[cherry-pick] Op(prelu/relu/f.relu/f.log_softmax) error message enhancement (#23792) (#23964)

paddle/fluid/operators/prelu_op.cc

@@ -31,10 +31,11 @@ class PReluOp : public framework::OperatorWithKernel {
     auto x_dim = ctx->GetInputDim("X");
     std::string mode = ctx->Attrs().Get<std::string>("mode");
     if (mode == "all") {
-      PADDLE_ENFORCE_EQ(
-          product(ctx->GetInputDim("Alpha")), 1,
-          platform::errors::InvalidArgument(
-              "For mode 'all', size of weight Alpha must be one."));
+      PADDLE_ENFORCE_EQ(product(ctx->GetInputDim("Alpha")), 1,
+                        platform::errors::InvalidArgument(
+                            "For mode 'all', size of weight Alpha must be one. "
+                            "But recevied alpha's size: %d.",
+                            product(ctx->GetInputDim("Alpha"))));
     } else if (mode == "channel") {
       PADDLE_ENFORCE_EQ(product(ctx->GetInputDim("Alpha")), x_dim[1],
                         platform::errors::InvalidArgument(
@@ -67,7 +68,11 @@ class PReluOp : public framework::OperatorWithKernel {
               "x's size: %d.",
               alpha_product, x_product));
     } else {
-      PADDLE_THROW("Unkown mode %s", mode);
+      PADDLE_THROW(platform::errors::InvalidArgument(
+          "Attr(mode) of prelu must be one of 'all', 'channel', or 'element'. "
+          "But recevied "
+          "mode: '%s'.",
+          mode));
     }
     ctx->ShareDim("X", /*->*/ "Out");
     ctx->ShareLoD("X", /*->*/ "Out");

python/paddle/fluid/tests/unittests/test_activation_op.py

@@ -463,7 +463,7 @@ class TestReluOpError(unittest.TestCase):
     def test_errors(self):
         with program_guard(Program()):
             # The input type must be Variable.
-            self.assertRaises(TypeError, fluid.layers.sqrt, 1)
+            self.assertRaises(TypeError, fluid.layers.relu, 1)
             # The input dtype must be float16, float32, float64.
             x_int32 = fluid.data(name='x_int32', shape=[12, 10], dtype='int32')
             self.assertRaises(TypeError, fluid.layers.relu, x_int32)

python/paddle/fluid/tests/unittests/test_prelu_op.py

@@ -23,21 +23,18 @@ from paddle.fluid import Program, program_guard
 from op_test import OpTest, skip_check_grad_ci
 
 
-class TestPReluAPIError(unittest.TestCase):
+class TestPReluOpError(unittest.TestCase):
     def test_errors(self):
-        with fluid.program_guard(fluid.Program(), fluid.Program()):
-            layer = fluid.PRelu(
-                mode='all',
-                param_attr=fluid.ParamAttr(
-                    initializer=fluid.initializer.Constant(1.0)))
-            # the input must be Variable.
-            x0 = fluid.create_lod_tensor(
-                np.array([-1, 3, 5, 5]), [[1, 1, 1, 1]], fluid.CPUPlace())
-            self.assertRaises(TypeError, layer, x0)
-            # the input dtype must be float32
-            data_t = fluid.data(
-                name="input", shape=[5, 200, 100, 100], dtype="float64")
-            self.assertRaises(TypeError, layer, data_t)
+        with program_guard(Program()):
+            # The input type must be Variable.
+            self.assertRaises(TypeError, fluid.layers.prelu, 0.1, 'all')
+            # The input dtype must be float16, float32, float64.
+            x_int32 = fluid.data(name='x_int32', shape=[12, 10], dtype='int32')
+            self.assertRaises(TypeError, fluid.layers.prelu, x_int32, 'all')
+            # support the input dtype is float32
+            x_fp16 = fluid.layers.data(
+                name='x_fp16', shape=[12, 10], dtype='float32')
+            fluid.layers.prelu(x_fp16, 'all')
 
 
 class PReluTest(OpTest):
@@ -79,39 +76,55 @@ class PReluTest(OpTest):
         self.check_grad(['X', 'Alpha'], 'Out')
 
 
-# TODO(minqiyang): Resume these test cases after fixing Python3 CI job issues
-if six.PY2:
-
-    @skip_check_grad_ci(
-        reason="[skip shape check] Input(Alpha) must be 1-D and only has one data in 'all' mode"
-    )
-    class TestModeAll(PReluTest):
-        def init_input_shape(self):
-            self.x_shape = (2, 3, 4, 5)
-
-        def init_attr(self):
-            self.attrs = {'mode': "all"}
+@skip_check_grad_ci(
+    reason="[skip shape check] Input(Alpha) must be 1-D and only has one data in 'all' mode"
+)
+class TestModeAll(PReluTest):
+    def init_input_shape(self):
+        self.x_shape = (2, 3, 4, 5)
 
-    class TestModeElt(PReluTest):
-        def init_input_shape(self):
-            self.x_shape = (3, 2, 5, 10)
+    def init_attr(self):
+        self.attrs = {'mode': "all"}
 
-        def init_attr(self):
-            self.attrs = {'mode': "element"}
 
+class TestModeElt(PReluTest):
+    def init_input_shape(self):
+        self.x_shape = (3, 2, 5, 10)
 
-class TestPReluOpError(unittest.TestCase):
-    def test_errors(self):
-        with program_guard(Program()):
-            # The input type must be Variable.
-            self.assertRaises(TypeError, fluid.layers.prelu, 1, 'all')
-            # The input dtype must be float16, float32, float64.
-            x_int32 = fluid.data(name='x_int32', shape=[12, 10], dtype='int32')
-            self.assertRaises(TypeError, fluid.layers.prelu, x_int32, 'all')
-            # support the input dtype is float32
-            x_fp16 = fluid.layers.data(
-                name='x_fp16', shape=[12, 10], dtype='float32')
-            fluid.layers.prelu(x_fp16, 'all')
+    def init_attr(self):
+        self.attrs = {'mode': "element"}
+
+
+def prelu_t(x, mode, param_attr=None, name=None):
+    helper = fluid.layer_helper.LayerHelper('prelu', **locals())
+    alpha_shape = [1, x.shape[1], 1, 1]
+    dtype = helper.input_dtype(input_param_name='x')
+    alpha = helper.create_parameter(
+        attr=helper.param_attr,
+        shape=alpha_shape,
+        dtype='float32',
+        is_bias=False,
+        default_initializer=fluid.initializer.ConstantInitializer(0.25))
+    out = helper.create_variable_for_type_inference(dtype)
+    helper.append_op(
+        type="prelu",
+        inputs={"X": x,
+                'Alpha': alpha},
+        attrs={"mode": mode},
+        outputs={"Out": out})
+    return out
+
+
+# error message test if mode is not one of 'all', 'channel', 'element'
+class TestModeError(unittest.TestCase):
+    def test_mode_error(self):
+        main_program = Program()
+        with fluid.program_guard(main_program, Program()):
+            x = fluid.data(name='x', shape=[2, 3, 4, 5])
+            try:
+                y = prelu_t(x, 'any')
+            except Exception as e:
+                assert (e.args[0].find('InvalidArgumentError') != -1)
 
 
 if __name__ == "__main__":

python/paddle/nn/functional/activation.py

@@ -206,8 +206,10 @@ def relu(input, inplace=False, name=None):
             )
         return core.ops.relu(input)
 
-    helper = LayerHelper('relu', **locals())
+    check_variable_and_dtype(input, 'input', ['float16', 'float32', 'float64'],
+                             'relu')
 
+    helper = LayerHelper('relu', **locals())
     outs = input if inplace else helper.create_variable_for_type_inference(
         input.dtype)
     helper.append_op(type='relu', inputs={'X': [input]}, outputs={'Out': outs})
@@ -263,7 +265,7 @@ def sigmoid(input, inplace=False, name=None):
             )
         return core.ops.sigmoid(input)
 
-    check_variable_and_dtype(input, 'X', ['float16', 'float32', 'float64'],
+    check_variable_and_dtype(input, 'input', ['float16', 'float32', 'float64'],
                              'sigmoid')
     helper = LayerHelper("sigmoid", **locals())
     outputs = helper.create_variable_for_type_inference(input.dtype)
@@ -329,8 +331,11 @@ def log_softmax(input, axis=None, dtype=None, name=None):
                                         False)
         return core.ops.log(outs_softmax)
 
-    helper = LayerHelper("log_softmax", **locals())
+    if dtype is None:
+        check_variable_and_dtype(
+            input, 'input', ['float16', 'float32', 'float64'], 'log_softmax')
 
+    helper = LayerHelper("log_softmax", **locals())
     outs_cast = input
     if dtype is not None:
         outs_cast = helper.create_variable_for_type_inference(dtype)