Merge pull request #5623 from guoshengCS/fix-H0-GRUOp

Fix data order of H0 in GRU Operator

Author: guoshengCS

paddle/operators/gru_op.h

@@ -24,16 +24,24 @@
 namespace paddle {
 namespace operators {
 
-using Tensor = framework::Tensor;
 using LoDTensor = framework::LoDTensor;
+using Tensor = framework::Tensor;
+
+template <typename Place, typename T>
+inline void ReorderInitState(const platform::DeviceContext& ctx,
+                             const framework::Tensor& src, const size_t* index,
+                             framework::Tensor* dst, bool indexed_src) {
+  math::CopyMatrixRowsFunctor<Place, T> row_shuffle;
+  dst->mutable_data<T>(src.dims(), ctx.GetPlace());
+  row_shuffle(ctx, src, index, *dst, indexed_src);
+}
 
 template <typename Place, typename T>
 class GRUKernel : public framework::OpKernel<T> {
  public:
   void BatchCompute(const framework::ExecutionContext& context) const {
     auto* input = context.Input<LoDTensor>("Input");
     auto* h0 = context.Input<Tensor>("H0");
-    const T* h0_data = h0 ? h0->data<T>() : nullptr;
     auto* weight = context.Input<Tensor>("Weight");
     const T* weight_data = weight->data<T>();
     auto* bias = context.Input<Tensor>("Bias");
@@ -66,7 +74,18 @@ class GRUKernel : public framework::OpKernel<T> {
     gru_value.gateWeight = const_cast<T*>(weight_data);
     gru_value.stateWeight =
         const_cast<T*>(weight_data + 2 * frame_size * frame_size);
-    gru_value.prevOutValue = const_cast<T*>(h0_data);
+    Tensor ordered_h0;
+    const size_t* order = batch_gate->lod()[2].data();
+    if (h0) {
+      // Since the batch computing for GRU reorders the input sequences
+      // according to their length. The initialized cell state also needs
+      // to reorder.
+      ReorderInitState<Place, T>(context.device_context(), *h0, order,
+                                 &ordered_h0, true);
+      gru_value.prevOutValue = ordered_h0.data<T>();
+    } else {
+      gru_value.prevOutValue = nullptr;
+    }
     auto batch_starts = batch_gate->lod()[0];
     size_t num_batch = batch_starts.size() - 1;
     for (size_t n = 0; n < num_batch; n++) {
@@ -102,7 +121,6 @@ class GRUGradKernel : public framework::OpKernel<T> {
  public:
   void BatchCompute(const framework::ExecutionContext& context) const {
     auto* h0 = context.Input<Tensor>("H0");
-    const T* h0_data = h0 ? h0->data<T>() : nullptr;
     auto* weight = context.Input<Tensor>("Weight");
     const T* weight_data = weight->data<T>();
     auto* batch_gate = context.Input<LoDTensor>("BatchGate");
@@ -135,6 +153,17 @@ class GRUGradKernel : public framework::OpKernel<T> {
     zero(dev_ctx, &batch_gate_grad, static_cast<T>(0.0));
     zero(dev_ctx, &batch_reset_hidden_prev_grad, static_cast<T>(0.0));
 
+    Tensor ordered_h0, ordered_h0_grad;
+    const size_t* order = batch_gate->lod()[2].data();
+    if (h0) {
+      ReorderInitState<Place, T>(context.device_context(), *h0, order,
+                                 &ordered_h0, true);
+    }
+    if (h0_grad) {
+      ordered_h0_grad.mutable_data<T>(h0_grad->dims(), context.GetPlace());
+      zero(context.device_context(), &ordered_h0_grad, static_cast<T>(0.0));
+    }
+
     bool is_reverse = context.Attr<bool>("is_reverse");
     batch_hidden_grad.set_lod(batch_hidden->lod());
     to_batch(dev_ctx, *hidden_grad, batch_hidden_grad, false, is_reverse);
@@ -176,14 +205,9 @@ class GRUGradKernel : public framework::OpKernel<T> {
           batch_reset_hidden_prev_grad.Slice(bstart, bend);
       gru_grad.resetOutputGrad = reset_hidden_prev_grad_t.data<T>();
       if (n == 0) {
-        gru_value.prevOutValue = const_cast<T*>(h0_data);
-        if (h0_grad) {
-          T* h0_grad_data = h0_grad->mutable_data<T>(context.GetPlace());
-          zero(dev_ctx, h0_grad, static_cast<T>(0.0));
-          gru_grad.prevOutGrad = h0_grad_data;
-        } else {
-          gru_grad.prevOutGrad = nullptr;
-        }
+        gru_value.prevOutValue = h0 ? ordered_h0.data<T>() : nullptr;
+        gru_grad.prevOutGrad =
+            h0 && h0_grad ? ordered_h0_grad.data<T>() : nullptr;
       } else {
         int bstart_pre = static_cast<int>(batch_starts[n - 1]);
         Tensor hidden_prev_t = batch_hidden->Slice(bstart_pre, bstart);
@@ -208,6 +232,10 @@ class GRUGradKernel : public framework::OpKernel<T> {
       math::ColwiseSum<Place, T> col_sum;
       col_sum(dev_ctx, batch_gate_grad, bias_grad);
     }
+    if (h0 && h0_grad) {
+      ReorderInitState<Place, T>(context.device_context(), ordered_h0_grad,
+                                 order, h0_grad, false);
+    }
   }
 
   void Compute(const framework::ExecutionContext& context) const override {

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

@@ -6,7 +6,8 @@
 
 
 class TestGRUOp(OpTest):
-    batch_size = 9
+    lod = [[0, 2, 6, 9]]
+    batch_size = lod[0][-1]
     frame_size = 5
     activate = {
         'identity': identity,
@@ -35,7 +36,7 @@ def seq_to_batch(lod, is_reverse):
                            seq_starts[sorted_seqs[i]] + batch_idx)
                 idx_in_seq.append(idx)
             idx_in_seq_list.append(idx_in_seq)
-        return idx_in_seq_list
+        return idx_in_seq_list, sorted_seqs
 
     def gru_step(self, x, h_p, w, b):
         batch_size = x.shape[0]
@@ -66,8 +67,8 @@ def gru(self):
         batch_hidden = self.outputs['BatchHidden']
         hidden = self.outputs['Hidden']
         idx_in_seq_list = self.idx_in_seq_list
-        h_p = self.inputs['H0'] if self.inputs.has_key('H0') else np.zeros(
-            (len(idx_in_seq_list[0]), self.frame_size))
+        h_p = self.inputs['H0'][self.sorted_seqs] if self.inputs.has_key(
+            'H0') else np.zeros((len(idx_in_seq_list[0]), self.frame_size))
         num_batch = len(idx_in_seq_list)
         end_idx = 0
         for batch_idx in range(num_batch):
@@ -84,8 +85,9 @@ def gru(self):
         return batch_gate, batch_reset_hidden_prev, hidden
 
     def set_data(self):
-        lod = [[0, 2, 6, self.batch_size]]
-        self.idx_in_seq_list = self.seq_to_batch(lod, self.is_reverse)
+        lod = self.lod
+        self.idx_in_seq_list, self.sorted_seqs = self.seq_to_batch(
+            lod, self.is_reverse)
         batch_size = self.batch_size
         frame_size = self.frame_size
         input = np.random.rand(batch_size, frame_size * 3).astype('float64')
@@ -146,7 +148,7 @@ class TestGRUOpReverse(TestGRUOp):
     def set_confs(self):
         self.is_reverse = True
         self.attrs = {
-            'activation': 'identity',
+            'activation': 'tanh',
             'gate_activation': 'sigmoid',
             'is_reverse': self.is_reverse
         }