第八课 tensorflow numpy_input_fn 以及队列性质

numpy_input_fn 以及队列性质

该函数的作用是从numpy的输入数据中，产生读取的featrues和labels数据。这样当我们在使用numpy的数据作为输入的时候就很方便。对于所有的input来说，都是要建立队列来进行读入，所以对于队列的处理就会比较麻烦，而numpy_input的数据将这些对队列的输入封装在一起方便了我们使用.

import tensorflow as tfimport numpy as npfrom tensorflow.python.estimator.inputs import numpy_iofrom tensorflow.python.training import coordinator

# 构造numpy数据进行测试age = np.arange(4) * 1.0height = np.arange(32, 36)x = {'age': age, 'height': height}y = np.arange(-32, -28)print('age shape:', age.shape, 'height shape:', height.shape, 'y shape:', y.shape)

('age shape:', (4,), 'height shape:', (4,), 'y shape:', (4,))

batch_size 与 num_epochs解析

with tf.Session() as session:    # 定义numpy input fn    input_fn = numpy_io.numpy_input_fn(x, y, batch_size=2, shuffle=False, num_epochs=1)    # 运行input_fn, 产生featrue和targets    featrues, targets = input_fn()    coord = coordinator.Coordinator()    threads = tf.train.start_queue_runners(session, coord=coord)    ret_features, ret_targets = session.run([featrues, targets])    print('featrues:', ret_features)    print('targets:', ret_targets)    coord.request_stop()    coord.join(threads)

('featrues:', {'age': array([ 0.,  1.]), 'height': array([32, 33])})('targets:', array([-32, -31]))

上面结果看到会输出2个值。对应到 input_fn = numpy_io.numpy_input_fn(x, y, batch_size=2, shuffle=False, num_epochs=1) 就是batch_size=2. shuffle是表示是否随机的意思. num_epochs=1表示对于整个输入数据来说一共灌入到队列中几份。当前的数据大小是4，那么 num_epochs=1 就是说队列中有 4*1=4分数据。总共是4分数据，batch_size=2，那么就是一共能够获取2=4/2份数据来进行训练，如果超过2，那么就会报队列溢出的异常。

with tf.Session() as session:    # 定义numpy input fn    input_fn = numpy_io.numpy_input_fn(x, y, batch_size=2, shuffle=False, num_epochs=1)    # 运行input_fn, 产生featrue和targets    featrues, targets = input_fn()    coord = coordinator.Coordinator()    threads = tf.train.start_queue_runners(session, coord=coord)    num_step = 3    for step in range(num_step):        ret_features, ret_targets = session.run([featrues, targets])        print('featrues:', ret_features)        print('targets:', ret_targets)    coord.request_stop()    coord.join(threads)

('featrues:', {'age': array([ 0.,  1.]), 'height': array([32, 33])})('targets:', array([-32, -31]))('featrues:', {'age': array([ 2.,  3.]), 'height': array([34, 35])})('targets:', array([-30, -29]))---------------------------------------------------------------------------OutOfRangeError                           Traceback (most recent call last)<ipython-input-5-b1111e315ccf> in <module>()     12     num_step = 3     13     for step in range(num_step):---> 14         ret_features, ret_targets = session.run([featrues, targets])     15      16         print('featrues:', ret_features)/Library/Python/2.7/site-packages/tensorflow/python/client/session.pyc in run(self, fetches, feed_dict, options, run_metadata)    893     try:    894       result = self._run(None, fetches, feed_dict, options_ptr,--> 895                          run_metadata_ptr)    896       if run_metadata:    897         proto_data = tf_session.TF_GetBuffer(run_metadata_ptr)/Library/Python/2.7/site-packages/tensorflow/python/client/session.pyc in _run(self, handle, fetches, feed_dict, options, run_metadata)   1122     if final_fetches or final_targets or (handle and feed_dict_tensor):   1123       results = self._do_run(handle, final_targets, final_fetches,-> 1124                              feed_dict_tensor, options, run_metadata)   1125     else:   1126       results = []/Library/Python/2.7/site-packages/tensorflow/python/client/session.pyc in _do_run(self, handle, target_list, fetch_list, feed_dict, options, run_metadata)   1319     if handle is None:   1320       return self._do_call(_run_fn, self._session, feeds, fetches, targets,-> 1321                            options, run_metadata)   1322     else:   1323       return self._do_call(_prun_fn, self._session, handle, feeds, fetches)/Library/Python/2.7/site-packages/tensorflow/python/client/session.pyc in _do_call(self, fn, *args)   1338         except KeyError:   1339           pass-> 1340       raise type(e)(node_def, op, message)   1341    1342   def _extend_graph(self):OutOfRangeError: FIFOQueue '_2_enqueue_input_1/fifo_queue' is closed and has insufficient elements (requested 2, current size 0)     [[Node: fifo_queue_DequeueUpTo_1 = QueueDequeueUpToV2[component_types=[DT_INT64, DT_DOUBLE, DT_INT64, DT_INT64], timeout_ms=-1, _device="/job:localhost/replica:0/task:0/cpu:0"](enqueue_input_1/fifo_queue, fifo_queue_DequeueUpTo_1/n)]]Caused by op u'fifo_queue_DequeueUpTo_1', defined at:  File "/System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/runpy.py", line 162, in _run_module_as_main    "__main__", fname, loader, pkg_name)  File "/System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/runpy.py", line 72, in _run_code    exec code in run_globals  File "/Library/Python/2.7/site-packages/ipykernel/__main__.py", line 3, in <module>    app.launch_new_instance()  File "/Library/Python/2.7/site-packages/traitlets-4.1.0-py2.7.egg/traitlets/config/application.py", line 589, in launch_instance    app.start()  File "/Library/Python/2.7/site-packages/ipykernel/kernelapp.py", line 474, in start    ioloop.IOLoop.instance().start()  File "/Library/Python/2.7/site-packages/zmq/eventloop/ioloop.py", line 162, in start    super(ZMQIOLoop, self).start()  File "/Library/Python/2.7/site-packages/tornado/ioloop.py", line 887, in start    handler_func(fd_obj, events)  File "/Library/Python/2.7/site-packages/tornado/stack_context.py", line 275, in null_wrapper    return fn(*args, **kwargs)  File "/Library/Python/2.7/site-packages/zmq/eventloop/zmqstream.py", line 440, in _handle_events    self._handle_recv()  File "/Library/Python/2.7/site-packages/zmq/eventloop/zmqstream.py", line 472, in _handle_recv    self._run_callback(callback, msg)  File "/Library/Python/2.7/site-packages/zmq/eventloop/zmqstream.py", line 414, in _run_callback    callback(*args, **kwargs)  File "/Library/Python/2.7/site-packages/tornado/stack_context.py", line 275, in null_wrapper    return fn(*args, **kwargs)  File "/Library/Python/2.7/site-packages/ipykernel/kernelbase.py", line 276, in dispatcher    return self.dispatch_shell(stream, msg)  File "/Library/Python/2.7/site-packages/ipykernel/kernelbase.py", line 228, in dispatch_shell    handler(stream, idents, msg)  File "/Library/Python/2.7/site-packages/ipykernel/kernelbase.py", line 390, in execute_request    user_expressions, allow_stdin)  File "/Library/Python/2.7/site-packages/ipykernel/ipkernel.py", line 196, in do_execute    res = shell.run_cell(code, store_history=store_history, silent=silent)  File "/Library/Python/2.7/site-packages/ipykernel/zmqshell.py", line 498, in run_cell    return super(ZMQInteractiveShell, self).run_cell(*args, **kwargs)  File "/Library/Python/2.7/site-packages/ipython-4.1.2-py2.7.egg/IPython/core/interactiveshell.py", line 2723, in run_cell    interactivity=interactivity, compiler=compiler, result=result)  File "/Library/Python/2.7/site-packages/ipython-4.1.2-py2.7.egg/IPython/core/interactiveshell.py", line 2825, in run_ast_nodes    if self.run_code(code, result):  File "/Library/Python/2.7/site-packages/ipython-4.1.2-py2.7.egg/IPython/core/interactiveshell.py", line 2885, in run_code    exec(code_obj, self.user_global_ns, self.user_ns)  File "<ipython-input-5-b1111e315ccf>", line 7, in <module>    featrues, targets = input_fn()  File "/Library/Python/2.7/site-packages/tensorflow/python/estimator/inputs/numpy_io.py", line 127, in input_fn    else queue.dequeue_up_to(batch_size))  File "/Library/Python/2.7/site-packages/tensorflow/python/ops/data_flow_ops.py", line 498, in dequeue_up_to    self._queue_ref, n=n, component_types=self._dtypes, name=name)  File "/Library/Python/2.7/site-packages/tensorflow/python/ops/gen_data_flow_ops.py", line 1430, in _queue_dequeue_up_to_v2    timeout_ms=timeout_ms, name=name)  File "/Library/Python/2.7/site-packages/tensorflow/python/framework/op_def_library.py", line 767, in apply_op    op_def=op_def)  File "/Library/Python/2.7/site-packages/tensorflow/python/framework/ops.py", line 2630, in create_op    original_op=self._default_original_op, op_def=op_def)  File "/Library/Python/2.7/site-packages/tensorflow/python/framework/ops.py", line 1204, in __init__    self._traceback = self._graph._extract_stack()  # pylint: disable=protected-accessOutOfRangeError (see above for traceback): FIFOQueue '_2_enqueue_input_1/fifo_queue' is closed and has insufficient elements (requested 2, current size 0)     [[Node: fifo_queue_DequeueUpTo_1 = QueueDequeueUpToV2[component_types=[DT_INT64, DT_DOUBLE, DT_INT64, DT_INT64], timeout_ms=-1, _device="/job:localhost/replica:0/task:0/cpu:0"](enqueue_input_1/fifo_queue, fifo_queue_DequeueUpTo_1/n)]]

正如上面的分析，这里 num_step=3，每次取2个，所以总量是6个就会队列溢出. 所以如果想要精确的控制，必须知道总的数据集有多大。但是，问题是实际可能我们只是知道大概的数据集大小，因为在离线训练的时候，我们需要不断的更新数据集，那么就会导致，这个数据集的大小是不确定，那么该如何处理呢？直观的做法就是，对队列的溢出异常进行捕获，就好了。

with tf.Session() as session:    # 定义numpy input fn    input_fn = numpy_io.numpy_input_fn(x, y, batch_size=2, shuffle=False, num_epochs=1)    # 运行input_fn, 产生featrue和targets    featrues, targets = input_fn()    coord = coordinator.Coordinator()    threads = tf.train.start_queue_runners(session, coord=coord)    num_step = 3    try:        for step in range(num_step):            # 这一行表示 如果coord 已经stop，就不要再进行下去了, 对应到 coord.request_stop()            # 因为 request_stop 不会立马stop, 但是通过should_stop能获取到状态            if coord.should_stop():                 break            ret_features, ret_targets = session.run([featrues, targets])            print('featrues:', ret_features)            print('targets:', ret_targets)    except tf.errors.OutOfRangeError as error:        print('ignore: ', error)    finally:        coord.request_stop()        coord.join(threads)

('featrues:', {'age': array([ 0.,  1.]), 'height': array([32, 33])})('targets:', array([-32, -31]))('featrues:', {'age': array([ 2.,  3.]), 'height': array([34, 35])})('targets:', array([-30, -29]))('ignore: ', OutOfRangeError())

当我们加入这个异常的保护，就可以很轻松的不用管数据集的数量问题了。除了上面方法，还有一种更为优雅的方法，就是队列可以自动循环的读取就好了。这里唯一要做的就是就 将 num_epochs=1 => num_epochs=None.那么，这种情况下就不会溢出了，队列会自己控制num_epochs，会循环的读入.

with tf.Session() as session:    # 定义numpy input fn, 注意: num_epochs=None，队列会循环的读入.    input_fn = numpy_io.numpy_input_fn(x, y, batch_size=2, shuffle=False, num_epochs=None)    # 运行input_fn, 产生featrue和targets    featrues, targets = input_fn()    coord = coordinator.Coordinator()    threads = tf.train.start_queue_runners(session, coord=coord)    num_step = 3    for step in range(num_step):        ret_features, ret_targets = session.run([featrues, targets])        print('featrues:', ret_features)        print('targets:', ret_targets)    coord.request_stop()    coord.join(threads)

('featrues:', {'age': array([ 0.,  1.]), 'height': array([32, 33])})('targets:', array([-32, -31]))('featrues:', {'age': array([ 2.,  3.]), 'height': array([34, 35])})('targets:', array([-30, -29]))('featrues:', {'age': array([ 0.,  1.]), 'height': array([32, 33])})('targets:', array([-32, -31]))

看到上面的最后一行的结果，就是 [0, 1] 这显然是第一个数据.

shuffle与threads

上面介绍了batch_size与num_epochs. 还有另外两个参数shuffle与threads.

• shuffle: 随机产出。这很好，方便产生随机的数据集
• threads: 是指读入数据时候的线程数.

这两个参数，有些情况，需要进行解释:

条件 解释 num_threads > 1 and num_epochs is not None num_epochs会应用到每一个线程上，所以这就会产生超过你实际设置的epochs的数量，所以如果想要限制epochs的数量，需要使用一个线程. 也就是说如果使用超过1个线程，那么最好是num_epochs设置为None. 否则，你无法控制这个数量 shuffle and num_threads > 1 and num_epochs is not None 多个线程同时读取放入到队列中，这可能会导致不会那么随机 if not shuffle and num_threads > 1 多线程无法保证顺序性

所以通过以上总结，最佳的参数,有两个组

1. num_threads > 1, shuffle=True, num_epochs = None
2. num_threads = 1, shuffle=Ture/False, num_epochs = None