From ab55d7933bd7efbdddebbcee237323505d80244a Mon Sep 17 00:00:00 2001 From: dangqingqing Date: Tue, 5 Sep 2017 10:36:46 +0800 Subject: [PATCH] revert scatter_op and other mirror changes. --- doc/howto/dev/new_op_cn.md | 48 ++++++++++++++----- paddle/operators/mul_op.h | 2 +- paddle/operators/scatter_op.cc | 4 +- paddle/operators/scatter_op.h | 10 ++-- .../paddle/v2/framework/tests/test_mul_op.py | 3 ++ .../v2/framework/tests/test_rowwise_add_op.py | 2 +- 6 files changed, 46 insertions(+), 23 deletions(-) diff --git a/doc/howto/dev/new_op_cn.md b/doc/howto/dev/new_op_cn.md index ec79b7f42b..5c523bf046 100644 --- a/doc/howto/dev/new_op_cn.md +++ b/doc/howto/dev/new_op_cn.md @@ -280,28 +280,50 @@ class TestMulOp(unittest.TestCase): 反向Op单测继承自`GradientChecker`，而`GradientChecker`集成自`unittest.TestCase`，所以反向单测函数需要`test_`开头。 - ``` - class MulGradOpTest(GradientChecker): - def test_mul(self): - op = create_op("mul") - inputs = { +``` +class TestMulGradOp(GradientChecker): + def setUp(self): + self.op = create_op("mul") + self.inputs = { 'X': np.random.random((32, 84)).astype("float32"), 'Y': np.random.random((84, 100)).astype("float32") } - self.compare_grad(op, inputs) + + def test_cpu_gpu_compare(self): + self.compare_grad(self.op, self.inputs) + + def test_normal(self): # mul op will enlarge the relative error self.check_grad( - op, inputs, set(["X", "Y"]), "Out", max_relative_error=0.5) - ``` + self.op, self.inputs, ["X", "Y"], "Out", max_relative_error=0.5) + + def test_ignore_x(self): + self.check_grad( + self.op, + self.inputs, ["Y"], + "Out", + max_relative_error=0.5, + no_grad_set={"X"}) + + def test_ignore_y(self): + self.check_grad( + self.op, + self.inputs, ["X"], + "Out", + max_relative_error=0.5, + no_grad_set={"Y"}) +``` + +下面解释一些关键的地方: - 调用`create_op("mul")`创建反向Op对应的前向Op。 - - 定义输入`inputs`。 - 调用`compare_grad`函数对比CPU、GPU计算结果。 - - 调用`check_grad`检查梯度稳定性，这里采用数值法检测梯度正确性。 - - 第一个参数`op` : 前向op。 - - 第二个参数`inputs` : 输入词典，词典的Key和`ProtoMaker`定义保持一致。 - - 第三个参数`set(["X", "Y"])` : 指定对输入变量`X`、`Y`做梯度检测。 + - `test_normal`中调用`check_grad`检查梯度稳定性，这里采用数值法检测梯度正确性。 + - 第一个参数`self.op` : 前向Op。 + - 第二个参数`self.inputs` : 输入词典，词典的Key和`ProtoMaker`定义保持一致。 + - 第三个参数`["X", "Y"]` : 指定对输入变量`X`、`Y`做梯度检测。 - 第四个参数`"Out"` : 指定前向网络最终的输出目标变量`Out` + - `test_ignore_x`和`test_ignore_y`分支测试只需要计算一个输入梯度的情况。 ### 编译和执行 diff --git a/paddle/operators/mul_op.h b/paddle/operators/mul_op.h index 66ed2f81c7..05a79e13b3 100644 --- a/paddle/operators/mul_op.h +++ b/paddle/operators/mul_op.h @@ -54,8 +54,8 @@ class MulGradKernel : public framework::OpKernel { auto* device_context = const_cast(ctx.device_context_); if (dx) { - // dx = dout * y'. dx: M x K, dout : M x N, y : K x N dx->mutable_data(ctx.GetPlace()); + // dx = dout * y'. dx: M x K, dout : M x N, y : K x N math::matmul(*dout, false, *y, true, 1, dx, 0, device_context); } if (dy) { diff --git a/paddle/operators/scatter_op.cc b/paddle/operators/scatter_op.cc index 9b5068f07c..35c185ad80 100644 --- a/paddle/operators/scatter_op.cc +++ b/paddle/operators/scatter_op.cc @@ -50,8 +50,8 @@ class ScatterGradOp : public framework::OperatorWithKernel { auto *dRef = ctx.Output(framework::GradVarName("Ref")); auto *Ref = ctx.Input("Ref"); - if (dRef) dRef->Resize(Ref->dims()); - if (dUpdates) dUpdates->Resize(Updates->dims()); + dRef->Resize(Ref->dims()); + dUpdates->Resize(Updates->dims()); } }; diff --git a/paddle/operators/scatter_op.h b/paddle/operators/scatter_op.h index 7551480211..e9595638a8 100644 --- a/paddle/operators/scatter_op.h +++ b/paddle/operators/scatter_op.h @@ -49,12 +49,10 @@ class ScatterGradientOpKernel : public framework::OpKernel { auto *dOut = ctx.Input(framework::GradVarName("Out")); // In place gradient: dRef = dO - if (dRef) dRef->ShareDataWith(*dOut); - if (dUpdates) { - dUpdates->mutable_data(ctx.GetPlace()); - // Gradient by Gather: dUpdates += dO[Index] - Gather(ctx.GetPlace(), dOut, Index, dUpdates); - } + dRef->ShareDataWith(*dOut); + dUpdates->mutable_data(ctx.GetPlace()); + // Gradient by Gather: dUpdates += dO[Index] + Gather(ctx.GetPlace(), dOut, Index, dUpdates); } }; diff --git a/python/paddle/v2/framework/tests/test_mul_op.py b/python/paddle/v2/framework/tests/test_mul_op.py index 92d2b80e87..b58e4266d1 100644 --- a/python/paddle/v2/framework/tests/test_mul_op.py +++ b/python/paddle/v2/framework/tests/test_mul_op.py @@ -24,6 +24,9 @@ class TestMulGradOp(GradientChecker): 'Y': np.random.random((84, 100)).astype("float32") } + def test_cpu_gpu_compare(self): + self.compare_grad(self.op, self.inputs) + def test_normal(self): # mul op will enlarge the relative error self.check_grad( diff --git a/python/paddle/v2/framework/tests/test_rowwise_add_op.py b/python/paddle/v2/framework/tests/test_rowwise_add_op.py index 403734e71a..2ddb85e2e7 100644 --- a/python/paddle/v2/framework/tests/test_rowwise_add_op.py +++ b/python/paddle/v2/framework/tests/test_rowwise_add_op.py @@ -16,7 +16,7 @@ class TestRowwiseAddOp(unittest.TestCase): self.outputs = {'Out': np.add(self.inputs['X'], self.inputs['b'])} -class RowwiseAddGradOpTest(GradientChecker): +class TestRowwiseAddGradOp(GradientChecker): def setUp(self): self.op = create_op("rowwise_add") self.inputs = { -- GitLab