bugfix: RNN does not initialize the state for the cell correctly (#27644)

1. fix a bug that paddle.nn.RNN does not initialize the state for the cell correctly;
2. add unittest for paddle.nn.RNN and paddle.nn.BiRNN

python/paddle/fluid/layers/rnn.py

@@ -531,6 +531,10 @@ def _rnn_dynamic_graph(cell,
     flat_inputs = flatten(inputs)
     time_steps = flat_inputs[0].shape[time_step_index]
 
+    if initial_states is None:
+        initial_states = cell.get_initial_states(
+            batch_ref=inputs, batch_dim_idx=1 if time_major else 0)
+
     if not time_major:
         inputs = map_structure(_transpose_batch_time, inputs)
 

python/paddle/fluid/tests/unittests/rnn/rnn_numpy.py

@@ -61,8 +61,8 @@ class SimpleRNNCell(LayerMixin):
             self.bias_ih = None
             self.bias_hh = None
 
-    def init_state(self, inputs):
-        batch_size = inputs.shape[0]
+    def init_state(self, inputs, batch_dim_index=0):
+        batch_size = inputs.shape[batch_dim_index]
         return np.zeros((batch_size, self.hidden_size), dtype=inputs.dtype)
 
     def forward(self, inputs, hx=None):
@@ -103,8 +103,8 @@ class GRUCell(LayerMixin):
             self.bias_ih = None
             self.bias_hh = None
 
-    def init_state(self, inputs):
-        batch_size = inputs.shape[0]
+    def init_state(self, inputs, batch_dim_index=0):
+        batch_size = inputs.shape[batch_dim_index]
         return np.zeros((batch_size, self.hidden_size), dtype=inputs.dtype)
 
     def forward(self, inputs, hx=None):
@@ -117,7 +117,6 @@ class GRUCell(LayerMixin):
         h_gates = np.matmul(pre_hidden, self.weight_hh.T)
         if self.bias_hh is not None:
             h_gates = h_gates + self.bias_hh
-
         x_r, x_z, x_c = np.split(x_gates, 3, 1)
         h_r, h_z, h_c = np.split(h_gates, 3, 1)
 
@@ -152,8 +151,8 @@ class LSTMCell(LayerMixin):
             self.bias_ih = None
             self.bias_hh = None
 
-    def init_state(self, inputs):
-        batch_size = inputs.shape[0]
+    def init_state(self, inputs, batch_dim_index=0):
+        batch_size = inputs.shape[batch_dim_index]
         init_h = np.zeros((batch_size, self.hidden_size), dtype=inputs.dtype)
         init_c = np.zeros((batch_size, self.hidden_size), dtype=inputs.dtype)
         return init_h, init_c
@@ -206,6 +205,9 @@ def rnn(cell,
     if is_reverse:
         inputs = np.flip(inputs, 0)
 
+    if initial_states is None:
+        initial_states = cell.init_state(inputs, 1)
+
     if sequence_length is None:
         mask = None
     else:

python/paddle/fluid/tests/unittests/rnn/test_rnn_cells_static.py

@@ -14,6 +14,7 @@
 
 import paddle
 paddle.framework.set_default_dtype("float64")
+paddle.enable_static()
 
 import numpy as np
 import unittest

python/paddle/fluid/tests/unittests/rnn/test_rnn_nets_static.py

@@ -15,6 +15,7 @@
 import paddle
 paddle.set_default_dtype("float64")
 from paddle.fluid.layers import sequence_mask
+paddle.enable_static()
 
 import numpy as np
 import unittest

python/paddle/fluid/tests/unittests/rnn/test_wrappers.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import paddle
+paddle.set_default_dtype("float64")
+from paddle.fluid.layers import sequence_mask
+
+import numpy as np
+import unittest
+
+from convert import convert_params_for_cell
+from rnn_numpy import GRUCell, RNN, BiRNN
+
+
+class TestRNNWrapper(unittest.TestCase):
+    def __init__(self, time_major=True, direction="forward", place="cpu"):
+        super(TestRNNWrapper, self).__init__("runTest")
+        self.time_major = time_major
+        self.direction = direction
+        self.place = paddle.CPUPlace() if place == "cpu" \
+            else paddle.CUDAPlace(0)
+
+    def setUp(self):
+        paddle.disable_static(self.place)
+        cell1 = GRUCell(16, 32)
+        cell2 = paddle.nn.GRUCell(16, 32)
+        convert_params_for_cell(cell1, cell2)
+        rnn1 = RNN(cell1,
+                   is_reverse=self.direction == "backward",
+                   time_major=self.time_major)
+        rnn2 = paddle.nn.RNN(cell2,
+                             is_reverse=self.direction == "backward",
+                             time_major=self.time_major)
+
+        self.rnn1 = rnn1
+        self.rnn2 = rnn2
+
+    def test_with_initial_state(self):
+        rnn1 = self.rnn1
+        rnn2 = self.rnn2
+
+        x = np.random.randn(12, 4, 16)
+        if not self.time_major:
+            x = np.transpose(x, [1, 0, 2])
+        prev_h = np.random.randn(4, 32)
+
+        y1, h1 = rnn1(x, prev_h)
+        y2, h2 = rnn2(paddle.to_tensor(x), paddle.to_tensor(prev_h))
+        np.testing.assert_allclose(y1, y2.numpy(), atol=1e-8, rtol=1e-5)
+        np.testing.assert_allclose(h1, h2.numpy(), atol=1e-8, rtol=1e-5)
+
+    def test_with_zero_state(self):
+        rnn1 = self.rnn1
+        rnn2 = self.rnn2
+
+        x = np.random.randn(12, 4, 16)
+        if not self.time_major:
+            x = np.transpose(x, [1, 0, 2])
+
+        y1, h1 = rnn1(x)
+        y2, h2 = rnn2(paddle.to_tensor(x))
+        np.testing.assert_allclose(y1, y2.numpy(), atol=1e-8, rtol=1e-5)
+        np.testing.assert_allclose(h1, h2.numpy(), atol=1e-8, rtol=1e-5)
+
+    def test_with_input_lengths(self):
+        rnn1 = self.rnn1
+        rnn2 = self.rnn2
+
+        x = np.random.randn(12, 4, 16)
+        if not self.time_major:
+            x = np.transpose(x, [1, 0, 2])
+        sequence_length = np.array([12, 10, 9, 8], dtype=np.int64)
+
+        y1, h1 = rnn1(x, sequence_length=sequence_length)
+
+        seq_len = paddle.to_tensor(sequence_length)
+        mask = sequence_mask(seq_len, dtype=paddle.get_default_dtype())
+        if self.time_major:
+            mask = paddle.transpose(mask, [1, 0])
+        y2, h2 = rnn2(paddle.to_tensor(x), sequence_length=seq_len)
+        y2 = paddle.multiply(y2, mask, axis=0)
+
+        np.testing.assert_allclose(y1, y2.numpy(), atol=1e-8, rtol=1e-5)
+        np.testing.assert_allclose(h1, h2.numpy(), atol=1e-8, rtol=1e-5)
+
+    def runTest(self):
+        self.test_with_initial_state()
+        self.test_with_zero_state()
+        self.test_with_input_lengths()
+
+
+class TestBiRNNWrapper(unittest.TestCase):
+    def __init__(self, time_major=True, place="cpu"):
+        super(TestBiRNNWrapper, self).__init__("runTest")
+        self.time_major = time_major
+        self.place = paddle.CPUPlace() if place == "cpu" \
+            else paddle.CUDAPlace(0)
+
+    def setUp(self):
+        paddle.disable_static(self.place)
+        fw_cell1 = GRUCell(16, 32)
+        bw_cell1 = GRUCell(16, 32)
+        fw_cell2 = paddle.nn.GRUCell(16, 32)
+        bw_cell2 = paddle.nn.GRUCell(16, 32)
+        convert_params_for_cell(fw_cell1, fw_cell2)
+        convert_params_for_cell(bw_cell1, bw_cell2)
+        rnn1 = BiRNN(fw_cell1, bw_cell1, time_major=self.time_major)
+        rnn2 = paddle.nn.BiRNN(fw_cell2, bw_cell2, time_major=self.time_major)
+
+        self.rnn1 = rnn1
+        self.rnn2 = rnn2
+
+    def test_with_initial_state(self):
+        rnn1 = self.rnn1
+        rnn2 = self.rnn2
+
+        x = np.random.randn(12, 4, 16)
+        if not self.time_major:
+            x = np.transpose(x, [1, 0, 2])
+        fw_prev_h = np.random.randn(4, 32)
+        bw_prev_h = np.random.randn(4, 32)
+
+        y1, (fw_h1, bw_h1) = rnn1(x, (fw_prev_h, bw_prev_h))
+        y2, (fw_h2, bw_h2) = rnn2(
+            paddle.to_tensor(x),
+            (paddle.to_tensor(fw_prev_h), paddle.to_tensor(bw_prev_h)))
+        np.testing.assert_allclose(y1, y2.numpy(), atol=1e-8, rtol=1e-5)
+        np.testing.assert_allclose(fw_h1, fw_h2.numpy(), atol=1e-8, rtol=1e-5)
+        np.testing.assert_allclose(bw_h1, bw_h2.numpy(), atol=1e-8, rtol=1e-5)
+
+    def test_with_zero_state(self):
+        rnn1 = self.rnn1
+        rnn2 = self.rnn2
+
+        x = np.random.randn(12, 4, 16)
+        if not self.time_major:
+            x = np.transpose(x, [1, 0, 2])
+
+        y1, (fw_h1, bw_h1) = rnn1(x)
+        y2, (fw_h2, bw_h2) = rnn2(paddle.to_tensor(x))
+        np.testing.assert_allclose(y1, y2.numpy(), atol=1e-8, rtol=1e-5)
+        np.testing.assert_allclose(fw_h1, fw_h2.numpy(), atol=1e-8, rtol=1e-5)
+        np.testing.assert_allclose(bw_h1, bw_h2.numpy(), atol=1e-8, rtol=1e-5)
+
+    def test_with_input_lengths(self):
+        rnn1 = self.rnn1
+        rnn2 = self.rnn2
+
+        x = np.random.randn(12, 4, 16)
+        if not self.time_major:
+            x = np.transpose(x, [1, 0, 2])
+        sequence_length = np.array([12, 10, 9, 8], dtype=np.int64)
+
+        y1, (fw_h1, bw_h1) = rnn1(x, sequence_length=sequence_length)
+
+        seq_len = paddle.to_tensor(sequence_length)
+        mask = sequence_mask(seq_len, dtype=paddle.get_default_dtype())
+        if self.time_major:
+            mask = paddle.transpose(mask, [1, 0])
+        y2, (fw_h2, bw_h2) = rnn2(paddle.to_tensor(x), sequence_length=seq_len)
+        y2 = paddle.multiply(y2, mask, axis=0)
+
+        np.testing.assert_allclose(y1, y2.numpy(), atol=1e-8, rtol=1e-5)
+        np.testing.assert_allclose(fw_h1, fw_h2.numpy(), atol=1e-8, rtol=1e-5)
+        np.testing.assert_allclose(bw_h1, bw_h2.numpy(), atol=1e-8, rtol=1e-5)
+
+    def runTest(self):
+        self.test_with_initial_state()
+        self.test_with_zero_state()
+        self.test_with_input_lengths()
+
+
+def load_tests(loader, tests, pattern):
+    suite = unittest.TestSuite()
+    devices = ["cpu", "gpu"] if paddle.fluid.is_compiled_with_cuda() \
+        else ["cpu"]
+    for direction in ["forward", "backward"]:
+        for device in devices:
+            for time_major in [False]:
+                suite.addTest(TestRNNWrapper(time_major, direction, device))
+            suite.addTest(TestBiRNNWrapper(time_major, device))
+    return suite

python/paddle/nn/layer/rnn.py

@@ -288,7 +288,7 @@ class SimpleRNNCell(RNNCellBase):
             `weight_hh`. Default: None.
         bias_ih_attr (ParamAttr, optional): The parameter attribute for the 
             `bias_ih`. Default: None.
-        bias_ih_attr (ParamAttr, optional): The parameter attribute for the 
+        bias_hh_attr (ParamAttr, optional): The parameter attribute for the 
             `bias_hh`. Default: None.
         name (str, optional): Name for the operation (optional, default is 
             None). For more information, please refer to :ref:`api_guide_Name`.
@@ -429,7 +429,7 @@ class LSTMCell(RNNCellBase):
             `weight_hh`. Default: None.
         bias_ih_attr (ParamAttr, optional): The parameter attribute for the 
             `bias_ih`. Default: None.
-        bias_ih_attr (ParamAttr, optional): The parameter attribute for the 
+        bias_hh_attr (ParamAttr, optional): The parameter attribute for the 
             `bias_hh`. Default: None.
         name (str, optional): Name for the operation (optional, default is 
             None). For more information, please refer to :ref:`api_guide_Name`.
@@ -582,7 +582,7 @@ class GRUCell(RNNCellBase):
             `weight_hh`. Default: None.
         bias_ih_attr (ParamAttr, optional): The parameter attribute for the 
             `bias_ih`. Default: None.
-        bias_ih_attr (ParamAttr, optional): The parameter attribute for the 
+        bias_hh_attr (ParamAttr, optional): The parameter attribute for the 
             `bias_hh`. Default: None.
         name (str, optional): Name for the operation (optional, default is 
             None). For more information, please refer to :ref:`api_guide_Name`.
@@ -778,12 +778,6 @@ class RNN(Layer):
                 initial_states=None,
                 sequence_length=None,
                 **kwargs):
-        if initial_states is None:
-            initial_states = self.cell.get_initial_states(
-                batch_ref=inputs,
-                dtype=inputs.dtype,
-                batch_dim_idx=self.batch_index)
-
         final_outputs, final_states = F.rnn(self.cell,
                                             inputs,
                                             initial_states=initial_states,
@@ -880,8 +874,6 @@ class BiRNN(Layer):
         if isinstance(initial_states, (list, tuple)):
             assert len(initial_states) == 2, \
                 "length of initial_states should be 2 when it is a list/tuple"
-        else:
-            initial_states = [initial_states, initial_states]
 
         outputs, final_states = F.birnn(self.cell_fw, self.cell_bw, inputs,
                                         initial_states, sequence_length,
@@ -968,7 +960,7 @@ class SimpleRNN(RNNMixin):
             `weight_hh` of each cell. Defaults to None.
         bias_ih_attr (ParamAttr, optional): The parameter attribute for the 
             `bias_ih` of each cells. Defaults to None.
-        bias_ih_attr (ParamAttr, optional): The parameter attribute for the 
+        bias_hh_attr (ParamAttr, optional): The parameter attribute for the 
             `bias_hh` of each cells. Defaults to None.
         name (str, optional): Name for the operation (optional, default is 
             None). For more information, please refer to :ref:`api_guide_Name`.
@@ -1111,7 +1103,7 @@ class LSTM(RNNMixin):
             `weight_hh` of each cell. Default: None.
         bias_ih_attr (ParamAttr, optional): The parameter attribute for the 
             `bias_ih` of each cells. Default: None.
-        bias_ih_attr (ParamAttr, optional): The parameter attribute for the 
+        bias_hh_attr (ParamAttr, optional): The parameter attribute for the 
             `bias_hh` of each cells. Default: None.
         name (str, optional): Name for the operation (optional, default is 
             None). For more information, please refer to :ref:`api_guide_Name`.
@@ -1247,7 +1239,7 @@ class GRU(RNNMixin):
             `weight_hh` of each cell. Default: None.
         bias_ih_attr (ParamAttr, optional): The parameter attribute for the 
             `bias_ih` of each cells. Default: None.
-        bias_ih_attr (ParamAttr, optional): The parameter attribute for the 
+        bias_hh_attr (ParamAttr, optional): The parameter attribute for the 
             `bias_hh` of each cells. Default: None.
         name (str, optional): Name for the operation (optional, default is 
             None). For more information, please refer to :ref:`api_guide_Name`.