Fix/PowBackwardKernel

Context

Fixes Pow Backward Kernel

Fixes #22 (closed)

Modified files

• PowImpl_backward_kernels.hpp, PowImpl.hpp, PowImpl.cpp and Test_PowImpl.cpp, add Pow Backward kernel and unit tests;

TODO

• Fix Pow Backward kernel

unit_tests/operator/Test_PowImpl.cpp

@@ -313,5 +313,171 @@ TEST_CASE("[cpu/operator] Pow", "[Pow][CPU]") {
             std::cout << "total time: " << duration.count() << "μs" << std::endl;
         }
     }
+
+
+    SECTION("PowImpl_cpu::backward()") {
+        SECTION("3D Tensors") {
+            const auto input0 = std::make_shared<Tensor>(Array3D<float, 2, 2, 2>(
+                {
+                    {
+                        {
+                            {2.0, 3.0},
+                            {4.0, 5.0}
+                        },
+                        {
+                            {6.0, 7.0},
+                            {8.0, 9.0}
+                        }
+                    }
+                }
+            ));
+            const auto input1 = std::make_shared<Tensor>(Array3D<float, 2, 2, 2>(
+                {
+                    {
+                        {
+                            {1.0, 2.0},
+                            {3.0, 2.0}
+                        },
+                        {
+                            {2.0, 3.0},
+                            {1.0, 0.5}
+                        }
+                    }
+                }
+            ));
+            const auto gradOut = std::make_shared<Tensor>(Array3D<float, 2, 2, 2>(
+                {
+                    {
+                        {
+                            {0.5, 1.0},
+                            {1.5, 2.0}
+                        },
+                        {
+                            {2.5, 3.0},
+                            {3.5, 4.0}
+                        }
+                    }
+                }
+            ));
+            const auto expectedGrad0 = std::make_shared<Tensor>(Array3D<float, 2, 2, 2>(
+                {
+                    {
+                        {
+                            {0.50000000,   6.00000000},
+                            {72.00000000,  20.00000000}
+                        },
+                        {
+                            {30.00000000, 441.00000000},
+                            {3.50000000,   0.66666669}
+                        }
+                    }
+                }
+            ));
+            const auto expectedGrad1 = std::make_shared<Tensor>(Array3D<float, 2, 2, 2>(
+                {
+                    {
+                        {
+                            {  0.693147182, 9.88751030},
+                            {1.33084259e+02, 8.04718933e+01}
+                        },
+                        {
+                            {1.61258362e+02, 2.00234143e+03},
+                            {5.82243652e+01, 2.63666954e+01}
+                        }
+                    }
+                }
+            ));
+            for(const auto T: {input0, input1, gradOut, expectedGrad0, expectedGrad1})
+            {
+                    T->setBackend("cpu") ;
+                    T->setDataType(DataType::Float32);
+            }
+            std::shared_ptr<Node> powOp = Pow();
+            auto opr = std::static_pointer_cast<OperatorTensor>(powOp-> getOperator());
+            opr->setDataType(DataType::Float32);
+            opr->setBackend("cpu");
+            opr->associateInput(0, input0);
+            opr->associateInput(1, input1);
+            opr->getOutput(0)->setGrad(gradOut);
+            opr->forward();
+
+            powOp->backward();
+            REQUIRE(approxEq<float>(*(opr->getInput(0)->grad()), *expectedGrad0));
+            REQUIRE(approxEq<float>(*(opr->getInput(1)->grad()), *expectedGrad1));
+        }
+        SECTION("Broadcasting") {
+            const auto input0 = std::make_shared<Tensor>(Array3D<float, 2, 2, 3>(
+                {
+                    {
+                        {
+                            {1.0, 2.0, 3.0},
+                            {4.0, 5.0, 6.0}
+                        },
+                        {
+                            {1.5, 2.5, 3.5},
+                            {4.5, 5.5, 6.5}
+                        }
+                    }
+                }
+            ));
+            const auto input1 = std::make_shared<Tensor>(Array1D<float, 3>(
+                {
+                    {0.1, 0.2, 0.3}
+                }
+            ));
+
+            const auto gradOut = std::make_shared<Tensor>(Array3D<float, 2, 2, 3>(
+                {
+                    {
+                        {
+                            {1.0, 2.0, 3.0},
+                            {4.0, 5.0, 6.0}
+                        },
+                        {
+                            {6.0, 5.0, 4.0},
+                            {3.0, 2.0, 1.0}
+                        }
+                    }
+                }
+            ));
+            const auto expectedGrad0 = std::make_shared<Tensor>(Array3D<float, 2, 2, 3>(
+                {
+                    {
+                        {
+                            {0.10000000, 0.22973967, 0.41711676},
+                            {0.11486985, 0.27594593, 0.51353097}
+                        },
+                        {
+                            {0.41655189, 0.48044977, 0.49926791},
+                            {0.07748720, 0.10227509, 0.08092485}
+                        }
+                    }
+                }
+            ));
+            const auto expectedGrad1 = std::make_shared<Tensor>(Array1D<float, 3>(
+                {
+                    {14.14779854, 22.99299049, 33.56402588}
+                }
+            ));
+
+            for(const auto T: {input0, input1, gradOut, expectedGrad0, expectedGrad1})
+            {
+                    T->setBackend("cpu") ;
+                    T->setDataType(DataType::Float32);
+            }
+            std::shared_ptr<Node> powOp = Pow();
+            auto opr = std::static_pointer_cast<OperatorTensor>(powOp-> getOperator());
+            opr->setDataType(DataType::Float32);
+            opr->setBackend("cpu");
+            opr->associateInput(0, input0);
+            opr->associateInput(1, input1);
+            opr->getOutput(0)->setGrad(gradOut);
+            powOp->forward();
+
+            powOp->backward();
+            REQUIRE(approxEq<float>(*(opr->getInput(0)->grad()), *expectedGrad0));
+            REQUIRE(approxEq<float>(*(opr->getInput(1)->grad()), *expectedGrad1));
+        }
+    }
 }
 } // namespace Aidge