Merge pull request #7719 from guoshengCS/enhance-lookup_table_op-padidx

Enhance lookup_table_op to support padding_idx.

paddle/framework/attribute.cc

@@ -61,6 +61,9 @@ Attribute GetAttrValue(const proto::OpDesc::Attr& attr_desc) {
       }
       return val;
     }
+    case proto::AttrType::LONG: {
+      return attr_desc.l();
+    }
     default:
       PADDLE_THROW("Unsupport attr type %d", attr_desc.type());
   }

paddle/framework/attribute.h

@@ -168,6 +168,32 @@ struct ExtractAttribute<bool> {
   const std::string& attr_name_;
 };
 
+template <>
+struct ExtractAttribute<int64_t> {
+  explicit ExtractAttribute(const std::string& attr_name)
+      : attr_name_(attr_name) {}
+
+  int64_t* operator()(Attribute& attr) const {
+    if (attr.type() == typeid(int)) {  // NOLINT
+      int val = boost::get<int>(attr);
+      attr = static_cast<int64_t>(val);
+    } else if (attr.type() == typeid(float)) {  // NOLINT
+      int val = boost::get<float>(attr);
+      attr = static_cast<int64_t>(val);
+    }
+    int64_t* attr_value = nullptr;
+    try {
+      attr_value = &boost::get<int64_t>(attr);
+    } catch (boost::bad_get& bad_get) {
+      PADDLE_THROW("Cannot get attribute %s by type int64_t, its type is %s",
+                   attr_name_, attr.type().name());
+    }
+    return attr_value;
+  }
+
+  const std::string& attr_name_;
+};
+
 // check whether a certain attribute fit its limits
 // an attribute can have more than one limits
 template <typename T>

paddle/framework/framework.proto

@@ -26,6 +26,7 @@ enum AttrType {
   BOOLEAN = 6;
   BOOLEANS = 7;
   BLOCK = 8;
+  LONG = 9;
 }
 
 // OpDesc describes an instance of a C++ framework::OperatorBase
@@ -44,6 +45,7 @@ message OpDesc {
     optional bool b = 10;
     repeated bool bools = 11;
     optional int32 block_idx = 12;
+    optional int64 l = 13;
   };
 
   message Var {

paddle/framework/op_desc.cc

@@ -283,6 +283,7 @@ struct SetAttrDescVisitor : public boost::static_visitor<void> {
     VectorToRepeated(v, attr_->mutable_bools());
   }
   void operator()(BlockDesc *desc) const { attr_->set_block_idx(desc->ID()); }
+  void operator()(int64_t v) const { attr_->set_l(v); }
   void operator()(boost::blank) const { PADDLE_THROW("Unexpected branch"); }
 };
 

paddle/framework/type_defs.h

@@ -35,7 +35,7 @@ using VariableNameMap = std::map<std::string, std::vector<std::string>>;
 using Attribute =
     boost::variant<boost::blank, int, float, std::string, std::vector<int>,
                    std::vector<float>, std::vector<std::string>, bool,
-                   std::vector<bool>, BlockDesc*>;
+                   std::vector<bool>, BlockDesc*, int64_t>;
 
 using AttributeMap = std::unordered_map<std::string, Attribute>;
 

paddle/operators/lookup_table_op.cc

@@ -66,6 +66,12 @@ class LookupTableOpMaker : public framework::OpProtoAndCheckerMaker {
                   "(boolean, default false) "
                   "Sparse update")
         .SetDefault(false);
+    AddAttr<int64_t>("padding_idx",
+                     "(int64, default -1) "
+                     "If the value is -1, it makes no effect to lookup. "
+                     "Otherwise the given value indicates padding the output "
+                     "with zeros whenever lookup encounters it in Ids.")
+        .SetDefault(-1);
     AddComment(R"DOC(
 Lookup Table Operator.
 

paddle/operators/lookup_table_op.cu

@@ -21,9 +21,11 @@ limitations under the License. */
 namespace paddle {
 namespace operators {
 
-template <typename T, int BlockDimX, int BlockDimY, int GridDimX>
+template <typename T, int BlockDimX, int BlockDimY, int GridDimX,
+          bool PaddingFlag>
 __global__ void LookupTable(T* output, const T* table, const int64_t* ids,
-                            const int64_t N, const int64_t K, const int64_t D) {
+                            const int64_t N, const int64_t K, const int64_t D,
+                            const int64_t padding_idx) {
   int idx = threadIdx.x;
   int idy = blockIdx.x + threadIdx.y * GridDimX;
 
@@ -34,8 +36,15 @@ __global__ void LookupTable(T* output, const T* table, const int64_t* ids,
     T* out = output + idy * D;
     const T* tab = table + id * D;
     for (int i = idx; i < D; i += BlockDimX) {
+      if (PaddingFlag) {
+        if (id == padding_idx)
+          out[i] = static_cast<T>(0);
+        else
+          out[i] = tab[i];
+      } else {
         out[i] = tab[i];
       }
+    }
     idy += BlockDimY * GridDimX;
   }
 }
@@ -67,6 +76,7 @@ class LookupTableCUDAKernel : public framework::OpKernel<T> {
     auto* table_t = context.Input<LoDTensor>("W");
     auto* ids_t = context.Input<LoDTensor>("Ids");
     auto* output_t = context.Output<LoDTensor>("Out");
+    int64_t padding_idx = context.Attr<int64_t>("padding_idx");
 
     size_t N = table_t->dims()[0];
     size_t D = table_t->dims()[1];
@@ -77,10 +87,17 @@ class LookupTableCUDAKernel : public framework::OpKernel<T> {
 
     dim3 threads(128, 8);
     dim3 grids(8, 1);
+
+    if (padding_idx == -1)
+      LookupTable<
+          T, 128, 8, 8,
+          false><<<grids, threads, 0, context.cuda_device_context().stream()>>>(
+          output, table, ids, N, K, D, padding_idx);
+    else
       LookupTable<
-        T, 128, 8,
-        8><<<grids, threads, 0, context.cuda_device_context().stream()>>>(
-        output, table, ids, N, K, D);
+          T, 128, 8, 8,
+          true><<<grids, threads, 0, context.cuda_device_context().stream()>>>(
+          output, table, ids, N, K, D, padding_idx);
   }
 };
 
@@ -91,6 +108,8 @@ class LookupTableGradCUDAKernel : public framework::OpKernel<T> {
     auto& dev_ctx =
         context.template device_context<platform::CUDADeviceContext>();
     bool is_sparse = context.Attr<bool>("is_sparse");
+    // Since paddings are not trainable and fixed in forward, the gradient of
+    // paddings makes no sense and we don't deal with it in backward.
     if (is_sparse) {
       auto* ids = context.Input<LoDTensor>("Ids");
       auto* table = context.Input<LoDTensor>("W");

paddle/operators/lookup_table_op.h

@@ -32,18 +32,32 @@ class LookupTableKernel : public framework::OpKernel<T> {
     auto* table_t = context.Input<LoDTensor>("W");      // float tensor
     auto* ids_t = context.Input<LoDTensor>("Ids");      // int tensor
     auto* output_t = context.Output<LoDTensor>("Out");  // float tensor
+    int64_t padding_idx = context.Attr<int64_t>("padding_idx");
 
     int N = table_t->dims()[0];
     int D = table_t->dims()[1];
     auto* ids = ids_t->data<int64_t>();
     auto* table = table_t->data<T>();
     auto* output = output_t->mutable_data<T>(context.GetPlace());
+
+    if (padding_idx == -1) {
+      for (int64_t i = 0; i < ids_t->numel(); ++i) {
+        PADDLE_ENFORCE_LT(ids[i], N);
+        PADDLE_ENFORCE_GE(ids[i], 0);
+        memcpy(output + i * D, table + ids[i] * D, D * sizeof(T));
+      }
+    } else {
       for (int64_t i = 0; i < ids_t->numel(); ++i) {
+        if (ids[i] == padding_idx) {
+          memset(output + i * D, 0, D * sizeof(T));
+        } else {
           PADDLE_ENFORCE_LT(ids[i], N);
           PADDLE_ENFORCE_GE(ids[i], 0);
           memcpy(output + i * D, table + ids[i] * D, D * sizeof(T));
         }
       }
+    }
+  }
 };
 
 template <typename T>
@@ -51,6 +65,8 @@ class LookupTableGradKernel : public framework::OpKernel<T> {
  public:
   void Compute(const framework::ExecutionContext& context) const override {
     bool is_sparse = context.Attr<bool>("is_sparse");
+    // Since paddings are not trainable and fixed in forward, the gradient of
+    // paddings makes no sense and we don't deal with it in backward.
     if (is_sparse) {
       auto* ids = context.Input<LoDTensor>("Ids");
       auto* table = context.Input<LoDTensor>("W");

paddle/pybind/print_operators_doc.cc

@@ -64,6 +64,8 @@ std::string AttrType(paddle::framework::proto::AttrType at) {
       return "bool array";
     case paddle::framework::proto::BLOCK:
       return "block id";
+    case paddle::framework::proto::LONG:
+      return "long";
   }
   return "UNKNOWN";  // not possible
 }

python/paddle/v2/fluid/layers/nn.py

@@ -185,20 +185,33 @@ def fc(input,
     return helper.append_activation(pre_activation)
 
 
-def embedding(input, size, is_sparse=False, param_attr=None, dtype='float32'):
+def embedding(input,
+              size,
+              is_sparse=False,
+              padding_idx=None,
+              param_attr=None,
+              dtype='float32'):
     """
     **Embedding Layer**
 
-    This layer is used to lookup a vector of IDs, provided by *input*, in a lookup table.
-    The result of this lookup is the embedding of each ID in the *input*.
+    This layer is used to lookup embeddings of IDs, provided by :attr:`input`, in 
+    a lookup table. The result of this lookup is the embedding of each ID in the
+    :attr:`input`.
 
     All the input variables are passed in as local variables to the LayerHelper
     constructor.
 
     Args:
-       input(Variable): Input to the function
-       size(tuple|list|None): Shape of the look up table parameter
-       is_sparse(bool): Boolean flag that specifying whether the input is sparse
+        input(Variable): The tensor variable containing the IDs.
+        size(tuple|list): The shape of the look up table parameter. It should
+            have two elements which indicate the size of the dictionary of
+            embeddings and the size of each embedding vector respectively.
+        is_sparse(bool): The flag indicating whether to use sparse update.
+        padding_idx(int|long|None): If :attr:`None`, it makes no effect to lookup.
+            Otherwise the given :attr:`padding_idx` indicates padding the output
+            with zeros whenever lookup encounters it in :attr:`input`. If 
+            :math:`padding_idx < 0`, the padding_idx to use in lookup is 
+            :math:`size[0] + dim`.
         param_attr(ParamAttr): Parameters for this layer
         dtype(np.dtype|core.DataType|str): The type of data : float32, float_16, int etc
 
@@ -218,12 +231,15 @@ def embedding(input, size, is_sparse=False, param_attr=None, dtype='float32'):
     w = helper.create_parameter(
         attr=helper.param_attr, shape=size, dtype=dtype, is_bias=False)
     tmp = helper.create_tmp_variable(dtype)
+    padding_idx = -1 if padding_idx is None else padding_idx if padding_idx >= 0 else (
+        size[0] + padding_idx)
     helper.append_op(
         type='lookup_table',
         inputs={'Ids': input,
                 'W': w},
         outputs={'Out': tmp},
-        attrs={'is_sparse': is_sparse})
+        attrs={'is_sparse': is_sparse,
+               'padding_idx': padding_idx})
     return tmp
 
 

python/paddle/v2/fluid/tests/test_lookup_table_op.py

@@ -33,5 +33,19 @@ class TestLookupTableOp(OpTest):
         self.check_grad(['W'], 'Out', no_grad_set=set('Ids'))
 
 
+class TestLookupTableOpWithPadding(TestLookupTableOp):
+    def test_check_output(self):
+        ids = np.squeeze(self.inputs['Ids'])
+        padding_idx = np.random.choice(ids, 1)[0]
+        self.outputs['Out'][ids == padding_idx] = np.zeros(31)
+        self.attrs = {'padding_idx': long(padding_idx)}
+        self.check_output()
+
+    def test_check_grad(self):
+        # Since paddings are not trainable and fixed in forward, the gradient of 
+        # paddings makes no sense and we don't test the gradient here.
+        pass
+
+
 if __name__ == "__main__":
     unittest.main()