!1921 Adapt TBE ops ApplyProximalAdagrad for GE.

Merge pull request !1921 from liuxiao/ApplyProximalAdagrad-for-GE

mindspore/ccsrc/transform/convert.cc

@@ -182,6 +182,7 @@ const char kNameBinaryCrossEntropy[] = "BinaryCrossEntropy";
 const char kNameBinaryCrossEntropyGrad[] = "BinaryCrossEntropyGrad";
 const char kNameSparseApplyAdagrad[] = "SparseApplyAdagrad";
 const char kNameSparseApplyFtrlD[] = "SparseApplyFtrlD";
+const char kNameApplyProximalAdagrad[] = "ApplyProximalAdagrad";
 const char kNameAcosh[] = "Acosh";
 const char kNameAcoshGrad[] = "AcoshGrad";
 const char kNameFloorMod[] = "FloorMod";
@@ -386,6 +387,7 @@ std::unordered_map<std::string, OpAdapterDescPtr> &DfGraphConvertor::get_adpt_ma
     {string(kNameBinaryCrossEntropyGrad), ADPT_DESC(BinaryCrossEntropyGrad)},
     {string(kNameSparseApplyAdagrad), ADPT_DESC(SparseApplyAdagradD)},
     {string(kNameSparseApplyFtrlD), ADPT_DESC(SparseApplyFtrlD)},
+    {string(kNameApplyProximalAdagrad), ADPT_DESC(ApplyProximalAdagrad)},
     {string(kNameAcosh), ADPT_DESC(Acosh)},
     {string(kNameAcoshGrad), ADPT_DESC(AcoshGrad)},
     {string(kNameFloorMod), ADPT_DESC(FloorMod)},

mindspore/ccsrc/transform/op_declare.cc

@@ -1155,6 +1155,12 @@ ATTR_MAP(SparseApplyAdagradD) = {{"lr", ATTR_DESC(lr, AnyTraits<float>())},
                                  {"use_locking", ATTR_DESC(use_locking, AnyTraits<bool>())}};
 OUTPUT_MAP(SparseApplyAdagradD) = {{0, OUTPUT_DESC(var)}};
 
+// ApplyProximalAdagrad
+INPUT_MAP(ApplyProximalAdagrad) = {{1, INPUT_DESC(var)}, {2, INPUT_DESC(accum)}, {3, INPUT_DESC(lr)},
+                                   {4, INPUT_DESC(l1)},  {5, INPUT_DESC(l2)},    {6, INPUT_DESC(grad)}};
+ATTR_MAP(ApplyProximalAdagrad) = {{"use_locking", ATTR_DESC(use_locking, AnyTraits<bool>())}};
+OUTPUT_MAP(ApplyProximalAdagrad) = {{0, OUTPUT_DESC(var)}};
+
 // SparseApplyFtrlD
 INPUT_MAP(SparseApplyFtrlD) = {{1, INPUT_DESC(var)},
                                {2, INPUT_DESC(accum)},

mindspore/ccsrc/transform/op_declare.h

@@ -442,6 +442,8 @@ DECLARE_OP_ADAPTER(BinaryCrossEntropyGrad)
 DECLARE_OP_USE_OUTPUT(BinaryCrossEntropyGrad)
 DECLARE_OP_ADAPTER(SparseApplyAdagradD)
 DECLARE_OP_USE_OUTPUT(SparseApplyAdagradD)
+DECLARE_OP_ADAPTER(ApplyProximalAdagrad)
+DECLARE_OP_USE_OUTPUT(ApplyProximalAdagrad)
 DECLARE_OP_ADAPTER(SpaceToDepth)
 DECLARE_OP_USE_OUTPUT(SpaceToDepth)
 DECLARE_OP_ADAPTER(DepthToSpace)

mindspore/ops/operations/nn_ops.py

@@ -3149,7 +3149,7 @@ class SparseApplyFtrl(PrimitiveWithInfer):
         validator.check_value_type("l1", l1, [float], self.name)
         validator.check_value_type("l2", l2, [float], self.name)
         validator.check_value_type("lr_power", lr_power, [float], self.name)
-        self.lr = validator.check_number("lr", lr, 0.0, Rel.GT, self.name)
+        self.lr = validator.check_number_range("lr", lr, 0.0, float("inf"), Rel.INC_LEFT, self.name)
         self.l1 = validator.check_number("l1", l1, 0.0, Rel.GE, self.name)
         self.l2 = validator.check_number("l2", l2, 0.0, Rel.GE, self.name)
         self.lr_power = validator.check_number("lr_power", lr_power, 0, Rel.LE, self.name)

tests/ut/python/ops/test_ops.py

@@ -244,12 +244,14 @@ class SparseApplyProximalAdagradNet(nn.Cell):
     def __init__(self):
         super(SparseApplyProximalAdagradNet, self).__init__()
         self.sparse_apply_proximal_adagrad = P.SparseApplyProximalAdagrad()
+        self.var = Parameter(Tensor(np.random.rand(3, 3).astype(np.float32)), name="var")
+        self.accum = Parameter(Tensor(np.random.rand(3, 3).astype(np.float32)), name="accum")
         self.lr = 0.01
         self.l1 = 0.0
         self.l2 = 0.0
 
-    def construct(self, var, accum, grad, indices):
-        out = self.sparse_apply_proximal_adagrad(var, accum, self.lr, self.l1, self.l2, grad, indices)
+    def construct(self, grad, indices):
+        out = self.sparse_apply_proximal_adagrad(self.var, self.accum, self.lr, self.l1, self.l2, grad, indices)
         return out
 
 
@@ -257,12 +259,14 @@ class ApplyProximalAdagradNet(nn.Cell):
     def __init__(self):
         super(ApplyProximalAdagradNet, self).__init__()
         self.apply_proximal_adagrad = P.ApplyProximalAdagrad()
+        self.var = Parameter(Tensor(np.random.rand(3, 3).astype(np.float32)), name="var")
+        self.accum = Parameter(Tensor(np.random.rand(3, 3).astype(np.float32)), name="accum")
         self.lr = 0.01
         self.l1 = 0.0
         self.l2 = 0.0
 
-    def construct(self, var, accum, grad):
-        out = self.apply_proximal_adagrad(var, accum, self.lr, self.l1, self.l2, grad)
+    def construct(self, grad):
+        out = self.apply_proximal_adagrad(self.var, self.accum, self.lr, self.l1, self.l2, grad)
         return out
 
 
@@ -1061,11 +1065,11 @@ test_case_nn_ops = [
         'skip': ['backward']}),
     ('ApplyProximalAdagrad', {
         'block': ApplyProximalAdagradNet(),
-        'desc_inputs': [[3, 3], [3, 3], [3, 3]],
+        'desc_inputs': [[3, 3]],
         'skip': ['backward']}),
     ('SparseApplyProximalAdagrad', {
         'block': SparseApplyProximalAdagradNet(),
-        'desc_inputs': [[3, 3], [3, 3], [3, 3], Tensor(np.ones((3,), np.int32))],
+        'desc_inputs': [[3, 3], Tensor(np.ones((3,), np.int32))],
         'skip': ['backward']}),
     ('Flatten_1', {
         'block': NetForFlatten(),