MultiProcessMapData with strict (#414)

.github/ISSUE_TEMPLATE.md

 Bug Reports/Feature Requests/Usage Questions Only:
 
-Bug Reports: PLEASE always include
+Bug reports or other problems with code: PLEASE always include
 1. What you did. (command you run and changes you made if using examples; post or describe your code if not)
-2. What you observed, e.g. logs.
+2. What you observed, e.g. as much as logs possible.
 3. What you expected, if not obvious.
 4. Your environment (TF version, cudnn version, number & type of GPUs), if it matters.
 5. About efficiency, PLEASE first read http://tensorpack.readthedocs.io/en/latest/tutorial/performance-tuning.html
@@ -14,10 +14,4 @@ Feature Requests:
 	It may not have to be added to tensorpack unless you have a good reason.
 3. Note that we don't implement papers at others' requests.
 
-Usage Questions, e.g.:
-"How do I do [this specific thing] in tensorpack?"
-"Why certain examples need to be written in this way?"
-We don't answer general machine learning questions like:
-"I want to do [this machine learning task]. What specific things do I need to do?"
-
 You can also use gitter (https://gitter.im/tensorpack/users) for more casual discussions.

tensorpack/dataflow/prefetch.py

@@ -67,20 +67,11 @@ def _zmq_catch_error(name):
 
 
 class _MultiProcessZMQDataFlow(DataFlow):
-    def __init__(self, ds):
+    def __init__(self):
         assert os.name != 'nt', "ZMQ IPC doesn't support windows!"
         self._reset_done = False
         self._procs = []
 
-        self.ds = ds
-        try:
-            self._size = ds.size()
-        except NotImplementedError:
-            self._size = -1
-
-    def size(self):
-        return self.ds.size()
-
     def reset_state(self):
         """
         All forked dataflows are reset **once and only once** in spawned processes.
@@ -265,10 +256,17 @@ class PrefetchDataZMQ(_MultiProcessZMQDataFlow):
         if nr_proc > 1:
             logger.info("[PrefetchDataZMQ] Will fork a dataflow more than one times. "
                         "This assumes the datapoints are i.i.d.")
+        try:
+            self._size = ds.size()
+        except NotImplementedError:
+            self._size = -1
 
     def _recv(self):
         return loads(self.socket.recv(copy=False).bytes)
 
+    def size(self):
+        return self.ds.size()
+
     def get_data(self):
         with self._guard, _zmq_catch_error('PrefetchDataZMQ'):
             for k in itertools.count():
@@ -311,7 +309,59 @@ class PrefetchOnGPUs(PrefetchDataZMQ):
                     proc.start()
 
 
-class MultiThreadMapData(ProxyDataFlow):
+class _ParallelMapData(ProxyDataFlow):
+    def __init__(self, ds, buffer_size):
+        super(_ParallelMapData, self).__init__(ds)
+        assert buffer_size > 0, buffer_size
+        self._buffer_size = buffer_size
+
+    def _recv(self):
+        pass
+
+    def _send(self, dp):
+        pass
+
+    def _recv_filter_none(self):
+        ret = self._recv()
+        assert ret is not None, \
+            "[{}] Map function cannot return None when strict mode is used.".format(type(self).__name__)
+        return ret
+
+    def _fill_buffer(self):
+        try:
+            for _ in range(self._buffer_size):
+                dp = next(self._iter)
+                self._send(dp)
+        except StopIteration:
+            logger.error(
+                "[{}] buffer_size cannot be larger than the size of the DataFlow!".format(type(self).__name__))
+            raise
+
+    def get_data_non_strict(self):
+        for dp in self._iter:
+            self._send(dp)
+            yield self._recv()
+
+        self._iter = self.ds.get_data()   # refresh
+        for _ in range(self._buffer_size):
+            self._send(next(self._iter))
+            yield self._recv()
+
+    def get_data_strict(self):
+        for dp in self._iter:
+            self._send(dp)
+            yield self._recv_filter_none()
+        self._iter = self.ds.get_data()   # refresh
+
+        # first clear the buffer, then fill
+        for k in range(self._buffer_size):
+            dp = self._recv_filter_none()
+            if k == self._buffer_size - 1:
+                self._fill_buffer()
+            yield dp
+
+
+class MultiThreadMapData(_ParallelMapData):
     """
     Same as :class:`MapData`, but start threads to run the mapping function.
     This is useful when the mapping function is the bottleneck, but you don't
@@ -367,11 +417,10 @@ class MultiThreadMapData(ProxyDataFlow):
             buffer_size (int): number of datapoints in the buffer
             strict (bool): use "strict mode", see notes above.
         """
-        super(MultiThreadMapData, self).__init__(ds)
+        super(MultiThreadMapData, self).__init__(ds, buffer_size)
 
         self._strict = strict
         self.nr_thread = nr_thread
-        self.buffer_size = buffer_size
         self.map_func = map_func
         self._threads = []
         self._evt = None
@@ -398,43 +447,20 @@ class MultiThreadMapData(ProxyDataFlow):
         # only call once, to ensure inq+outq has a total of buffer_size elements
         self._fill_buffer()
 
-    def _fill_buffer(self):
-        n = self.buffer_size - self._in_queue.qsize() - self._out_queue.qsize()
-        assert n >= 0, n
-        if n == 0:
-            return
-        try:
-            for _ in range(n):
-                self._in_queue.put(next(self._iter))
-        except StopIteration:
-            logger.error("[MultiThreadMapData] buffer_size cannot be larger than the size of the DataFlow!")
-            raise
-
     def _recv(self):
-        ret = self._out_queue.get()
-        if ret is None:
-            assert not self._strict, \
-                "[MultiThreadMapData] Map function cannot return None when strict mode is used."
-        return ret
+        return self._out_queue.get()
 
-    def get_data(self):
-        with self._guard:
-            for dp in self._iter:
+    def _send(self, dp):
         self._in_queue.put(dp)
-                yield self._recv()
 
-            self._iter = self.ds.get_data()
+    def get_data(self):
+        with self._guard:
             if self._strict:
-                # first call get() to clear the queues, then fill
-                for k in range(self.buffer_size):
-                    dp = self._recv()
-                    if k == self.buffer_size - 1:
-                        self._fill_buffer()
+                for dp in self.get_data_strict():
                     yield dp
             else:
-                for _ in range(self.buffer_size):
-                    self._in_queue.put(next(self._iter))
-                    yield self._recv()
+                for dp in self.get_data_non_strict():
+                    yield dp
 
     def __del__(self):
         if self._evt is not None:
@@ -447,10 +473,20 @@ class MultiThreadMapData(ProxyDataFlow):
 ThreadedMapData = MultiThreadMapData
 
 
-class MultiProcessMapDataZMQ(_MultiProcessZMQDataFlow):
+class MultiProcessMapDataZMQ(_ParallelMapData, _MultiProcessZMQDataFlow):
     """
     Same as :class:`MapData`, but start processes to run the mapping function,
     and communicate with ZeroMQ pipe.
+
+    Note:
+        1. Processes run in parallel and can take different time to run the
+           mapping function. Therefore the order of datapoints won't be
+           preserved, and datapoints from one pass of `df.get_data()` might get
+           mixed with datapoints from the next pass.
+
+           You can use **strict mode**, where `MultiProcessMapData.get_data()`
+           is guranteed to produce the exact set which `df.get_data()`
+           produces. Although the order of data still isn't preserved.
     """
     class _Worker(mp.Process):
         def __init__(self, identity, map_func, pipename, hwm):
@@ -472,30 +508,32 @@ class MultiProcessMapDataZMQ(_MultiProcessZMQDataFlow):
                 dp = self.map_func(dp)
                 socket.send(dumps(dp), copy=False)
 
-    def __init__(self, ds, nr_proc, map_func, buffer_size=200):
+    def __init__(self, ds, nr_proc, map_func, buffer_size=200, strict=False):
         """
         Args:
             ds (DataFlow): the dataflow to map
             nr_proc(int): number of threads to use
             map_func (callable): datapoint -> datapoint | None
             buffer_size (int): number of datapoints in the buffer
+            strict (bool): use "strict mode", see notes above.
         """
-        super(MultiProcessMapDataZMQ, self).__init__(ds)
+        _ParallelMapData.__init__(self, ds, buffer_size)
+        _MultiProcessZMQDataFlow.__init__(self)
         self.nr_proc = nr_proc
         self.map_func = map_func
-        self.buffer_size = buffer_size
+        self._strict = strict
         self._procs = []
         self._guard = DataFlowReentrantGuard()
 
     def _reset_once(self):
         self.context = zmq.Context()
         self.socket = self.context.socket(zmq.ROUTER)
-        self.socket.set_hwm(self.buffer_size * 2)
+        self.socket.set_hwm(self._buffer_size * 2)
         pipename = _get_pipe_name('dataflow-map')
         _bind_guard(self.socket, pipename)
 
         self._proc_ids = [u'{}'.format(k).encode('utf-8') for k in range(self.nr_proc)]
-        worker_hwm = int(self.buffer_size * 2 // self.nr_proc)
+        worker_hwm = int(self._buffer_size * 2 // self.nr_proc)
         self._procs = [MultiProcessMapDataZMQ._Worker(
             self._proc_ids[k], self.map_func, pipename, worker_hwm)
             for k in range(self.nr_proc)]
@@ -507,14 +545,8 @@ class MultiProcessMapDataZMQ(_MultiProcessZMQDataFlow):
         self._start_processes()
         self._fill_buffer()
 
-    def _fill_buffer(self):
-        # Filling the buffer.
-        try:
-            for _ in range(self.buffer_size):
-                self._send(next(self._iter))
-        except StopIteration:
-            logger.error("[MultiProcessMapData] buffer_size cannot be larger than the size of the DataFlow!")
-            raise
+    def reset_state(self):
+        _MultiProcessZMQDataFlow.reset_state(self)
 
     def _send(self, dp):
         # round-robin assignment
@@ -529,14 +561,12 @@ class MultiProcessMapDataZMQ(_MultiProcessZMQDataFlow):
 
     def get_data(self):
         with self._guard, _zmq_catch_error('MultiProcessMapData'):
-            for dp in self._iter:
-                self._send(dp)
-                yield self._recv()
-
-            self._iter = self.ds.get_data()   # refresh
-            for _ in range(self.buffer_size):
-                self._send(next(self._iter))
-                yield self._recv()
+            if self._strict:
+                for dp in self.get_data_strict():
+                    yield dp
+            else:
+                for dp in self.get_data_non_strict():
+                    yield dp
 
 
 MultiProcessMapData = MultiProcessMapDataZMQ  # alias
@@ -549,13 +579,13 @@ if __name__ == '__main__':
 
         def get_data(self):
             for k in range(self._size):
-                yield [0]
+                yield [k]
 
         def size(self):
             return self._size
 
     ds = Zero(300)
-    ds = MultiProcessMapData(ds, 3, lambda x: [x[0] + 1])
+    ds = MultiProcessMapData(ds, 3, lambda x: [x[0] + 1], strict=True)
     ds.reset_state()
     for k in ds.get_data():
         print("Bang!", k)