* fix pad javadoc and @see links. (#72) Signed-off-by: Robert Altena <Rob@Ra-ai.com> * [WIP] More fixes (#73) * special tests for ConstantTadHelper/ConstantShapeHelper Signed-off-by: raver119 <raver119@gmail.com> * release methods for data buffers Signed-off-by: raver119 <raver119@gmail.com> * delete temporary buffer Java side Signed-off-by: raver119 <raver119@gmail.com> * delete temporary buffer Java side Signed-off-by: raver119 <raver119@gmail.com> * delete temporary TadPack C++/Java side (#74) Signed-off-by: raver119 <raver119@gmail.com> * Zoo model TF import test updates (#75) * argLine fix, update compression_gru comment * updated comment for xception * undid but commented argLine change * updated xlnet comment * copyright headers * - new NDArray methods like()/ulike() (#77) - fix for depthwise_conv2d_bp + special test Signed-off-by: raver119 <raver119@gmail.com> * upsampling2d fix CUDA Signed-off-by: raver119 <raver119@gmail.com> * DL4J trace logging (#79) * MLN/CG trace logging for debugging Signed-off-by: AlexDBlack <blacka101@gmail.com> * Tiny tweak Signed-off-by: AlexDBlack <blacka101@gmail.com> * strided_slice_bp shape fn leak fix Signed-off-by: raver119 <raver119@gmail.com> * SameDiff fixes and naming (#78) * remove SDVariable inplace methods * import methods * npe fix in OpVal * removed SameDiff inplace ops from tests * Naming updates, moved to centralized methods in SameDiff, should use op_#:# for everything * quick fixes * javadoc * SDVariable eval with placeholders * use regex match * better matching * initial commit Signed-off-by: raver119 <raver119@gmail.com> * initial commit Signed-off-by: raver119 <raver119@gmail.com> * fix javadoc. (#76) * fix javadoc. Signed-off-by: Robert Altena <Rob@Ra-ai.com> * replace most @see with @link s. Signed-off-by: Robert Altena <Rob@Ra-ai.com> * 4 additional tests Signed-off-by: raver119 <raver119@gmail.com> * launch context reorganization Signed-off-by: raver119 <raver119@gmail.com> * LaunchContext reorganization Signed-off-by: raver119 <raver119@gmail.com> * per-device LaunchContext Signed-off-by: raver119 <raver119@gmail.com> * Various DL4J/ND4J fixes (#81) * #7954 Force refresh of UI when switching tabs on overview page Signed-off-by: AlexDBlack <blacka101@gmail.com> * #8017 Concurrent modification exception (synchronize) fix Signed-off-by: AlexDBlack <blacka101@gmail.com> * #8033 Don't initialize updater in middle of writing memory crash dump Signed-off-by: AlexDBlack <blacka101@gmail.com> * #8208 Fix shape checks for ND4J int[] creator methods Signed-off-by: AlexDBlack <blacka101@gmail.com> * #6385 #7992 Keras import naming fixes + cleanup Signed-off-by: AlexDBlack <blacka101@gmail.com> * #8016 Upsampling3D - add NDHWC format support Signed-off-by: AlexDBlack <blacka101@gmail.com> * ContextBuffers as separate entity Signed-off-by: raver119 <raver119@gmail.com> * Refactor NativeOps.h to export C functions * Actually export functions from NativeOps.h * Adapt the Java wrappers in ND4J generated with JavaCPP * Create C wrappers for some of the C++ classes currently used by ND4J * ContextBuffers as separate entity Signed-off-by: raver119 <raver119@gmail.com> * remove duplicate code in createBufferDetached. (#83) Signed-off-by: Robert Altena <Rob@Ra-ai.com> * Keras model import - updater lr fix (#84) * Keras model import - updater lr fix Signed-off-by: eraly <susan.eraly@gmail.com> * Keras model import - updater lr fix, cleanup Signed-off-by: eraly <susan.eraly@gmail.com> * ContextBuffers as separate entity Signed-off-by: raver119 <raver119@gmail.com> * ContextBuffers as separate entity Signed-off-by: raver119 <raver119@gmail.com> * Fix functions of OpaqueVariablesSet * thread-local buffers/affinity Signed-off-by: raver119 <raver119@gmail.com> * thread safety for LaunchContext Signed-off-by: raver119 <raver119@gmail.com> * more of thread safety Signed-off-by: raver119 <raver119@gmail.com> * one more multi threaded test Signed-off-by: raver119 <raver119@gmail.com> * SameDiff Convolution Config validation, better output methods (#82) * Conv Config validation & tests Signed-off-by: Ryan Nett <rnett@skymind.io> * stackOutputs utility method Signed-off-by: Ryan Nett <rnett@skymind.io> * use constructor for validation, support negative kernel sizes (infered from weights) Signed-off-by: Ryan Nett <rnett@skymind.io> * better output methods Signed-off-by: Ryan Nett <rnett@skymind.io> * move output to be with fit and evaluate Signed-off-by: Ryan Nett <rnett@skymind.io> * fixes Signed-off-by: Ryan Nett <rnett@skymind.io> * more fixes Signed-off-by: Ryan Nett <rnett@skymind.io> * refactor duplicate code from pad methods. (#86) * refactor duplicate code from pad methods. Signed-off-by: Robert Altena <Rob@Ra-ai.com> * replace switch with if. Signed-off-by: Robert Altena <Rob@Ra-ai.com> * Various ND4J/DL4J fixes and improvements (#87) * Reshape and reallocate - small fixes Signed-off-by: AlexDBlack <blacka101@gmail.com> * Reshape and reallocate - small fixes Signed-off-by: AlexDBlack <blacka101@gmail.com> * #6488 ElementWiseVertex broadcast support Signed-off-by: AlexDBlack <blacka101@gmail.com> * Constructors and broadcast supported it Transforms.max/min Signed-off-by: AlexDBlack <blacka101@gmail.com> * #8054 ElementWiseVertex now supports broadcast inputs Signed-off-by: AlexDBlack <blacka101@gmail.com> * #8057 Nd4j.create overload dtype fix Signed-off-by: AlexDBlack <blacka101@gmail.com> * #7551 ND4J Shape validation fix Signed-off-by: AlexDBlack <blacka101@gmail.com> * [WIP] Numpy boolean import (#91) * numpy bool type Signed-off-by: raver119 <raver119@gmail.com> * numpy bool java side Signed-off-by: raver119 <raver119@gmail.com> * remove create method with unused parameter. (#89) * remove create method with unused parameter. * removed more unused methods. Signed-off-by: Robert Altena <Rob@Ra-ai.com> * removing more unused code. Signed-off-by: Robert Altena <Rob@Ra-ai.com> * last removal of unused code. Signed-off-by: Robert Altena <Rob@Ra-ai.com> * remove createSparse methods. (#92) Signed-off-by: Robert Altena <Rob@Ra-ai.com> * Various ND4J/DL4J fixes (#90) * Deprecate Old*Op instances Signed-off-by: AlexDBlack <blacka101@gmail.com> * #8063 #8054 Broadcast exceptions + cleanup inplace ops Signed-off-by: AlexDBlack <blacka101@gmail.com> * Small fix Signed-off-by: AlexDBlack <blacka101@gmail.com> * Remove bad test condition Signed-off-by: AlexDBlack <blacka101@gmail.com> * #7993 Fix shape function issue in crop_and_resize op Signed-off-by: AlexDBlack <blacka101@gmail.com> * DL4J SameDiff lambda layer fix Signed-off-by: AlexDBlack <blacka101@gmail.com> * #8029 Fix for pnorm backprop math Signed-off-by: AlexDBlack <blacka101@gmail.com> * #8038 Fix Op profiler NaN/Inf triggering + add tests (#93) Signed-off-by: AlexDBlack <blacka101@gmail.com> * createUninitializedDetached refactoring. (#94) * wip * update interface, add null implementations. * Breaking one test in a weird way. Signed-off-by: Robert Altena <Rob@Ra-ai.com> * createUninitializedDetached refactored. Signed-off-by: Robert Altena <Rob@Ra-ai.com> * cuda build fix for issues introduced by recent refactoring Signed-off-by: raver119 <raver119@gmail.com> * [WIP] More of CUDA (#95) * initial commit Signed-off-by: raver119 <raver119@gmail.com> * Implementation of hashcode cuda helper. Working edition. * Fixed parallel test input arangements. * Fixed tests for hashcode op. * Fixed shape calculation for image:crop_and_resize op and test. * NativeOps tests. Initial test suite. * Added tests for indexReduce methods. * Added test on execBroadcast with NDArray as dimensions. * Added test on execBroadcastBool with NDArray as dimensions. * Added tests on execPairwiseTransform and execPairwiseTransofrmBool. * Added tests for execReduce with scalar results. * Added reduce tests for non-empty dims array. * Added tests for reduce3. * Added tests for execScalar. * Added tests for execSummaryStats. * - provide cpu/cuda code for batch_to_space - testing it Signed-off-by: Yurii <yurii@skymind.io> * - remove old test for batch_to_space (had wrong format and numbers were not checked) Signed-off-by: Yurii <yurii@skymind.io> * Fixed complilation errors with test. * Added test for execTransformFloat. * Added test for execTransformSame. * Added test for execTransformBool. * Added test for execTransformStrict. * Added tests for execScalar/execScalarBool with TADs. * Added test for flatten. * - provide cpu/cuda code for space_to_Batch operaion Signed-off-by: Yurii <yurii@skymind.io> * Added test for concat. * comment unnecessary stuff in s_t_b Signed-off-by: Yurii <yurii@skymind.io> * Added test for specialConcat. * Added tests for memcpy/set routines. * Fixed pullRow cuda test. * Added pullRow test. * Added average test. * - correct typo in NDArray::applyPairwiseTransform(nd4j::pairwise::BoolOps op...) Signed-off-by: Yurii <yurii@skymind.io> * - debugging and fixing cuda tests in JavaInteropTests file Signed-off-by: Yurii <yurii@skymind.io> * - correct some tests Signed-off-by: Yurii <yurii@skymind.io> * Added test for shuffle. * Fixed ops declarations. * Restored omp and added shuffle test. * Added convertTypes test. * Added tests for execRandom. Eliminated usage of RandomBuffer with NativeOps. * Added sort tests. * Added tests for execCustomOp. * - further debuging and fixing tests terminated with crash Signed-off-by: Yurii <yurii@skymind.io> * Added tests for calculateOutputShapes. * Addded Benchmarks test. * Commented benchmark tests. * change assertion Signed-off-by: raver119 <raver119@gmail.com> * Added tests for apply_sgd op. Added cpu helper for that op. * Implement cuda helper for aplly_sgd op. Fixed tests for NativeOps. * Added test for assign broadcastable. * Added tests for assign_bp op. * Added tests for axpy op. * - assign/execScalar/execTransformAny signature change - minor test fix Signed-off-by: raver119 <raver119@gmail.com> * Fixed axpy op. * meh Signed-off-by: raver119 <raver119@gmail.com> * - fix tests for nativeOps::concat Signed-off-by: Yurii <yurii@skymind.io> * sequential transform/scalar Signed-off-by: raver119 <raver119@gmail.com> * allow nested parallelism Signed-off-by: raver119 <raver119@gmail.com> * assign_bp leak fix Signed-off-by: raver119 <raver119@gmail.com> * block setRNG fix Signed-off-by: raver119 <raver119@gmail.com> * enable parallelism by default Signed-off-by: raver119 <raver119@gmail.com> * enable nested parallelism by default Signed-off-by: raver119 <raver119@gmail.com> * Added cuda implementation for row_count helper. * Added implementation for tnse gains op helper. * - take into account possible situations when input arrays are empty in reduce_ cuda stuff Signed-off-by: Yurii <yurii@skymind.io> * Implemented tsne/edge_forces op cuda-based helper. Parallelized cpu-based helper for edge_forces. * Added kernel for tsne/symmetrized op heleper. * Implementation of tsne/symmetrized op cuda helper. Working edition. * Eliminated waste printfs. * Added test for broadcastgradientargs op. * host-only fallback for empty reduce float Signed-off-by: raver119 <raver119@gmail.com> * - some tests fixes Signed-off-by: Yurii <yurii@skymind.io> * - correct the rest of reduce_ stuff Signed-off-by: Yurii <yurii@skymind.io> * - further correction of reduce_ stuff Signed-off-by: Yurii <yurii@skymind.io> * Added test for Cbow op. Also added cuda implementation for cbow helpers. * - improve code of stack operation for scalar case Signed-off-by: Yurii <yurii@skymind.io> * - provide cuda kernel for gatherND operation Signed-off-by: Yurii <yurii@skymind.io> * Implementation of cbow helpers with cuda kernels. * minor tests tweaks Signed-off-by: raver119 <raver119@gmail.com> * minor tests tweaks Signed-off-by: raver119 <raver119@gmail.com> * - further correction of cuda stuff Signed-off-by: Yurii <yurii@skymind.io> * Implementatation of cbow op helper with cuda kernels. Working edition. * Skip random testing for cudablas case. * lstmBlockCell context fix Signed-off-by: raver119 <raver119@gmail.com> * Added tests for ELU and ELU_BP ops. * Added tests for eq_scalar, gt_scalar, gte_scalar and lte_scalar ops. * Added tests for neq_scalar. * Added test for noop. * - further work on clipbynorm_bp Signed-off-by: Yurii <yurii@skymind.io> * - get rid of concat op call, use instead direct concat helper call Signed-off-by: Yurii <yurii@skymind.io> * lstmBlockCell context fix Signed-off-by: raver119 <raver119@gmail.com> * Added tests for lrelu and lrelu_bp. * Added tests for selu and selu_bp. * Fixed lrelu derivative helpers. * - some corrections in lstm Signed-off-by: Yurii <yurii@skymind.io> * operator * result shape fix Signed-off-by: raver119 <raver119@gmail.com> * - correct typo in lstmCell Signed-off-by: Yurii <yurii@skymind.io> * few tests fixed Signed-off-by: raver119 <raver119@gmail.com> * CUDA inverse broadcast bool fix Signed-off-by: raver119 <raver119@gmail.com> * disable MMAP test for CUDA Signed-off-by: raver119 <raver119@gmail.com> * BooleanOp syncToDevice Signed-off-by: raver119 <raver119@gmail.com> * meh Signed-off-by: raver119 <raver119@gmail.com> * additional data types for im2col/col2im Signed-off-by: raver119 <raver119@gmail.com> * Added test for firas_sparse op. * one more RandomBuffer test excluded Signed-off-by: raver119 <raver119@gmail.com> * Added tests for flatten op. * Added test for Floor op. * bunch of tests fixed Signed-off-by: raver119 <raver119@gmail.com> * mmulDot tests fixed Signed-off-by: raver119 <raver119@gmail.com> * more tests fixed Signed-off-by: raver119 <raver119@gmail.com> * Implemented floordiv_bp op and tests. * Fixed scalar case with cuda implementation for bds. * - work on cuda kernel for clip_by_norm backprop op is completed Signed-off-by: Yurii <yurii@skymind.io> * Eliminate cbow crach. * more tests fixed Signed-off-by: raver119 <raver119@gmail.com> * more tests fixed Signed-off-by: raver119 <raver119@gmail.com> * Eliminated abortion with batched nlp test. * more tests fixed Signed-off-by: raver119 <raver119@gmail.com> * Fixed shared flag initializing. * disabled bunch of cpu workspaces tests Signed-off-by: raver119 <raver119@gmail.com> * scalar operators fix: missing registerSpecialUse call Signed-off-by: raver119 <raver119@gmail.com> * Fixed logdet for cuda and tests. * - correct clipBynorm_bp Signed-off-by: Yurii <yurii@skymind.io> * Fixed crop_and_resize shape datatype. * - correct some mmul tests Signed-off-by: Yurii <yurii@skymind.io> * build fix Signed-off-by: raver119 <raver119@gmail.com> * exclude two methods for JNI Signed-off-by: raver119 <raver119@gmail.com> * exclude two methods for JNI Signed-off-by: raver119 <raver119@gmail.com> * exclude two methods for JNI (#97) Signed-off-by: raver119 <raver119@gmail.com> * temporary stack fix Signed-off-by: raver119 <raver119@gmail.com> * round robin affinity test Signed-off-by: raver119 <raver119@gmail.com> * get rid of legacy CudaContext methods Signed-off-by: raver119 <raver119@gmail.com> * get rid of legacy ContextPool classes/methods Signed-off-by: raver119 <raver119@gmail.com> * one legacy test removed Signed-off-by: raver119 <raver119@gmail.com> * few more fields rearranged Signed-off-by: raver119 <raver119@gmail.com> * OpaqueLaunchContext Signed-off-by: raver119 <raver119@gmail.com> * OpaqueLaunchContext++ Signed-off-by: raver119 <raver119@gmail.com> * more of OpaqueLaunchContext methods Signed-off-by: raver119 <raver119@gmail.com> * LaunchContext -> CudaContext Signed-off-by: raver119 <raver119@gmail.com> * AffinityManger changes Signed-off-by: raver119 <raver119@gmail.com> * AffinityManger changes Signed-off-by: raver119 <raver119@gmail.com> * cusolver handles Signed-off-by: raver119 <raver119@gmail.com> * typo Signed-off-by: raver119 <raver119@gmail.com> * cusolver method Signed-off-by: raver119 <raver119@gmail.com> * cusolver handle propagated Signed-off-by: raver119 <raver119@gmail.com> * blas/solver handles Signed-off-by: raver119 <raver119@gmail.com> * one more test Signed-off-by: raver119 <raver119@gmail.com> * legacy concat implementations replaced with new CustomOp Signed-off-by: raver119 <raver119@gmail.com> * one more test Signed-off-by: raver119 <raver119@gmail.com> * concat now uses way more blocks Signed-off-by: raver119 <raver119@gmail.com> * print Signed-off-by: raver119 <raver119@gmail.com> * no more triple template mmul Signed-off-by: raver119 <raver119@gmail.com> * bunch of kernels have dtypes reconsidered Signed-off-by: raver119 <raver119@gmail.com> * bunch of kernels have dtypes reconsidered Signed-off-by: raver119 <raver119@gmail.com> * bitonic sort reorganized Signed-off-by: raver119 <raver119@gmail.com> * bunch of cpu stuff removed from cuda scope Signed-off-by: raver119 <raver119@gmail.com> * bunch of cpu stuff removed from cuda scope Signed-off-by: raver119 <raver119@gmail.com> * type conversions moved to generic impl Signed-off-by: raver119 <raver119@gmail.com> * cpu data types pass Signed-off-by: raver119 <raver119@gmail.com> * non_max_suppression Signed-off-by: raver119 <raver119@gmail.com> * sortByValue fix Signed-off-by: raver119 <raver119@gmail.com> * ignore all mixed datatype tests for mmul Signed-off-by: raver119 <raver119@gmail.com> * special handling of OpProfiler exceptions Signed-off-by: raver119 <raver119@gmail.com> * - one failing concat test in cpp - Nd4j.tile now uses op internally Signed-off-by: raver119 <raver119@gmail.com> * get back dtype exception for legacy arrays deserialization Signed-off-by: raver119 <raver119@gmail.com>
508 lines
17 KiB
C++
508 lines
17 KiB
C++
/*******************************************************************************
|
|
* Copyright (c) 2015-2018 Skymind, Inc.
|
|
*
|
|
* This program and the accompanying materials are made available under the
|
|
* terms of the Apache License, Version 2.0 which is available at
|
|
* https://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.
|
|
*
|
|
* SPDX-License-Identifier: Apache-2.0
|
|
******************************************************************************/
|
|
|
|
|
|
//
|
|
// Created by raver on 8/4/2018.
|
|
//
|
|
|
|
#include "testlayers.h"
|
|
#include <ops/declarable/CustomOperations.h>
|
|
#include <NDArray.h>
|
|
#include <ops/ops.h>
|
|
#include <GradCheck.h>
|
|
#include <array>
|
|
|
|
|
|
using namespace nd4j;
|
|
|
|
|
|
class DeclarableOpsTests15 : public testing::Test {
|
|
public:
|
|
|
|
DeclarableOpsTests15() {
|
|
printf("\n");
|
|
fflush(stdout);
|
|
}
|
|
};
|
|
|
|
TEST_F(DeclarableOpsTests15, Test_NormalizeMoments_1) {
|
|
auto d = NDArrayFactory::create<double>('c', {10, 10});
|
|
auto w = NDArrayFactory::create<double>(10);
|
|
auto x = NDArrayFactory::create<double>('c', {10});
|
|
auto y = NDArrayFactory::create<double>('c', {10});
|
|
|
|
auto z0 = NDArrayFactory::create<double>('c', {10});
|
|
auto z1 = NDArrayFactory::create<double>('c', {10});
|
|
|
|
nd4j::ops::normalize_moments op;
|
|
auto result = op.execute({&w, &x, &y}, {&z0, &z1}, {1e-4}, {}, {});
|
|
ASSERT_EQ(Status::OK(), result);
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, Test_Add_1) {
|
|
auto x = NDArrayFactory::create<int>('c', {5}, {1, 1, 1, 1, 1});
|
|
auto y = NDArrayFactory::create<int>('c', {5}, {1, 1, 1, 1, 1});
|
|
auto e = NDArrayFactory::create<int>('c', {5}, {2, 2, 2, 2, 2});
|
|
|
|
nd4j::ops::add op;
|
|
auto result = op.execute({&x, &y}, {&x}, {}, {}, {});
|
|
ASSERT_EQ(Status::OK(), result);
|
|
ASSERT_EQ(e, x);
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, Test_Half_assign_1) {
|
|
auto x = NDArrayFactory::create<float16>('c', {2, 5});
|
|
int y = 1;
|
|
x.assign(y);
|
|
|
|
ASSERT_EQ(10, x.sumNumber().e<int>(0));
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, test_avgpooling_edge_1) {
|
|
int inOutH = 5;// 35;
|
|
int inOutW = 5;// 35;
|
|
int inOutC = 10;// 192;
|
|
|
|
auto x = NDArrayFactory::create<double>('c', {1, inOutH, inOutW, inOutC});
|
|
x.linspace(1.0);
|
|
|
|
nd4j::ops::avgpool2d op;
|
|
auto result = op.execute({&x}, {}, {3,3, 1,1, 0,0, 1,1, 1, 0, 1});
|
|
ASSERT_EQ(Status::OK(), result->status());
|
|
|
|
auto z = result->at(0);
|
|
|
|
int totalPadHeight = (inOutH - 1) * 1 + 3 - inOutH;
|
|
int padTop = totalPadHeight / 2;
|
|
int padBottom = totalPadHeight - totalPadHeight / 2;
|
|
|
|
int k = 3;
|
|
|
|
auto m = NDArrayFactory::create<double>('c', {1, inOutH, inOutW, inOutC});
|
|
auto c = NDArrayFactory::create<double>('c', {1, inOutH, inOutW, inOutC});
|
|
|
|
for (int h = 0; h < inOutH; h++) {
|
|
for (int w = 0; w < inOutW; w++) {
|
|
int hFrom = h - padTop;
|
|
int wFrom = w - padBottom;
|
|
|
|
int hTo = hFrom + k;
|
|
int wTo = wFrom + k;
|
|
|
|
hFrom = nd4j::math::nd4j_max<int>(0, hFrom);
|
|
wFrom = nd4j::math::nd4j_max<int>(0, wFrom);
|
|
|
|
hTo = nd4j::math::nd4j_min<int>(inOutH, hTo);
|
|
wTo = nd4j::math::nd4j_min<int>(inOutW, wTo);
|
|
|
|
int idxOut[4];
|
|
int idxIn[4];
|
|
for (int ch = 0; ch < inOutC; ch++) {
|
|
idxOut[1] = h;
|
|
idxOut[2] = w;
|
|
idxOut[3] = ch;
|
|
idxIn[3] = ch;
|
|
|
|
for (int kh = hFrom; kh < hTo; kh++) {
|
|
for (int kw = wFrom; kw < wTo; kw++) {
|
|
idxIn[1] = kh;
|
|
idxIn[2] = kw;
|
|
|
|
auto inVal = x.e<double>(0, kh, kw, ch);
|
|
m.p(0, h, w, ch, inVal + m.e<double>(0, h, w, ch));
|
|
c.p(0, h, w, ch, 1 + c.e<int>(0, h, w, ch));
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
m /= c;
|
|
|
|
ASSERT_EQ(m, *z);
|
|
|
|
delete result;
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, Test_standarize_1) {
|
|
auto x = NDArrayFactory::create<float>('c', {5}, {1, 1, 1, 1, 1});
|
|
auto e = NDArrayFactory::create<float>('c', {5}, {0, 0, 0, 0, 0});
|
|
|
|
nd4j::ops::standardize op;
|
|
auto result = op.execute({&x}, {&x}, {}, {0}, {});
|
|
ASSERT_EQ(Status::OK(), result);
|
|
ASSERT_EQ(e, x);
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, Test_standarize_bp_1) {
|
|
auto x = NDArrayFactory::create<float>('c', {5}, {1., 1., 1., 1., 1.});
|
|
auto eps = NDArrayFactory::create<float>('c', {5}, {0., 0., 0., 0., 0.});
|
|
|
|
nd4j::ops::standardize_bp op;
|
|
auto result = op.execute({&x, &eps}, {}, {0}, {});
|
|
ASSERT_EQ(Status::OK(), result->status());
|
|
delete result;
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, Test_depthwise_bp_1) {
|
|
auto in = NDArrayFactory::create<float>('c', {4, 8, 64, 64});
|
|
auto w = NDArrayFactory::create<float>('c', {2, 2, 8, 2});
|
|
auto b = NDArrayFactory::create<float>('c', {1, 16});
|
|
auto grad = NDArrayFactory::create<float>('c', {4, 16, 64, 64});
|
|
|
|
auto gradI = in.like();
|
|
auto gradW = w.like();
|
|
auto gradB = b.like();
|
|
|
|
nd4j:ops::depthwise_conv2d_bp op;
|
|
auto status = op.execute({&in, &w, &b, &grad}, {&gradI, &gradW, &gradB}, {}, {2, 2, 1, 1, 0, 0, 1, 1, 1, 0}, {});
|
|
ASSERT_EQ(Status::OK(), status);
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, test_matmul_bp_1) {
|
|
auto a = NDArrayFactory::create<double>('c', {1, 3});
|
|
auto b = NDArrayFactory::create<double>('c', {1, 4});
|
|
auto gI = NDArrayFactory::create<double>('c', {3, 4});
|
|
|
|
auto gA = NDArrayFactory::create<double>('c', {1, 3});
|
|
auto gB = NDArrayFactory::create<double>('c', {1, 4});
|
|
|
|
nd4j::ops::matmul_bp op;
|
|
auto status = op.execute({&a, &b, &gI}, {&gA, &gB}, {}, {1, 0, 0}, {});
|
|
ASSERT_EQ(Status::OK(), status);
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, test_non_decreasing_1) {
|
|
auto x = NDArrayFactory::create<double>(1.0);
|
|
auto z = NDArrayFactory::create<bool>(false);
|
|
auto e = NDArrayFactory::create<bool>(true);
|
|
|
|
nd4j::ops::is_non_decreasing op;
|
|
Context ctx(1);
|
|
ctx.setInputArray(0, &x);
|
|
ctx.setOutputArray(0, &z);
|
|
|
|
auto status = op.execute(&ctx);
|
|
ASSERT_EQ(Status::OK(), status);
|
|
ASSERT_EQ(e, z);
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, test_check_numeric_1) {
|
|
auto x = NDArrayFactory::create<float>('c', {3},{1.f, 2.f, 3.f});
|
|
auto y = NDArrayFactory::string("shouldn't ever trigger");
|
|
|
|
nd4j::ops::check_numerics op;
|
|
auto result = op.execute({&x, &y}, {}, {});
|
|
ASSERT_EQ(Status::OK(), result->status());
|
|
|
|
auto z = result->at(0);
|
|
|
|
ASSERT_EQ(x, *z);
|
|
|
|
delete result;
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, test_check_numeric_2) {
|
|
auto x = NDArrayFactory::create<float>('c', {3},{1.f, 2.f, std::numeric_limits<float>::infinity()});
|
|
auto y = NDArrayFactory::string("should trigger");
|
|
auto z = NDArrayFactory::create<float>('c', {3} );
|
|
|
|
nd4j::ops::check_numerics op;
|
|
try {
|
|
auto status = op.execute({&x, &y}, {&z}, {}, {}, {});
|
|
ASSERT_TRUE(false);
|
|
} catch (std::invalid_argument &e) {
|
|
//
|
|
}
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, test_check_numeric_3) {
|
|
auto x = NDArrayFactory::create<float>('c', {3},{1.f, 2.f, std::numeric_limits<float>::quiet_NaN()});
|
|
auto y = NDArrayFactory::string("should trigger");
|
|
auto z = NDArrayFactory::create<float>('c', {3} );
|
|
|
|
nd4j::ops::check_numerics op;
|
|
try {
|
|
auto status = op.execute({&x, &y}, {&z}, {}, {}, {});
|
|
ASSERT_TRUE(false);
|
|
} catch (std::invalid_argument &e) {
|
|
//
|
|
}
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, Test_layer_norm_1) {
|
|
auto x = NDArrayFactory::create<float>('c', {1, 5}, {1., 2., 3., 4., 5.});
|
|
auto g = NDArrayFactory::create<float>('c', {1, 5}, {1., 2., 3., 4., 5.});
|
|
auto b = NDArrayFactory::create<float>('c', {1, 5}, {1., 2., 3., 4., 5.});
|
|
|
|
nd4j::ops::layer_norm op;
|
|
auto result = op.execute({&x, &g, &b}, {}, {0}, {});
|
|
ASSERT_EQ(Status::OK(), result->status());
|
|
delete result;
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, Test_layer_norm_bp_1) {
|
|
auto x = NDArrayFactory::create<float>('c', {1, 5}, {1., 2., 3., 4., 5.});
|
|
auto g = NDArrayFactory::create<float>('c', {1, 5}, {1., 2., 3., 4., 5.});
|
|
auto b = NDArrayFactory::create<float>('c', {1, 5}, {1., 2., 3., 4., 5.});
|
|
auto eps = NDArrayFactory::create<float>('c', {1, 5}, {0., 0., 0., 0., 0.});
|
|
|
|
nd4j::ops::layer_norm_bp op;
|
|
auto result = op.execute({&x, &g, &b, &eps}, {}, {0}, {});
|
|
ASSERT_EQ(Status::OK(), result->status());
|
|
delete result;
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, test_hashCode_1) {
|
|
auto x = NDArrayFactory::create<int>('c', {10});
|
|
auto y = NDArrayFactory::create<int>('c', {10});
|
|
|
|
x.linspace(1.);
|
|
y.linspace(2.);
|
|
|
|
nd4j::ops::hashcode op;
|
|
auto resultA0 = op.execute({&x}, {}, {}, {}, false, nd4j::DataType::INT64);
|
|
auto resultA1 = op.execute({&x}, {}, {}, {}, false, nd4j::DataType::INT64);
|
|
auto resultB0 = op.execute({&y}, {}, {}, {}, false, nd4j::DataType::INT64);
|
|
// resultA0->at(0)->printIndexedBuffer("A0");
|
|
// resultA1->at(0)->printIndexedBuffer("A1");
|
|
// resultB0->at(0)->printIndexedBuffer("B0");
|
|
ASSERT_EQ(*resultA0->at(0), *resultA1->at(0));
|
|
ASSERT_NE(*resultA0->at(0), *resultB0->at(0));
|
|
|
|
delete resultA0;
|
|
delete resultA1;
|
|
delete resultB0;
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, test_hashCode_2) {
|
|
auto x = NDArrayFactory::create<int>('c', {1027});
|
|
auto y = NDArrayFactory::create<int>('c', {1027});
|
|
|
|
x.linspace(1.);
|
|
y.linspace(2.);
|
|
|
|
nd4j::ops::hashcode op;
|
|
auto resultA0 = op.execute({&x}, {}, {}, {}, false, nd4j::DataType::INT64);
|
|
auto resultA1 = op.execute({&x}, {}, {}, {}, false, nd4j::DataType::INT64);
|
|
auto resultB0 = op.execute({&y}, {}, {}, {}, false, nd4j::DataType::INT64);
|
|
|
|
// resultA0->at(0)->printIndexedBuffer("A0");
|
|
// resultA1->at(0)->printIndexedBuffer("A1");
|
|
// resultB0->at(0)->printIndexedBuffer("B0");
|
|
|
|
ASSERT_EQ(*resultA0->at(0), *resultA1->at(0));
|
|
ASSERT_NE(*resultA0->at(0), *resultB0->at(0));
|
|
|
|
delete resultA0;
|
|
delete resultA1;
|
|
delete resultB0;
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, test_reshape_to_scalar_1) {
|
|
auto array = NDArrayFactory::create<float>(119.f);
|
|
auto e = NDArrayFactory::create<float>('c', {1, 1}, {119.f});
|
|
|
|
nd4j::ops::reshape op;
|
|
auto result = op.execute({&array}, {}, {1, 1});
|
|
ASSERT_EQ(Status::OK(), result->status());
|
|
|
|
auto z = result->at(0);
|
|
|
|
ASSERT_EQ(e, *z);
|
|
|
|
delete result;
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, test_reshape_to_scalar_2) {
|
|
auto array = NDArrayFactory::create<float>(119.f);
|
|
auto e = NDArrayFactory::create<float>('c', {1, 1}, {119.f});
|
|
auto z = NDArrayFactory::create<float>('c', {1, 1});
|
|
|
|
nd4j::ops::reshape op;
|
|
auto result = op.execute({&array}, {&z}, {}, {1, 1}, {});
|
|
ASSERT_EQ(Status::OK(), result);
|
|
ASSERT_EQ(e, z);
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, test_rank_1) {
|
|
auto array = NDArrayFactory::create<float>('c', {4, 64});
|
|
auto e = NDArrayFactory::create<int>('c', {}, {2});
|
|
auto z = NDArrayFactory::create<int>('c', {});
|
|
|
|
nd4j::ops::rank op;
|
|
auto result = op.execute({&array}, {&z}, {}, {}, {});
|
|
ASSERT_EQ(Status::OK(), result);
|
|
ASSERT_EQ(e, z);
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, test_rank_2) {
|
|
auto array = NDArrayFactory::create<float>('c', {4, 64});
|
|
auto e = NDArrayFactory::create<int>('c', {}, {2});
|
|
|
|
nd4j::ops::rank op;
|
|
auto result = op.execute({&array}, {}, {});
|
|
ASSERT_EQ(Status::OK(), result->status());
|
|
|
|
auto z = result->at(0);
|
|
|
|
ASSERT_EQ(e, *z);
|
|
|
|
delete result;
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, test_concat_column_1) {
|
|
auto x = NDArrayFactory::create<double>('c', {2, 1}, {1, 1});
|
|
auto y = NDArrayFactory::create<double>('c', {2, 1}, {0, 0});
|
|
auto e = NDArrayFactory::create<double>('c', {2, 2}, {1, 0, 1, 0});
|
|
auto z = NDArrayFactory::create<double>('c', {2, 2});
|
|
|
|
nd4j::ops::concat op;
|
|
auto status = op.execute({&x, &y}, {&z}, {}, {1}, {});
|
|
ASSERT_EQ(Status::OK(), status);
|
|
|
|
z.printIndexedBuffer("z");
|
|
|
|
ASSERT_EQ(e, z);
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, test_concat_large_1) {
|
|
std::array<NDArray*, 2000> arrays;
|
|
Context context(1);
|
|
Nd4jLong axis = 0;
|
|
|
|
// we crate bunch of arrays, filled with specific values
|
|
for (int e = 0; e < arrays.size(); e++) {
|
|
auto array = NDArrayFactory::create_<float>('c', {1, 300});
|
|
array->assign(e);
|
|
context.setInputArray(e, array, true);
|
|
}
|
|
|
|
auto z = NDArrayFactory::create<float>('c', {2000, 300});
|
|
context.setOutputArray(0, &z, false);
|
|
context.setIArguments(&axis, 1);
|
|
|
|
nd4j::ops::concat op;
|
|
op.execute(&context);
|
|
|
|
for (int e = 0; e < arrays.size(); e++) {
|
|
auto row = z.tensorAlongDimension(e, {1});
|
|
|
|
ASSERT_NEAR((float) e, row->e<float>(0), 1e-5f);
|
|
|
|
delete row;
|
|
}
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, test_concat_large_2) {
|
|
std::array<NDArray*, 10> arrays;
|
|
Context context(1);
|
|
Nd4jLong axis = 0;
|
|
|
|
// we crate bunch of arrays, filled with specific values
|
|
for (int e = 0; e < arrays.size(); e++) {
|
|
auto array = NDArrayFactory::create_<float>('c', {1, 5, 20});
|
|
array->assign(e);
|
|
context.setInputArray(e, array, true);
|
|
}
|
|
|
|
auto z = NDArrayFactory::create<float>('c', {arrays.size(), 5, 20});
|
|
context.setOutputArray(0, &z, false);
|
|
context.setIArguments(&axis, 1);
|
|
|
|
nd4j::ops::concat op;
|
|
op.execute(&context);
|
|
|
|
for (int e = 0; e < arrays.size(); e++) {
|
|
auto row = z.tensorAlongDimension(e, {1, 2});
|
|
|
|
ASSERT_NEAR((float) e, row->meanNumber().e<float>(0), 1e-5f);
|
|
|
|
delete row;
|
|
}
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, test_lstmBlock_1) {
|
|
auto x0 = NDArrayFactory::create<Nd4jLong>(5);
|
|
auto x1 = NDArrayFactory::create<float>('c', {5, 1, 4}, {0.7787856f, 0.80119777f, 0.72437465f, 0.23089433f, 0.72714126f, 0.18039072f, 0.50563407f, 0.89252293f, 0.5461209f, 0.92336726f, 0.085571885f, 0.7937801f, 0.65908563f, 0.55552566f, 0.15962744f, 0.30874777f, 0.15476847f, 0.46954823f, 0.9938899f, 0.6112741f});
|
|
auto x2 = NDArrayFactory::create<float>('c', {1, 3}, {0.7717289f, 0.9280778f, 0.98455656f});
|
|
auto x3 = NDArrayFactory::create<float>('c', {1, 3}, {0.94414854f, 0.5956861f, 0.8668989f});
|
|
auto x4 = NDArrayFactory::create<float>('c', {7, 12}, {0.460692f, 0.042572856f, 0.08420354f, -0.09538093f, -0.11416581f, -0.53166187f, 0.40133476f, -0.24381405f, 0.30778718f, 0.52713746f, 0.16253126f, -0.034891903f, 0.011679292f, -0.19076681f, 0.14710993f, -0.3704369f, 0.51872355f, 0.13536876f, -0.5568739f, -0.08727971f, 0.07601875f, -0.074174374f, -0.5345982f, -0.3581748f, -0.28263924f, -0.25141674f, 0.43328637f, -0.50227314f, -0.26641843f, -0.38241976f, -0.19636461f, -0.04020852f, -0.27312332f, 0.5207915f, -0.37247592f, -0.4713087f, -0.25670746f, -0.14942765f, -0.015806139f, -0.22531253f, 0.5582536f, 0.3093416f, 0.3221351f, -0.0964683f, 0.14318448f, 0.42279094f, -0.46992f, -0.43399644f, -0.51704615f, -0.11854091f, 0.21697259f, -0.049382925f, 0.14059627f, 0.3912331f, -0.41345632f, 0.5067368f, -0.3420229f, 0.485789f, 0.044918716f, 0.26209074f, 0.12357575f, 0.21778125f, -0.53791714f, 0.18346387f, 0.054183125f, 0.5480431f, 0.03675288f, -0.26656917f, -0.018610716f, 0.19917983f, 0.5566165f, 0.43570566f, -0.35720813f, 0.31097364f, -0.47134516f, -0.289197f, 0.091138184f, 0.13300979f, -0.36592877f, -0.17540845f, 0.21732038f, 0.4393713f, 0.42800313f, 0.5006979f});
|
|
auto x5 = NDArrayFactory::create<float>('c', {1, 3});
|
|
auto x6 = NDArrayFactory::create<float>('c', {1, 3});
|
|
auto x7 = NDArrayFactory::create<float>('c', {1, 3});
|
|
auto x8 = NDArrayFactory::create<float>('c', {12});
|
|
|
|
nd4j::ops::lstmBlock op;
|
|
auto result = op.execute({&x0, &x1, &x2, &x3, &x4, &x5, &x6, &x7, &x8}, {2.0, 0.3}, {0, 0});
|
|
ASSERT_EQ(Status::OK(), result->status());
|
|
|
|
auto z = result->at(0);
|
|
|
|
// z->printIndexedBuffer("Z");
|
|
|
|
delete result;
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, test_lstmBlock_2) {
|
|
int seqLen = 32;
|
|
int bS = 64;
|
|
int nIn = 32;
|
|
|
|
auto x0 = NDArrayFactory::create<Nd4jLong>(5);
|
|
auto x1 = NDArrayFactory::create<float>('f', {bS, nIn, seqLen});
|
|
auto x2 = NDArrayFactory::create<float>('f', {bS, nIn}); // nIn == nOut
|
|
auto x3 = NDArrayFactory::create<float>('f', {bS, nIn});
|
|
auto x4 = NDArrayFactory::create<float>('f', {2 * nIn, 4 * nIn});
|
|
auto x5 = NDArrayFactory::create<float>('f', {nIn});
|
|
auto x6 = NDArrayFactory::create<float>('f', {nIn});
|
|
auto x7 = NDArrayFactory::create<float>('f', {nIn});
|
|
auto x8 = NDArrayFactory::create<float>('f', {4 * nIn});
|
|
|
|
nd4j::ops::lstmBlock op;
|
|
auto result = op.execute({&x0, &x1, &x2, &x3, &x4, &x5, &x6, &x7, &x8}, {1.0, 0.0}, {0, 1});
|
|
ASSERT_EQ(Status::OK(), result->status());
|
|
|
|
auto z = result->at(0);
|
|
|
|
delete result;
|
|
}
|
|
|
|
TEST_F(DeclarableOpsTests15, test_lstmBlock_3) {
|
|
|
|
int seqLen = 3;
|
|
int bS = 2;
|
|
int nIn = 4;
|
|
|
|
NDArray f('f', {bS, nIn, seqLen}, nd4j::DataType::FLOAT32);
|
|
NDArray cLast('f', {bS, nIn}, nd4j::DataType::FLOAT32);
|
|
|
|
f = 2;
|
|
cLast = 3;
|
|
|
|
for (int t = 0; t < seqLen; ++t) {
|
|
|
|
//section 1
|
|
//auto ft = f({0,0, 0,0, t,t+1});
|
|
//auto temp = ft * cLast;
|
|
|
|
|
|
// section 2
|
|
auto ft = f({0,0, 0,0, t,t+1});
|
|
auto temp1 = ft.reshape('f', {bS, nIn});
|
|
auto temp2 = temp1 * cLast;
|
|
}
|
|
} |