Skip to content

Commit

Permalink
[NeoMLTest] Add DnnCrossEntropyLossLayerTest
Browse files Browse the repository at this point in the history 
Signed-off-by: Kirill Golikov <[email protected]>
  • Loading branch information
@favorart
favorart committed Sep 20, 2024
1 parent 60df378 commit 444e4f3
Show file tree
Hide file tree
Showing 2 changed files with 244 additions and 0 deletions.
1 change: 1 addition & 0 deletions NeoML/test/src/CMakeLists.txt
View file
Original file line number Diff line number Diff line change
Expand Up @@ -12,6 +12,7 @@ target_sources(${PROJECT_NAME} INTERFACE
${CMAKE_CURRENT_SOURCE_DIR}/DnnBlobTest.cpp
${CMAKE_CURRENT_SOURCE_DIR}/DnnChannelwiseWith1x1BlockTest.cpp
${CMAKE_CURRENT_SOURCE_DIR}/DnnConfusionMatrixLayer.cpp
${CMAKE_CURRENT_SOURCE_DIR}/DnnCrossEntropyLossLayerTest.cpp
${CMAKE_CURRENT_SOURCE_DIR}/DnnCtcTest.cpp
${CMAKE_CURRENT_SOURCE_DIR}/DnnDistributedTest.cpp
${CMAKE_CURRENT_SOURCE_DIR}/DnnDropoutTest.cpp
Expand Down
243 changes: 243 additions & 0 deletions NeoML/test/src/DnnCrossEntropyLossLayerTest.cpp
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,243 @@
/* Copyright © 2024 ABBYY
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.
--------------------------------------------------------------------------------------------------------------*/

#include <common.h>
#pragma hdrstop

#include <TestFixture.h>
#include <DnnSimpleTest.h>

using namespace NeoML;
using namespace NeoMLTest;

namespace NeoMLTest {

struct CCrossEntropyLossLayerTestParams final {
CCrossEntropyLossLayerTestParams( float target, float result, float lossValue ) :
Target( target ), Result( result ), LossValue( lossValue ) {}

float Target;
float Result;
float LossValue;
};

class CCrossEntropyLossLayerTest :
public CNeoMLTestFixture, public ::testing::WithParamInterface<CCrossEntropyLossLayerTestParams> {
public:
static bool InitTestFixture() { return true; }
static void DeinitTestFixture() {}
};

} // namespace NeoMLTest

//---------------------------------------------------------------------------------------------------------------------

TEST_F( CCrossEntropyLossLayerTest, ZeroBackwardDiffTest )
{
const auto met = MathEngine().GetType();
if( met != MET_Cpu && met != MET_Cuda ) {
NEOML_HILIGHT( GTEST_LOG_( INFO ) ) << "Skipped rest of test for MathEngine type=" << met << " because no implementation.\n";
return;
}

const float rawData[8]{
0.7f, 0.3f,
0.3f, 0.7f,
0.2f, 0.8f,
0.4f, 0.6f
};
const float rawExpectedDiff[8]{
-0.10032f, 0.10032f,
0.10032f, -0.10032f,
0.08859f, -0.08859f,
0.0f, 0.0f
};
const float rawFloatLabels[8]{
1.f, 0.f,
0.f, 1.f,
0.f, 1.f,
0.f, 0.f
};
const int rawIntLabels[4]{ 0, 1, 1, -1 };

CRandom random;
CDnn dnn( random, MathEngine() );

CPtr<CSourceLayer> data = Source( dnn, "data" );
CPtr<CSourceLayer> label = Source( dnn, "label" );

CPtr<CDnnSimpleTestDummyLearningLayer> learn = new CDnnSimpleTestDummyLearningLayer( MathEngine() );
learn->SetName( "learn" );
learn->Connect( *data );
dnn.AddLayer( *learn );

CPtr<CCrossEntropyLossLayer> loss = CrossEntropyLoss()( learn.Ptr(), label.Ptr() );

CPtr<CDnnBlob> dataBlob = CDnnBlob::CreateDataBlob( MathEngine(), CT_Float, 2, 2, 2 );
dataBlob->CopyFrom( rawData );
data->SetBlob( dataBlob );

CPtr<CDnnBlob> expectedDiff = CDnnBlob::CreateDataBlob( MathEngine(), CT_Float, 2, 2, 2 );
expectedDiff->CopyFrom( rawExpectedDiff );
learn->ExpectedDiff = expectedDiff;

CPtr<CDnnBlob> labelBlob = CDnnBlob::CreateDataBlob( MathEngine(), CT_Float, 2, 2, 2 );
labelBlob->CopyFrom( rawFloatLabels );
label->SetBlob( labelBlob );

dnn.RunAndBackwardOnce();

labelBlob = CDnnBlob::CreateDataBlob( MathEngine(), CT_Int, 2, 2, 1 );
labelBlob->CopyFrom( rawIntLabels );
label->SetBlob( labelBlob );

dnn.RunAndBackwardOnce();
}

TEST_F( CCrossEntropyLossLayerTest, NoSoftmax )
{
const auto met = MathEngine().GetType();
if( met != MET_Cpu && met != MET_Cuda ) {
NEOML_HILIGHT( GTEST_LOG_( INFO ) ) << "Skipped rest of test for MathEngine type=" << met << " because no implementation.\n";
return;
}

const float resultBuff[]{ 1.f / 15, 2.f / 15, 3.f / 15, 4.f / 15, 5.f / 15,
5.f / 15, 4.f / 15, 3.f / 15, 2.f / 15, 1.f / 15,
0.2f, 0.2f, 0.2f, 0.2f, 0.2f,
0.2f, 0.2f, 0.2f, 0.2f, 0.2f,
0.2f, 0.2f, 0.2f, 0.2f, 0.2f,
0.2f, 0.2f, 0.2f, 0.2f, 0.2f,
0.2f, 0.2f, 0.2f, 0.2f, 0.2f
};

CRandom random;
CDnn dnn( random, MathEngine() );

CPtr<CSourceLayer> result = Source( dnn, "result" );
CPtr<CSourceLayer> target = Source( dnn, "target" );

CPtr<CDnnSimpleTestDummyLearningLayer> learn = new CDnnSimpleTestDummyLearningLayer( MathEngine() );
learn->Connect( *result );
dnn.AddLayer( *learn );

CPtr<CCrossEntropyLossLayer> loss = CrossEntropyLoss( /*softmax*/false )( learn.Ptr(), target.Ptr() );

CPtr<CDnnBlob> resultBlob = CDnnBlob::CreateDataBlob( MathEngine(), CT_Float, 1, 7, 5 );
resultBlob->CopyFrom( resultBuff );
result->SetBlob( resultBlob );

CPtr<CDnnBlob> floatTargetBlob = CDnnBlob::CreateDataBlob( MathEngine(), CT_Float, 1, 7, 5 );
floatTargetBlob->Fill( 0.f );

floatTargetBlob->GetData().SetValueAt( 2, 1 );
floatTargetBlob->GetObjectData( 1 ).SetValueAt( 4, 1 );
for( int i = 2; i < 7; ++i ) {
floatTargetBlob->GetObjectData( i ).SetValueAt( 3, 1 );
}
target->SetBlob( floatTargetBlob );

CPtr<CDnnBlob> expectedDiffBLob = resultBlob->GetCopy();
expectedDiffBLob->Fill( 1.f / 7 );

expectedDiffBLob->GetData().SetValueAt( 2, -4.f / 7 );
expectedDiffBLob->GetObjectData( 1 ).SetValueAt( 4, -2 );
for( int i = 2; i < 7; ++i ) {
expectedDiffBLob->GetObjectData( i ).SetValueAt( 3, -4.f / 7 );
}
learn->ExpectedDiff = expectedDiffBLob;

dnn.RunAndBackwardOnce();

{
dnn.DeleteLayer( *learn );
loss->Connect( 0, *result );
dnn.RunAndBackwardOnce();

CMemoryFile file;
{
CArchive archive( &file, CArchive::store );
archive.Serialize( dnn );
}
file.SeekToBegin();
{
CArchive archive( &file, CArchive::load );
archive.Serialize( dnn );
}

result = CheckCast<CSourceLayer>( dnn.GetLayer( result->GetName() ) );
target = CheckCast<CSourceLayer>( dnn.GetLayer( target->GetName() ) );
loss = CheckCast<CCrossEntropyLossLayer>( dnn.GetLayer( loss->GetName() ) );

result->SetBlob( resultBlob );

learn->Connect( *result );
loss->Connect( *learn );
dnn.AddLayer( *learn );
}

CPtr<CDnnBlob> intTargetBlob = CDnnBlob::CreateDataBlob( MathEngine(), CT_Int, 1, 7, 1 );
intTargetBlob->FillObject<int>( 0, 2 );
intTargetBlob->FillObject<int>( 1, 4 );
for( int i = 2; i < 7; ++i ) {
intTargetBlob->GetObjectData<int>( i ).SetValue( 3 );
}
target->SetBlob( intTargetBlob );

dnn.RunAndBackwardOnce();
}

TEST_P( CCrossEntropyLossLayerTest, BinaryCrossEntropyLossSignTest )
{
const CCrossEntropyLossLayerTestParams params = GetParam();

CRandom random;
CDnn dnn( random, MathEngine() );

CTextStream debugOutput;
dnn.SetLog( &debugOutput );

CSourceLayer* result = Source( dnn, "result" );
CSourceLayer* target = Source( dnn, "target" );

CPtr<CBinaryCrossEntropyLossLayer> loss = BinaryCrossEntropyLoss()( result, target );

CPtr<CDnnBlob> resultBlob = CDnnBlob::CreateDataBlob( MathEngine(), CT_Float, 1, 1, 1 );
resultBlob->Fill( params.Result );
result->SetBlob( resultBlob );

CPtr<CDnnBlob> targetBlob = CDnnBlob::CreateDataBlob( MathEngine(), CT_Float, 1, 1, 1 );
targetBlob->Fill( params.Target );
target->SetBlob( targetBlob );

dnn.RunAndBackwardOnce();

EXPECT_TRUE( FloatEq( loss->GetLastLoss(), params.LossValue ) );
}

INSTANTIATE_TEST_CASE_P( CnnCrossEntropyLossTestInstantiation, CCrossEntropyLossLayerTest,
::testing::Values(
CCrossEntropyLossLayerTestParams( -1.f, -2.f, 0.126928f ),
CCrossEntropyLossLayerTestParams( -1.f, -0.999f, 0.313531f ),
CCrossEntropyLossLayerTestParams( -1.f, -1.f, 0.313262f ),
CCrossEntropyLossLayerTestParams( -1.f, 1.f, 1.313262f ),
CCrossEntropyLossLayerTestParams( 1.f, 1.f, 0.313262f ),
CCrossEntropyLossLayerTestParams( 1.f, -1.f, 1.313262f ),
CCrossEntropyLossLayerTestParams( 1.f, 3.1415926f, 4.2306e-2f ),
CCrossEntropyLossLayerTestParams( -1.f, -0.451f, 0.49286f ),
CCrossEntropyLossLayerTestParams( 1.f, 10000.f, 0.f ),
CCrossEntropyLossLayerTestParams( -1.f, -10000.f, 0.f )
)
);

0 comments on commit 444e4f3

Please sign in to comment.