Merge pull request #36 from antinucleon/op

Add BatchNorm, change interface

Makefile

@@ -63,14 +63,14 @@ endif
 
 BIN = tests/test_simple_engine
 OBJ = narray_function_cpu.o
-OBJCXX11 = reshape_cpu.o dag_engine.o simple_engine.o narray.o c_api.o operator.o symbol.o storage.o fully_connected_cpu.o static_graph.o activation_cpu.o graph_executor.o softmax_cpu.o elementwise_sum_cpu.o pooling_cpu.o convolution_cpu.o io.o iter_mnist.o
+OBJCXX11 = batch_norm_cpu.o reshape_cpu.o dag_engine.o simple_engine.o narray.o c_api.o operator.o symbol.o storage.o fully_connected_cpu.o static_graph.o activation_cpu.o graph_executor.o softmax_cpu.o elementwise_sum_cpu.o pooling_cpu.o convolution_cpu.o io.o iter_mnist.o
 CUOBJ =
 SLIB = lib/libmxnet.so
 ALIB = lib/libmxnet.a
 LIB_DEP = $(DMLC_CORE)/libdmlc.a
 
 ifeq ($(USE_CUDA), 1)
-	CUOBJ += reshape_gpu.o narray_function_gpu.o fully_connected_gpu.o activation_gpu.o elementwise_sum_gpu.o pooling_gpu.o softmax_gpu.o convolution_gpu.o
+	CUOBJ += batch_norm_gpu.o reshape_gpu.o narray_function_gpu.o fully_connected_gpu.o activation_gpu.o elementwise_sum_gpu.o pooling_gpu.o softmax_gpu.o convolution_gpu.o
 endif
 
 .PHONY: clean all test lint doc
@@ -105,6 +105,8 @@ convolution_cpu.o: src/operator/convolution.cc
 convolution_gpu.o: src/operator/convolution.cu
 reshape_cpu.o: src/operator/reshape.cc
 reshape_gpu.o: src/operator/reshape.cu
+batch_norm_cpu.o: src/operator/batch_norm.cc
+batch_norm_gpu.o: src/operator/batch_norm.cu
 io.o: src/io/io.cc
 iter_mnist.o: src/io/iter_mnist.cc
 

include/mxnet/c_api.h

@@ -358,6 +358,16 @@ MXNET_DLL int MXSymbolListArguments(SymbolHandle symbol,
 MXNET_DLL int MXSymbolListReturns(SymbolHandle symbol,
                                   mx_uint *out_size,
                                   const char ***out_str_array);
+/*!
+ * \brief List auxiliary states in the symbol.
+ * \param symbol the symbol
+ * \param out_size output size
+ * \param out_str_array pointer to hold the output string array
+ * \return 0 when success, -1 when failure happens
+ */
+MXNET_DLL int MXSymbolListAuxiliaryStates(SymbolHandle symbol,
+                                          mx_uint *out_size,
+                                          const char ***out_str_array);
 /*!
  * \brief Compose the symbol on other symbols.
  *
@@ -406,6 +416,9 @@ MXNET_DLL int MXSymbolGrad(SymbolHandle sym,
  * \param out_shape_size sizeof the returning array of out_shapes
  * \param out_shape_ndim returning array of shape dimensions of eachs input shape.
  * \param out_shape_data returning array of pointers to head of the input shape.
+ * \param aux_shape_size sizeof the returning array of aux_shapes
+ * \param aux_shape_ndim returning array of shape dimensions of eachs auxiliary shape.
+ * \param aux_shape_data returning array of pointers to head of the auxiliary shape.
  * \param complete whether infer shape completes or more information is needed.
  * \return 0 when success, -1 when failure happens
  */
@@ -420,6 +433,9 @@ MXNET_DLL int MXSymbolInferShape(SymbolHandle sym,
                                  mx_uint *out_shape_size,
                                  const mx_uint **out_shape_ndim,
                                  const mx_uint ***out_shape_data,
+                                 mx_uint *aux_shape_size,
+                                 const mx_uint **aux_shape_ndim,
+                                 const mx_uint ***aux_shape_data,
                                  int *complete);
 //--------------------------------------------
 // Part 4: Executor interface
@@ -428,9 +444,10 @@ MXNET_DLL int MXSymbolInferShape(SymbolHandle sym,
  * \brief Executor forward method
  *
  * \param handle executor handle
+ * \param is_train bool value to indicate whether the forward pass is for evaluation
  * \return 0 when success, -1 when failure happens
  */
-MXNET_DLL int MXExecutorForward(ExecutorHandle handle);
+MXNET_DLL int MXExecutorForward(ExecutorHandle handle, bool is_train);
 /*!
  * \brief Excecutor run backward
  *
@@ -466,6 +483,8 @@ MXNET_DLL int MXExecutorHeads(ExecutorHandle handle,
  * \param in_args in args array
  * \param arg_grad_store arg grads handle array
  * \param grad_req_type grad req array
+ * \param aux_states_len length of auxiliary states
+ * \param aux_states auxiliary states array
  * \param out output executor handle
  * \return 0 when success, -1 when failure happens
  */
@@ -476,6 +495,8 @@ MXNET_DLL int MXExecutorBind(SymbolHandle symbol_handle,
                              NArrayHandle *in_args,
                              NArrayHandle *arg_grad_store,
                              mx_uint *grad_req_type,
+                             mx_uint aux_states_len,
+                             NArrayHandle *aux_states,
                              ExecutorHandle *out);
 
 //--------------------------------------------

include/mxnet/narray.h

@@ -414,6 +414,7 @@ struct NArrayFunctionReg
 }  // namespace mxnet
 
 namespace dmlc {
+/*!\brief traits */
 DMLC_DECLARE_TRAITS(has_saveload, mxnet::NArray, true);
 }  // namespace dmlc
 #endif  // MXNET_NARRAY_H_

include/mxnet/operator.h

@@ -80,12 +80,15 @@ class Operator {
    * \param req the request types of saving operation, can only be kWriteTo or kWriteInplace.
    * \param out_data array of output data, pointer is used to indicate that this is holder
    *        the space of TBlob in out_data must be pre-allocated with InferShape
+   * \param aux_states Auxiliary states of operator. Normally operator doesn't 
+   *        need, epecial case like Batch Norm requires.
    * \sa OpReqType, OpContext
    */
   virtual void Forward(const OpContext &ctx,
                        const std::vector<TBlob> &in_data,
                        const std::vector<OpReqType> &req,
-                       const std::vector<TBlob> &out_data) = 0;
+                       const std::vector<TBlob> &out_data,
+                       const std::vector<TBlob> &aux_states) = 0;
   /*!
    * \brief Perform a Backward Operation, write gradient to the in_grad.
    *
@@ -111,14 +114,16 @@ class Operator {
    * \param out_data the array of output data.
    * \param req request types of the saving operation, can be all types.
    * \param in_grad the array of gradient we need to write to.
+   * \param aux_states Auxiliary states of operator. Normally operator doesn't need
    * \sa OperatorProperty, OpReqType, OpContext
    */
   virtual void Backward(const OpContext &ctx,
                         const std::vector<TBlob> &out_grad,
                         const std::vector<TBlob> &in_data,
                         const std::vector<TBlob> &out_data,
                         const std::vector<OpReqType> &req,
-                        const std::vector<TBlob> &in_grad) {
+                        const std::vector<TBlob> &in_grad,
+                        const std::vector<TBlob> &aux_states) {
     LOG(FATAL) << "Backward is not implemented";
   }
 };
@@ -158,6 +163,13 @@ class OperatorProperty {
   virtual std::vector<std::string> ListReturns() const {
     return {"output"};
   }
+  /*!
+   * \brief Get name of auxilary states of Operator
+   * \return name of return values.
+   */
+  virtual std::vector<std::string> ListAuxiliaryStates() const {
+    return {};
+  }
   /*! \return number of real return values of the Operator */
   virtual int NumReturns() const {
     return 1;
@@ -189,11 +201,14 @@ class OperatorProperty {
    *
    * \param out_shape the shape of outputs of the operator
    *     InferShape will modify the vector to fill output TShape
+   * \param aux_shape the shape of auxiliary states of the operator
+   *     InferShape will modify the vector to fill output TShape
    * \return true if the shape inference is successful, false if there is not enough information.
    * \throws dmlc::Error if the known arg_shapes are inconsistent.
    */
   virtual bool InferShape(std::vector<TShape> *in_shape,
-                          std::vector<TShape> *out_shape) const = 0;
+                          std::vector<TShape> *out_shape,
+                          std::vector<TShape> *aux_shape) const = 0;
   /*!
    * \brief Copy this OperatorProperty.
    * \return a pointer to the copied OperatorProperty

include/mxnet/symbolic.h

@@ -128,10 +128,12 @@ class StaticGraph {
    *
    * \param topo_order The topological order of node index, as created by TopoSort.
    * \param node_out_shapes The shapes of the each outputs of nodes in the graph.
+   * \param node_aux_shapes The shapes of the each auxiliary states of nodes in the graph.
    * \return if the shape inference is successful, return true, else return false.
    */
   bool InferNodeShapes(const std::vector<uint32_t> &topo_order,
-                       std::vector<std::vector<TShape> > *node_out_shapes) const;
+                       std::vector<std::vector<TShape> > *node_out_shapes,
+                       std::vector<std::vector<TShape> > *node_aux_shapes) const;
   /*!
    * \brief infer the shapes of outputs and unknown input arguments
    * \param in_shape the shape of input arguments of the operator
@@ -144,10 +146,13 @@ class StaticGraph {
    *
    * \param out_shape the shape of outputs of the operator
    *     InferShape will modify the vector to fill output TShape
+   * \param aux_shape the shape of auxiliary states of the operator
+   *     InferShape will modify the vector to fill output TShape
    * \return if the shape inference is successful, return true, else return false.
    */
   bool InferShape(std::vector<TShape>* in_shape,
-                  std::vector<TShape>* out_shape) const;
+                  std::vector<TShape>* out_shape,
+                  std::vector<TShape>* aux_shape) const;
   /*!
    * \brief Add a full backward pass in the static graph.
    *  This function will add gradient nodes for each heads,
@@ -204,6 +209,8 @@ class Symbol {
   std::vector<std::string> ListArguments() const;
   /*! \return get the descriptions of outputs for this symbol */
   std::vector<std::string> ListReturns() const;
+  /*! \return get the descriptions of auxiliary data for this symbol */
+  std::vector<std::string> ListAuxiliaryStates() const;
   /*!
    * \brief get the index th element from the returned tuple.
    * \param index index of multi output
@@ -272,22 +279,26 @@ class Symbol {
    *     common practice: set the shape of data input, and usually weight's shape can be infered
    *
    * \param out_shapes Use to store the infered shapes of outputs.
+   * \param aux_shapes Use to store the infered shapes of auxiliary states
    * \return true if the shape inference is successful, false if there is not enough information.
    * \throws dmlc::Error if the known arg_shapes are inconsistent.
    */
   bool InferShape(std::vector<TShape> *arg_shapes,
-                  std::vector<TShape> *out_shapes) const;
+                  std::vector<TShape> *out_shapes,
+                  std::vector<TShape> *aux_shapes) const;
   /*!
    * \brief infer the shapes by providing shapes of known arguments.
    * \param known_arg_shapes map of argument name to shape of arguments with known shapes.
    * \param arg_shapes used to store infered shapes of arguments.
    * \param out_shapes used to store infered shapes of outputs.
+   * \param aux_shapes Use to store the infered shapes of auxiliary states
    * \return true if the shape inference is successful, false if there is not enough information.
    * \throws dmlc::Error if the known arg_shapes are inconsistent.
    */
   bool InferShape(const std::unordered_map<std::string, TShape> &known_arg_shapes,
                   std::vector<TShape> *arg_shapes,
-                  std::vector<TShape> *out_shapes) const;
+                  std::vector<TShape> *out_shapes,
+                  std::vector<TShape> *aux_shapes) const;
   /*!
    * \brief get number of outputs of this symbol
    * \return number of outputs
@@ -378,7 +389,7 @@ class Executor {
    * \brief Perform a Forward operation of Operator
    *  After this operation, user can get the result by using function head.
    */
-  virtual void Forward() = 0;
+  virtual void Forward(bool is_train) = 0;
   /*!
    * \brief Perform a Backward operation of the Operator.
    *  This must be called after Forward.
@@ -400,13 +411,15 @@ class Executor {
    * \param in_args the NArray that stores the input arguments to the symbol.
    * \param arg_grad_store NArray that is used to store the gradient output of the input arguments.
    * \param grad_req_type requirment type of gradient saving. Can only be in {kNullOp, kAddTo, kWriteTo}.
+   * \param aux_states NArray that is used as internal state in op
    * \return a new executor.
    */
   static Executor *Bind(Symbol symbol,
                         Context ctx,
                         const std::vector<NArray> &in_args,
                         const std::vector<NArray> &arg_grad_store,
-                        const std::vector<OpReqType> &grad_req_type);
+                        const std::vector<OpReqType> &grad_req_type,
+                        const std::vector<NArray> &aux_states);
 };  // class operator
 }  // namespace mxnet
 #endif  // MXNET_SYMBOLIC_H_

python/mxnet/executor.py

@@ -23,9 +23,16 @@ def __init__(self, handle):
             raise TypeError("Handle type error")
         self.handle = handle
 
-    def forward(self):
-        """Do forward."""
-        check_call(_LIB.MXExecutorForward(self.handle))
+    def forward(self, is_train=True):
+        """Do forward.
+
+        Parameters
+        ----------
+        is_train: bool
+            whether this forward is for evaluation purpose
+            Note: for test only network, please indicate in Bind (TODO)
+        """
+        check_call(_LIB.MXExecutorForward(self.handle, is_train))
 
     def backward(self, grads):
         """Do backward on heads' gradient.

python/mxnet/symbol.py

@@ -1,5 +1,5 @@
 # coding: utf-8
-# pylint: disable=invalid-name, protected-access, fixme
+# pylint: disable=invalid-name, protected-access, fixme, too-many-arguments
 """Symbol support of mxnet"""
 from __future__ import absolute_import
 
@@ -123,6 +123,20 @@ def list_returns(self):
             self.handle, ctypes.byref(size), ctypes.byref(sarr)))
         return [py_str(sarr[i]) for i in range(size.value)]
 
+    def list_auxiliary_states(self):
+        """List all auxiliary states in the symbool.
+
+        Returns
+        -------
+        args: list of string
+            List of all the auxiliary
+        """
+        size = ctypes.c_uint()
+        sarr = ctypes.POINTER(ctypes.c_char_p)()
+        check_call(_LIB.MXSymbolListAuxiliaryStates(
+            self.handle, ctypes.byref(size), ctypes.byref(sarr)))
+        return [py_str(sarr[i]) for i in range(size.value)]
+
     def infer_shape(self, *args, **kwargs):
         """Infer the shape of outputs and arguments of given known shapes of arguments.
 
@@ -147,6 +161,9 @@ def infer_shape(self, *args, **kwargs):
         out_shapes : list of tuple or None
             List of shapes of outputs.
             The order is in the same order as list_returns()
+        aux_shapes : list of tuple or None
+            List of shapes of outputs.
+            The order is in the same order as list_auxiliary()
         """
         # pylint: disable=too-many-locals
         if len(args) != 0 and len(kwargs) != 0:
@@ -176,6 +193,9 @@ def infer_shape(self, *args, **kwargs):
         out_shape_size = mx_uint()
         out_shape_ndim = ctypes.POINTER(mx_uint)()
         out_shape_data = ctypes.POINTER(ctypes.POINTER(mx_uint))()
+        aux_shape_size = mx_uint()
+        aux_shape_ndim = ctypes.POINTER(mx_uint)()
+        aux_shape_data = ctypes.POINTER(ctypes.POINTER(mx_uint))()
         complete = ctypes.c_int()
         check_call(_LIB.MXSymbolInferShape(
             self.handle, len(indptr) - 1,
@@ -188,13 +208,18 @@ def infer_shape(self, *args, **kwargs):
             ctypes.byref(out_shape_size),
             ctypes.byref(out_shape_ndim),
             ctypes.byref(out_shape_data),
+            ctypes.byref(aux_shape_size),
+            ctypes.byref(aux_shape_ndim),
+            ctypes.byref(aux_shape_data),
             ctypes.byref(complete)))
         if complete.value != 0:
             arg_shapes = [
                 tuple(arg_shape_data[i][:arg_shape_ndim[i]]) for i in range(arg_shape_size.value)]
             out_shapes = [
                 tuple(out_shape_data[i][:out_shape_ndim[i]]) for i in range(out_shape_size.value)]
-            return (arg_shapes, out_shapes)
+            aux_shapes = [
+                tuple(aux_shape_data[i][:aux_shape_ndim[i]]) for i in range(aux_shape_size.value)]
+            return (arg_shapes, out_shapes, aux_shapes)
         else:
             return (None, None)
         # pylint: enable=too-many-locals
@@ -212,7 +237,7 @@ def debug_str(self):
             self.handle, ctypes.byref(debug_str)))
         return py_str(debug_str.value)
 
-    def bind(self, ctx, args, args_grad, reqs):
+    def bind(self, ctx, args, args_grad, reqs, aux_states=None):
         """bind current symbol to get an executor.
 
         Parameters
@@ -225,15 +250,20 @@ def bind(self, ctx, args, args_grad, reqs):
             input args' gradient
         reqs: Array of enum
             graident requirements
+        aux_states: Array of NArray
+            input auxiliary states to the symbol
         """
         # TODO(bing): consider a more friendly interface
         # For example, pass in args_grad by dict
         enum = {"null" : 0, "write_to" : 1, "in_place":2, "add_to" : 3}
         if not isinstance(ctx, Context):
             raise TypeError("Context type error")
+        if aux_states == None:
+            aux_states = []
         args_handle = c_array(NArrayHandle, [item.handle for item in args])
         args_grad_handle = c_array(NArrayHandle, [item.handle for item in args_grad])
         reqs_array = c_array(mx_uint, [mx_uint(enum[item]) for item in reqs])
+        aux_args_handle = c_array(NArrayHandle, [item.handle for item in aux_states])
         handle = ExecutorHandle()
         check_call(_LIB.MXExecutorBind(self.handle,
                                        mx_uint(ctx.device_mask),
@@ -242,6 +272,8 @@ def bind(self, ctx, args, args_grad, reqs):
                                        args_handle,
                                        args_grad_handle,
                                        reqs_array,
+                                       len(aux_states),
+                                       aux_args_handle,
                                        ctypes.byref(handle)))
         return Executor(handle)