写的：
结合data_reader.py代码，xxx部分应该是namenode，也就是broker list，那么topic在哪里指定呢。
看data_io.cc代码，应该是通过AddPath接口指定的，但是data_reader.py的_decode_path，并没有解析出fpath，因此AddPath的传递的也是个空，而DataSharding的代码看着也不支持对kafka数据分片。很疑惑，这块是不是少写了代码。请指教。

    def _decode_path(self, path):

        '''

        hdfs://namenode/path

        kafka://namenode

        '''

        namenode = ""

        fs_type = None

        fpath = None

        if path.startswith('hdfs://'):

            fs_type = pybind.fs.hdfs

            arr = path.split('/', 3)

            assert len(arr) == 4

            namenode = arr[2]

            fpath = '/'+arr[3]

        elif path.startswith('kafka://'):

            fs_type = pybind.fs.kafka

            arr = path.split('/', 2)

            assert len(arr) == 3

            namenode = arr[2]

        else:

            assert '://' not in path, "Unsupported path: %s" % path

            fpath = path


        return fs_type, namenode, fpath

Start with data_reader.py, especially _decode_path, then trace the AddPath call in data_io.cc and the DataSharding handling for Kafka paths. Verify how the broker list and topic are expected to be supplied; the issue is resolved when topic selection is supported or the documented configuration clearly matches the implementation.

Written by the indexing model from the issue text.