test_LayerGrad.cpp 46.2 KB
Newer Older
Z
zhangjinchao01 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
/* Copyright (c) 2016 Baidu, Inc. All Rights Reserve.

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 <gtest/gtest.h>
#include <string>
Q
qijun 已提交
17
#include <vector>
Z
zhangjinchao01 已提交
18
#include "ModelConfig.pb.h"
Q
qijun 已提交
19
#include "paddle/gserver/layers/DataLayer.h"
Z
zhangjinchao01 已提交
20
#include "paddle/trainer/Trainer.h"
21
#include "paddle/math/MathUtils.h"
Z
zhangjinchao01 已提交
22 23

#include "LayerGradUtil.h"
Q
qijun 已提交
24
#include "TestUtil.h"
Z
zhangjinchao01 已提交
25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53

using namespace paddle;  // NOLINT
using namespace std;     // NOLINT

P_DECLARE_bool(use_gpu);
P_DECLARE_int32(gpu_id);
P_DECLARE_double(checkgrad_eps);
P_DECLARE_bool(thread_local_rand_use_global_seed);
P_DECLARE_bool(prev_batch_state);

TEST(Operator, dot_mul) {
  TestConfig config;
  config.layerConfig.set_size(10);

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 10, 0});
  config.inputDefs.push_back({INPUT_DATA, "layer_1", 10, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  OperatorConfig& operatorConf = *config.layerConfig.add_operator_confs();
  operatorConf.set_type("dot_mul");
  operatorConf.set_dotmul_scale(-1);

  testOperatorGrad(config, operatorConf, 100, false, false);
}

TEST(Projection, context) {
  for (auto contextStart : {-5, -3, -1, 0, 3}) {
    for (auto contextLength : {1, 2, 5, 7}) {
54
      for (auto batchSize : {1, 2, 5, 20, 50}) {
Z
zhangjinchao01 已提交
55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137
        for (auto trainablePadding : {false, true}) {
          LOG(INFO) << " contextStart=" << contextStart
                    << " contextLength=" << contextLength
                    << " batchSize=" << batchSize
                    << " trainablePadding=" << trainablePadding;
          ProjectionConfig conf;
          conf.set_type("context");
          conf.set_input_size(10);
          conf.set_context_start(contextStart);
          conf.set_context_length(contextLength);
          conf.set_trainable_padding(trainablePadding);
          conf.set_output_size(conf.context_length() * conf.input_size());
          int pad =
              std::max(0, -conf.context_start()) +
              std::max(0, conf.context_start() + conf.context_length() - 1);
          for (auto useGpu : {false, true}) {
            testProjectionGrad(
                conf, INPUT_SEQUENCE_DATA,
                trainablePadding ? conf.input_size() * pad : 0, batchSize,
                useGpu,
                contextStart + contextLength <= 1);  // = testState
          }
        }
      }
    }
  }
}

TEST(Projection, trans_fc) {
  ProjectionConfig conf;
  conf.set_type("trans_fc");
  conf.set_input_size(50);
  conf.set_output_size(20);
  for (auto useGpu : {false, true}) {
    testProjectionGrad(conf, INPUT_DATA, /* parameterSize */ 1000,
                       /* batchSize */ 100, useGpu);
  }
}

TEST(Projection, fc) {
  ProjectionConfig conf;
  conf.set_type("fc");
  conf.set_input_size(10);
  conf.set_output_size(20);
  for (auto useGpu : {false, true}) {
    testProjectionGrad(conf, INPUT_DATA, /* parameterSize */ 200,
                       /* batchSize */ 100, useGpu);
  }
}

TEST(Projection, dot_mul) {
  ProjectionConfig conf;
  conf.set_type("dot_mul");
  conf.set_input_size(20);
  conf.set_output_size(20);
  for (auto useGpu : {false, true}) {
    testProjectionGrad(conf, INPUT_DATA, /* parameterSize */ 20,
                       /* batchSize */ 100, useGpu);
  }
}

TEST(Projection, table) {
  ProjectionConfig conf;
  conf.set_type("table");
  conf.set_input_size(10);
  conf.set_output_size(20);
  for (auto useGpu : {false, true}) {
    testProjectionGrad(conf, INPUT_LABEL, /* parameterSize */ 200,
                       /* batchSize */ 100, useGpu);
  }
}

TEST(Projection, identity) {
  ProjectionConfig conf;
  conf.set_type("identity");
  conf.set_input_size(10);
  conf.set_output_size(10);
  for (auto useGpu : {false, true}) {
    testProjectionGrad(conf, INPUT_DATA, /* parameterSize */ 0,
                       /* batchSize */ 100, useGpu);
  }
}

X
xuwei06 已提交
138 139 140 141 142 143 144 145 146 147 148
TEST(Projection, scaling) {
  ProjectionConfig conf;
  conf.set_type("scaling");
  conf.set_input_size(10);
  conf.set_output_size(10);
  for (auto useGpu : {false}) {
    testProjectionGrad(conf, INPUT_DATA, /* parameterSize */ 1,
                       /* batchSize */ 100, useGpu);
  }
}

149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171
#ifndef PADDLE_ONLY_CPU
TEST(Projection, conv) {
  const int NUM_FILTERS = 16;
  const int FILTER_SIZE = 2;
  const int FILTER_SIZE_Y = 3;
  const int CHANNELS = 3;
  const int IMAGE_SIZE = 16;

  ProjectionConfig conf;
  conf.set_type("conv");
  conf.set_num_filters(NUM_FILTERS);

  ConvConfig* conv = conf.mutable_conv_conf();
  conv->set_filter_size(FILTER_SIZE);
  conv->set_filter_size_y(FILTER_SIZE_Y);
  conv->set_channels(CHANNELS);
  conv->set_padding(0);
  conv->set_padding_y(1);
  conv->set_stride(2);
  conv->set_stride_y(2);
  conv->set_groups(1);
  conv->set_filter_channels(conv->channels() / conv->groups());
  conv->set_img_size(IMAGE_SIZE);
172 173 174 175 176 177 178
  int output_x =
      outputSize(conv->img_size(), conv->filter_size(), conv->padding(),
                 conv->stride(), /* caffeMode */ true);
  int output_y =
      outputSize(conv->img_size(), conv->filter_size_y(), conv->padding_y(),
                 conv->stride_y(), /* caffeMode */ true);
  conv->set_output_x(output_x);
179
  conf.set_input_size(IMAGE_SIZE * IMAGE_SIZE * CHANNELS);
180
  conf.set_output_size(output_x * output_y * NUM_FILTERS);
181

182 183
  testProjectionGrad(
      conf, INPUT_DATA,
184 185 186 187 188
      /* parameterSize */ NUM_FILTERS * CHANNELS * FILTER_SIZE * FILTER_SIZE_Y,
      /* batchSize */ 100, true, false, NUM_FILTERS, true);
}
#endif

L
Update  
liaogang 已提交
189 190 191 192 193 194
TEST(Layer, BilinearInterpLayer) {
  TestConfig config;
  config.layerConfig.set_type("bilinear_interp");
  config.biasSize = 0;
  config.inputDefs.push_back({INPUT_DATA, "layer_0", 4096, 0});

L
liaogang 已提交
195 196 197 198 199 200
  LayerInputConfig* input = config.layerConfig.add_inputs();
  BilinearInterpConfig* bilinear = input->mutable_bilinear_interp_conf();
  bilinear->set_img_size_x(32);
  bilinear->set_img_size_y(32);
  bilinear->set_num_channels(4);

L
liaogang 已提交
201 202 203 204 205 206 207
  for (auto useGpu : {false, true}) {
    for (auto outSize : {32, 64}) {
      bilinear->set_out_size_x(outSize);
      bilinear->set_out_size_y(outSize);
      testLayerGrad(config, "bilinear_interp", 10, false, useGpu);
    }
  }
L
Update  
liaogang 已提交
208 209
}

Z
zhangjinchao01 已提交
210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254
TEST(Layer, concat) {
  TestConfig config;
  config.biasSize = 0;
  config.layerConfig.set_type("concat");
  config.layerConfig.set_size(15);
  config.layerConfig.set_active_type("sigmoid");

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 5, 0});
  config.layerConfig.add_inputs();
  config.inputDefs.push_back({INPUT_DATA, "layer_1", 10, 0});
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "concat", 100, false, useGpu);
  }
}

TEST(Layer, AddtoLayer) {
  TestConfig config;
  config.biasSize = 0;
  config.layerConfig.set_type("addto");
  config.layerConfig.set_size(10);
  config.layerConfig.set_active_type("sigmoid");

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 10, 0});
  config.layerConfig.add_inputs();
  config.inputDefs.push_back({INPUT_DATA, "layer_1", 10, 0});
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "addto", 100, false, useGpu);
  }
}

TEST(Layer, CRFLayer) {
  TestConfig config;
  config.layerConfig.set_type("crf");
  config.layerConfig.set_size(10);
  config.biasSize = 0;

  config.inputDefs.push_back({INPUT_SEQUENCE_DATA, "layer_0", 10, 120});
  config.inputDefs.push_back({INPUT_SEQUENCE_LABEL, "layer_1", 10, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

255 256 257
  // Not support GPU now
  testLayerGrad(config, "crf", 100, /* trans */ false, /* useGpu */ false,
                false /*useWeight*/, 0.03 /*epsilon*/);
Z
zhangjinchao01 已提交
258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329
}

TEST(Layer, CTCLayer) {
  TestConfig config;
  config.layerConfig.set_type("ctc");
  config.layerConfig.set_norm_by_times(false);
  config.layerConfig.set_size(10);
  config.biasSize = 0;

  config.inputDefs.push_back({INPUT_SEQUENCE_DATA, "layer_0", 10, 0});
  config.inputDefs.push_back({INPUT_SEQUENCE_LABEL, "layer_1", 10, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "ctc", 100, /* trans */ false, /* useGpu */ useGpu);
  }
}

TEST(Layer, cosSimLayer) {
  TestConfig config;
  config.layerConfig.set_type("cos");
  config.layerConfig.set_size(1);
  config.biasSize = 0;

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 50, 0});
  config.inputDefs.push_back({INPUT_DATA, "layer_1", 50, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "cos", 100, false, useGpu);
  }
}

TEST(Layer, CosSimVecMatLayer) {
  TestConfig config;
  config.layerConfig.set_type("cos_vm");
  config.layerConfig.set_size(5);  // output size
  config.layerConfig.set_cos_scale(2.0);

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 20, 0});
  config.layerConfig.add_inputs();
  config.inputDefs.push_back({INPUT_DATA, "layer_1", 100, 0});
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "cos_vm", 100, false, useGpu);
  }
}

void testConvLayer(const string& type, bool trans, bool useGpu) {
  TestConfig config;
  config.biasSize = 16;
  config.layerConfig.set_type(type);
  config.layerConfig.set_num_filters(16);
  config.layerConfig.set_partial_sum(1);
  config.layerConfig.set_shared_biases(true);

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 768, 288});
  LayerInputConfig* input = config.layerConfig.add_inputs();
  ConvConfig* conv = input->mutable_conv_conf();
  conv->set_filter_size(2);
  conv->set_filter_size_y(3);
  conv->set_channels(3);
  conv->set_padding(0);
  conv->set_padding_y(1);
  conv->set_stride(2);
  conv->set_stride_y(2);
  conv->set_groups(1);
  conv->set_filter_channels(conv->channels() / conv->groups());
  conv->set_img_size(16);
330 331 332
  conv->set_output_x(outputSize(conv->img_size(), conv->filter_size(),
                                conv->padding(), conv->stride(),
                                /* caffeMode */ true));
Z
zhangjinchao01 已提交
333 334 335 336
  config.layerConfig.set_size(conv->output_x() * conv->output_x() *
                              config.layerConfig.num_filters());

  testLayerGrad(config, "conv", 100, trans, useGpu);
337 338
  // Use small batch_size and useWeight=true to test biasGrad
  testLayerGrad(config, "conv", 2, trans, useGpu, true, 0.02);
Z
zhangjinchao01 已提交
339 340 341 342 343 344 345 346 347 348
}

TEST(Layer, convLayer) {
  testConvLayer("exconv", /* trans= */ false, /* useGpu= */ false);
#ifndef PADDLE_ONLY_CPU
  testConvLayer("exconv", /* trans= */ false, /* useGpu= */ true);
  testConvLayer("cudnn_conv", /* trans= */ false, /* useGpu= */ true);
#endif
}

W
wangyang59 已提交
349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370

void testConvTransLayer(const string& type, bool trans, bool useGpu) {
  TestConfig config;
  config.biasSize = 3;
  config.layerConfig.set_type(type);
  config.layerConfig.set_num_filters(3);
  config.layerConfig.set_partial_sum(1);
  config.layerConfig.set_shared_biases(true);

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 1024, 288});
  LayerInputConfig* input = config.layerConfig.add_inputs();
  ConvConfig* conv = input->mutable_conv_conf();
  conv->set_filter_size(2);
  conv->set_filter_size_y(3);
  conv->set_channels(16);
  conv->set_padding(0);
  conv->set_padding_y(1);
  conv->set_stride(2);
  conv->set_stride_y(2);
  conv->set_groups(1);
  conv->set_filter_channels(3 / conv->groups());
  conv->set_img_size(16);
371 372 373
  conv->set_output_x(outputSize(conv->img_size(), conv->filter_size(),
                                conv->padding(), conv->stride(),
                                /* caffeMode */ true));
W
wangyang59 已提交
374 375 376 377 378

  config.layerConfig.set_size(conv->img_size() * conv->img_size() *
                              config.layerConfig.num_filters());

  testLayerGrad(config, "convTrans", 100, trans, useGpu);
379 380
  // Use small batch_size and useWeight=true to test biasGrad
  testLayerGrad(config, "convTrans", 2, trans, useGpu, true, 0.02);
W
wangyang59 已提交
381 382 383
}

TEST(Layer, convTransLayer) {
384 385 386
  for (auto useGpu : {false, true}) {
    testConvTransLayer("exconvt", /* trans= */ false, /* useGpu= */ useGpu);
  }
W
wangyang59 已提交
387 388
}

Z
zhangjinchao01 已提交
389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406
TEST(Layer, blockExpandLayer) {
  TestConfig config;
  config.biasSize = 0;
  config.layerConfig.set_type("blockexpand");

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 6144, 0});
  LayerInputConfig* input = config.layerConfig.add_inputs();
  BlockExpandConfig* blockExpand = input->mutable_block_expand_conf();
  blockExpand->set_img_size_x(64);
  blockExpand->set_img_size_y(32);
  blockExpand->set_channels(3);
  blockExpand->set_padding_x(0);
  blockExpand->set_padding_y(0);
  blockExpand->set_block_x(4);
  blockExpand->set_block_y(32);
  blockExpand->set_stride_x(2);
  blockExpand->set_stride_y(2);
  blockExpand->set_output_x(
407 408 409
      outputSize(blockExpand->img_size_x(), blockExpand->block_x(),
                 blockExpand->padding_x(), blockExpand->stride_x(),
                 /* caffeMode */ false));
Z
zhangjinchao01 已提交
410
  blockExpand->set_output_y(
411 412 413
      outputSize(blockExpand->img_size_y(), blockExpand->block_y(),
                 blockExpand->padding_y(), blockExpand->stride_y(),
                 /* caffeMode */ false));
Z
zhangjinchao01 已提交
414 415 416 417 418 419 420 421
  config.layerConfig.set_size(blockExpand->block_x() * blockExpand->block_y() *
                              blockExpand->channels());

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "blockexpand", 100, false, useGpu);
  }
}

422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439
TEST(Layer, maxoutLayer) {
  TestConfig config;
  config.biasSize = 0;
  config.layerConfig.set_type("maxout");

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 4096, 0});
  LayerInputConfig* input = config.layerConfig.add_inputs();
  MaxOutConfig* maxout = input->mutable_maxout_conf();

  maxout->set_img_size_x(32);
  maxout->set_img_size_y(32);
  maxout->set_channels(4);
  maxout->set_groups(2);

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "maxout", 10, false, useGpu);
  }
}
Z
zhangjinchao01 已提交
440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541
void testFcLayer(string format, size_t nnz) {
  TestConfig config;
  config.biasSize = 4096;
  config.layerConfig.set_type("fc");
  config.layerConfig.set_size(4096);
  config.layerConfig.set_active_type("sigmoid");
  config.layerConfig.set_drop_rate(0.1);

  config.inputDefs.push_back(
      {INPUT_DATA, "layer_0", 8192, nnz, ParaSparse(format)});
  config.layerConfig.add_inputs();

  LOG(INFO) << config.inputDefs[0].sparse.sparse << " "
            << config.inputDefs[0].sparse.format;

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "fc", 100, /* trans */ false, useGpu,
                  /* weight */ true);
  }
}

TEST(Layer, fcLayer) {
  testFcLayer("", 4096 * 4096 * 2);
  testFcLayer("csc", 4096 * 40);
  testFcLayer("csr", 4096 * 40);
}

TEST(Layer, SelectiveFullyConnectedLayer) {
  TestConfig config;
  size_t nin = 16;
  size_t nout = 256;
  config.layerConfig.set_type("selective_fc");
  config.layerConfig.set_size(nout);
  config.layerConfig.set_active_type("sigmoid");
  config.layerConfig.set_has_selected_colums(true);
  config.layerConfig.set_selective_fc_pass_generation(false);
  config.biasSize = nout;

  config.inputDefs.push_back({INPUT_DATA, "input0", nin, nin * nout});
  config.layerConfig.add_inputs();
  config.inputDefs.push_back(
      {INPUT_SPARSE_NON_VALUE_DATA, "index", nout, 0, ParaSparse("csr", true)});
  config.layerConfig.add_inputs();

  testLayerGrad(config, "selective_fc", 100,
                /* trans= */ false, /* useGup= */ false, false);
#ifndef PADDLE_ONLY_CPU
  testLayerGrad(config, "selective_fc", 100,
                /* trans= */ false, /* useGup= */ true, false);
#endif
}

TEST(Layer, DataNormLayer) {
  TestConfig config;
  config.layerConfig.set_type("data_norm");
  config.layerConfig.set_size(20);
  config.biasSize = 0;

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 20, 100});
  config.inputDefs.back().isStatic = true;
  config.layerConfig.add_inputs();

  for (auto strategy : {"z-score", "min-max", "decimal-scaling"}) {
    config.layerConfig.set_data_norm_strategy(strategy);
    // The parameters are static, so not support GPU now
    testLayerGrad(config, "data_norm", 200, /* trans */ false,
                  /* useGpu */ false);
  }
}

TEST(Layer, hsigmoidLayer) {
  TestConfig config;
  config.layerConfig.set_type("hsigmoid");
  config.layerConfig.set_num_classes(5);
  config.layerConfig.set_size(1);
  config.biasSize = config.layerConfig.num_classes() - 1;

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 50, 200});
  config.inputDefs.push_back({INPUT_LABEL, "layer_1", 5, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  // Not support GPU now
  testLayerGrad(config, "hsigmoid", 100, /* trans */ false, /* useGpu */ false);
}

TEST(Layer, multi_cross) {
  TestConfig config;
  config.layerConfig.set_type("multi-class-cross-entropy");
  config.biasSize = 0;

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 50, 0});
  config.inputDefs.push_back({INPUT_LABEL, "layer_1", 10, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "multi-class-cross-entropy", 100, /* trans */ false,
                  useGpu);
  }
}

H
Haonan 已提交
542
TEST(Layer, multi_binary_label_sparse_mat) {
Z
zhangjinchao01 已提交
543 544 545 546 547 548 549 550 551
  TestConfig config;
  config.layerConfig.set_type("multi_binary_label_cross_entropy");
  config.biasSize = 0;

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 50, 0});
  config.inputDefs.push_back({INPUT_SPARSE_NON_VALUE_DATA, "layer_1", 50, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

552 553 554 555
  for (auto useGpu : {false, true}) {
      testLayerGrad(config, "multi_binary_label_cross_entropy", 100,
                    /* trans */ false, useGpu);
  }
Z
zhangjinchao01 已提交
556 557
}

H
Haonan 已提交
558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573
TEST(layer, multi_binary_label_id) {
  TestConfig config;
  config.layerConfig.set_type("multi_binary_label_cross_entropy");
  config.biasSize = 0;

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 50, 0});
  config.inputDefs.push_back({INPUT_LABEL, "layer_1", 10, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
      testLayerGrad(config, "multi_binary_label_cross_entropy", 100,
                    /* trans */ false, useGpu);
  }
}

Z
zhangjinchao01 已提交
574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941
TEST(Layer, multi_cross_with_selfnorm) {
  TestConfig config;
  config.layerConfig.set_type("multi_class_cross_entropy_with_selfnorm");
  config.layerConfig.set_softmax_selfnorm_alpha(0.1);
  config.biasSize = 0;

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 50, 0});
  config.inputDefs.push_back({INPUT_LABEL, "layer_1", 10, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  // Not support GPU now
  testLayerGrad(config, "multi_class_cross_entropy_with_selfnorm", 100,
                /* trans */ false,
                /* useGpu */ false);
}

TEST(Layer, multi_cross_soft) {
  TestConfig config;
  config.layerConfig.set_type("soft_binary_class_cross_entropy");
  config.biasSize = 0;

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 10, 0});
  config.inputDefs.push_back({INPUT_DATA_TARGET, "layer_1", 10, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "soft_binary_class_cross_entropy", 100,
                  /* trans */ false, useGpu);
  }
}

TEST(Layer, square_error) {
  TestConfig config;
  config.layerConfig.set_type("square_error");
  config.biasSize = 0;

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 10, 0});
  config.inputDefs.push_back({INPUT_DATA_TARGET, "layer_1", 10, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "square_error", 100, /* trans */ false, useGpu);
  }
}

TEST(Layer, sparse_square_error) {
  TestConfig config;
  config.layerConfig.set_type("square_error");
  config.biasSize = 0;

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 50, 0});
  config.inputDefs.push_back({INPUT_SPARSE_NON_VALUE_DATA, "layer_1", 50, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  // "GpuSparseMatrix" as label is not supported
  testLayerGrad(config, "square_error", 100, /* trans */ false,
                /* useGpu */ false);
}

TEST(Layer, sparse_float_square_error) {
  TestConfig config;
  config.layerConfig.set_type("square_error");
  config.biasSize = 0;

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 50, 0});
  config.inputDefs.push_back({INPUT_SPARSE_FLOAT_VALUE_DATA, "layer_1", 50, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  // "GpuSparseMatrix" as label is not supported
  testLayerGrad(config, "square_error", 100, /* trans */ false,
                /* useGpu */ false);
}

TEST(Layer, square_error_weighted) {
  TestConfig config;
  config.layerConfig.set_type("square_error");
  config.biasSize = 0;
  config.testAccumulate = false;

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 10, 0});
  config.inputDefs.push_back({INPUT_DATA_TARGET, "layer_1", 10, 0});
  config.inputDefs.push_back({INPUT_DATA_TARGET, "layer_2", 1, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "square_error", 100, /* trans */ false, useGpu);
  }
}

TEST(Layer, huber_two_class) {
  TestConfig config;
  config.layerConfig.set_type("huber");
  config.biasSize = 0;

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 1, 0});
  config.inputDefs.push_back({INPUT_LABEL, "layer_1", 2, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "huber", 100, /* trans */ false, useGpu);
  }
}

void testExpandLayer(string trans_type, bool hasSubseq) {
  TestConfig config;
  config.layerConfig.set_type("expand");

  config.inputDefs.push_back(
      {trans_type == "non-seq" ? INPUT_DENSE_DIM_DATA : INPUT_SEQUENCE_DATA,
       "layer_0", 10, 0});
  config.inputDefs.push_back(
      {hasSubseq ? INPUT_HASSUB_SEQUENCE_DATA : INPUT_SEQUENCE_DATA, "layer_1",
       10, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();
  config.layerConfig.set_trans_type(trans_type);
  LOG(INFO) << " trans_type=" << trans_type << " hasSubseq=" << hasSubseq;

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "expand", 30, false, useGpu);
  }
}

TEST(Layer, ExpandLayer) {
  testExpandLayer("non-seq", false);  // non-seq expand to seq
  testExpandLayer("non-seq", true);   // non-seq expand to hasSubseq
  testExpandLayer("seq", true);       // seq expand to hasSubseq
}

void testDegradeLayer(bool hasSubseq, string layer_type, string trans_type) {
  TestConfig config;
  config.layerConfig.set_type(layer_type);
  config.layerConfig.set_size(10);
  config.biasSize = 0;

  config.inputDefs.push_back(
      {hasSubseq ? INPUT_HASSUB_SEQUENCE_DATA : INPUT_SEQUENCE_DATA, "layer_0",
       10, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.set_trans_type(trans_type);

  auto testDegradeLayerGrad = [](TestConfig& config, string layer_type) {
    for (auto useGpu : {false, true}) {
      testLayerGrad(config, layer_type, 100, false, useGpu);
    }
  };

  if (layer_type == "average") {
    for (auto strategy : {"average", "sum", "squarerootn"}) {
      LOG(INFO) << " hasSubseq=" << hasSubseq << " trans_type=" << trans_type
                << " average_strategy=" << strategy;
      config.layerConfig.set_average_strategy(strategy);
      testDegradeLayerGrad(config, layer_type);
    }
  } else {
    LOG(INFO) << " hasSubseq=" << hasSubseq << " trans_type=" << trans_type;
    testDegradeLayerGrad(config, layer_type);
  }
}

TEST(Layer, MaxLayer) {
  testDegradeLayer(false, "max", "non-seq");  // seq max to non-seq
  testDegradeLayer(true, "max", "non-seq");   // hasSubseq max to non-seq
  testDegradeLayer(true, "max", "seq");       // hasSubseq max to seq
}

TEST(Layer, SequenceLastInstanceLayer) {
  testDegradeLayer(false, "seqlastins",
                   "non-seq");  // seq seqlastins to non-seq
  testDegradeLayer(true, "seqlastins",
                   "non-seq");  // hasSubseq seqlastins to non-seq
  testDegradeLayer(true, "seqlastins", "seq");  // hasSubseq seqlastins to seq
}

TEST(Layer, AverageLayer) {
  testDegradeLayer(false, "average", "non-seq");  // seq average to non-seq
  testDegradeLayer(true, "average", "non-seq");  // hasSubseq average to non-seq
  testDegradeLayer(true, "average", "seq");      // hasSubseq average to seq
}

TEST(Layer, SequenceConcatLayer) {
  TestConfig config;
  config.layerConfig.set_type("seqconcat");
  config.layerConfig.set_size(10);
  config.biasSize = 0;

  config.inputDefs.push_back({INPUT_SEQUENCE_DATA, "layer_0", 10, 0});
  config.layerConfig.add_inputs();
  config.inputDefs.push_back({INPUT_SEQUENCE_DATA, "layer_1", 10, 0});
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "seqconcat", 100, false, useGpu);
  }
}

TEST(Layer, SequenceReshapeLayer) {
  TestConfig config;
  config.layerConfig.set_type("seqreshape");
  config.layerConfig.set_size(10);

  config.inputDefs.push_back({INPUT_SEQUENCE_DATA, "layer_0", 100, 0});
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "seqreshape", 100, false, useGpu);
  }
}

TEST(Layer, ConvShiftLayer) {
  TestConfig config;
  config.layerConfig.set_type("conv_shift");
  config.layerConfig.set_size(10);

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 10, 0});
  config.inputDefs.push_back({INPUT_DATA, "layer_1", 3, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  // Not support GPU now
  testLayerGrad(config, "conv_shift", 100, false, false);
}

TEST(Layer, PowerLayer) {
  TestConfig config;
  config.layerConfig.set_type("power");
  config.layerConfig.set_size(10);

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 1, 0});
  config.inputDefs.push_back({INPUT_DATA, "layer_1", 10, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "power", 100, false, useGpu);
  }
}

TEST(Layer, ConvexCombinationLayer) {
  TestConfig config;
  config.layerConfig.set_type("convex_comb");
  config.layerConfig.set_size(20);
  config.biasSize = 0;

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 5, 0});
  config.inputDefs.push_back({INPUT_DATA, "layer_1", 100, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "convex_comb", 100, false, useGpu);
  }
}

TEST(Layer, InterpolationLayer) {
  TestConfig config;
  config.layerConfig.set_type("interpolation");
  config.layerConfig.set_size(10);
  config.biasSize = 0;

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 1, 0});
  config.inputDefs.push_back({INPUT_DATA, "layer_1", 10, 0});
  config.inputDefs.push_back({INPUT_DATA, "layer_2", 10, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "interpolation", 100, false, useGpu);
  }
}

TEST(Layer, OuterProdLayer) {
  TestConfig config;
  config.layerConfig.set_type("out_prod");
  config.layerConfig.set_size(100);

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 10, 0});
  config.layerConfig.add_inputs();
  config.inputDefs.push_back({INPUT_DATA, "layer_1", 10, 0});
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "out_prod", 100, false, useGpu);
  }
}

TEST(Layer, SlopeInterceptLayer) {
  TestConfig config;
  config.layerConfig.set_type("slope_intercept");
  config.layerConfig.set_size(10);
  config.layerConfig.set_slope(1.0);
  config.layerConfig.set_intercept(0.1);

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 10, 0});
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "slope_intercept", 100, false, useGpu);
  }
}

TEST(Layer, ScalingLayer) {
  TestConfig config;
  config.layerConfig.set_type("scaling");
  config.layerConfig.set_size(10);
  config.biasSize = 0;

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 1, 0});
  config.layerConfig.add_inputs();
  config.inputDefs.push_back({INPUT_DATA, "layer_1", 10, 0});
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "scaling", 100, false, useGpu);
  }
}

void testNormLayer(const string& normType, bool trans, bool useGpu) {
  TestConfig config;
  config.layerConfig.set_type("norm");
  config.layerConfig.set_active_type("relu");

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 3136, 0});
  LayerInputConfig* input = config.layerConfig.add_inputs();
  NormConfig* norm = input->mutable_norm_conf();
  norm->set_norm_type(normType);
  norm->set_channels(16);
  norm->set_size(5);
  norm->set_scale(0.001);
  norm->set_pow(0.75);
  norm->set_blocked(0);
  norm->set_img_size(14);
  norm->set_output_x(norm->img_size());
  if (norm->norm_type() == "cmrnorm" ||
      norm->norm_type() == "cmrnorm-projection") {
    norm->set_scale(norm->scale() / norm->size());
  } else {
    norm->set_scale(norm->scale() / (norm->size() * norm->size()));
  }

  config.layerConfig.set_size(norm->output_x() * norm->output_x() *
                              norm->channels());
  config.biasSize = 0;

  testLayerGrad(config, "norm", 100, trans, useGpu);
}

#ifndef PADDLE_ONLY_CPU
TEST(Layer, NormLayer) {
  testNormLayer("cmrnorm-projection", /* trans= */ false, /* useGpu= */ true);
}
#endif

void setPoolConfig(TestConfig* config, PoolConfig* pool,
                   const string& poolType) {
  (*config).biasSize = 0;
  (*config).layerConfig.set_type("pool");
  (*config).layerConfig.set_num_filters(16);

942 943 944
  int kw = 3, kh = 3;
  int pw = 0, ph = 0;
  int sw = 2, sh = 2;
Z
zhangjinchao01 已提交
945 946
  pool->set_pool_type(poolType);
  pool->set_channels(16);
947 948 949 950 951 952 953 954
  pool->set_size_x(kw);
  pool->set_size_y(kh);
  pool->set_start(0);
  pool->set_padding(pw);
  pool->set_padding_y(ph);
  pool->set_stride(sw);
  pool->set_stride_y(sh);

955 956
  int ow = outputSize(pool->img_size(), kw, pw, sw, /* caffeMode */ false);
  int oh = outputSize(pool->img_size_y(), kh, ph, sh, /* caffeMode */ false);
957 958
  pool->set_output_x(ow);
  pool->set_output_y(oh);
Z
zhangjinchao01 已提交
959 960 961 962 963 964 965 966 967
}

void testPoolLayer(const string& poolType, bool trans, bool useGpu) {
  TestConfig config;
  config.inputDefs.push_back({INPUT_DATA, "layer_0", 3136, 0});
  LayerInputConfig* input = config.layerConfig.add_inputs();
  PoolConfig* pool = input->mutable_pool_conf();

  pool->set_img_size(14);
968 969 970
  pool->set_img_size_y(14);
  setPoolConfig(&config, pool, poolType);
  config.layerConfig.set_size(pool->output_x() * pool->output_y() *
Z
zhangjinchao01 已提交
971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986
                              pool->channels());

  testLayerGrad(config, "pool", 100, trans, useGpu);
}

#ifndef PADDLE_ONLY_CPU
void testPoolLayer2(const string& poolType, bool trans, bool useGpu) {
  TestConfig config;
  config.inputDefs.push_back({INPUT_DATA, "layer_0", 3200, 0});
  LayerInputConfig* input = config.layerConfig.add_inputs();
  PoolConfig* pool = input->mutable_pool_conf();

  pool->set_size_y(4);
  pool->set_stride_y(3);
  pool->set_img_size(10);
  pool->set_img_size_y(20);
987
  setPoolConfig(&config, pool, poolType);
Z
zhangjinchao01 已提交
988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011
  pool->set_output_y((pool->img_size_y() - pool->start() - pool->size_y()) /
                         ((float)pool->stride_y()) +
                     1.5);
  config.layerConfig.set_size(pool->output_x() * pool->output_y() *
                              pool->channels());

  testLayerGrad(config, "pool", 100, trans, useGpu);
}
#endif

TEST(Layer, PoolLayer) {
  testPoolLayer("avg-projection", /* trans= */ false, /* useGpu= */ false);
  testPoolLayer("max-projection", /* trans= */ false, /* useGpu= */ false);

#ifndef PADDLE_ONLY_CPU
  testPoolLayer("avg-projection", /* trans= */ false, /* useGpu= */ true);
  testPoolLayer("max-projection", /* trans= */ false, /* useGpu= */ true);
  testPoolLayer("cudnn-max-pool", /* trans= */ false, /* useGpu= */ true);
  testPoolLayer("cudnn-avg-pool", /* trans= */ false, /* useGpu= */ true);
  testPoolLayer2("cudnn-max-pool", /* trans= */ false, /* useGpu= */ true);
  testPoolLayer2("cudnn-avg-pool", /* trans= */ false, /* useGpu= */ true);
#endif
}

Q
qijun 已提交
1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023
void testSppLayer(const string& poolType, const int pyramidHeight, bool trans,
                  bool useGpu) {
  TestConfig config;
  config.layerConfig.set_type("spp");
  config.inputDefs.push_back({INPUT_DATA, "layer_0", 3200, 0});
  LayerInputConfig* input = config.layerConfig.add_inputs();
  SppConfig* sppConfig = input->mutable_spp_conf();
  sppConfig->set_pool_type(poolType);
  sppConfig->set_pyramid_height(pyramidHeight);
  sppConfig->set_channels(16);
  sppConfig->set_img_size(10);
  sppConfig->set_img_size_y(20);
Q
qijun 已提交
1024 1025
  int outputSize = (std::pow(4, sppConfig->pyramid_height()) - 1) / (4 - 1);
  config.layerConfig.set_size(outputSize * sppConfig->channels());
Q
qijun 已提交
1026 1027 1028 1029 1030
  testLayerGrad(config, "spp", 100, trans, useGpu);
}

TEST(Layer, SpatialPyramidPoolLayer) {
  for (auto useGpu : {false, true}) {
1031 1032 1033 1034
    for (auto pyramidHeight : {1, 2, 3}) {
      testSppLayer("avg-projection", pyramidHeight, false, useGpu);
      testSppLayer("max-projection", pyramidHeight, false, useGpu);
    }
Q
qijun 已提交
1035 1036 1037
  }
}

Z
zhangjinchao01 已提交
1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054
TEST(Layer, rankCostLayer) {
  TestConfig config;
  config.layerConfig.set_type("rank-cost");
  config.biasSize = 0;

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 1, 0});
  config.inputDefs.push_back({INPUT_DATA, "layer_1", 1, 0});
  config.inputDefs.push_back({INPUT_DATA_TARGET, "layer_2", 1, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "rank-cost", 100, false, useGpu);
  }
}

X
xuwei06 已提交
1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067
TEST(Layer, sumCostLayer) {
  TestConfig config;
  config.layerConfig.set_type("sum_cost");
  config.biasSize = 0;

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 1, 0});
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "sum_cost", 100, false, useGpu);
  }
}

Z
zhangjinchao01 已提交
1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127
TEST(Layer, weightedRankCostLayer) {
  TestConfig config;
  config.layerConfig.set_type("rank-cost");
  config.biasSize = 0;

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 1, 0});
  config.inputDefs.push_back({INPUT_DATA, "layer_1", 1, 0});
  config.inputDefs.push_back({INPUT_DATA_TARGET, "layer_2", 1, 0});
  config.inputDefs.push_back({INPUT_DATA_TARGET, "layer_3", 1, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "weighted-rank-cost", 100, false, useGpu);
  }
}

TEST(Layer, TensorLayer) {
  TestConfig config;
  config.layerConfig.set_type("tensor");
  config.layerConfig.set_size(10);
  config.layerConfig.set_active_type("sigmoid");
  config.biasSize = config.layerConfig.size();

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 5, 250});
  config.inputDefs.push_back({INPUT_DATA, "layer_1", 5, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "tensor", 100, false, useGpu);
  }
}

TEST(Layer, RecurrentLayer) {
  TestConfig config;
  config.layerConfig.set_type("recurrent");
  config.layerConfig.set_size(4);
  config.layerConfig.set_active_type("tanh");
  config.biasSize = 4;

  config.inputDefs.push_back(
      {INPUT_SEQUENCE_DATA, "layer_0", /* dim= */ 4, /* paraSize= */ 16});
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    for (auto reversed : {false, true}) {
      config.layerConfig.set_reversed(reversed);
      config.testState = !reversed;
      testLayerGrad(config, "recurrent", 50, /* trans= */ false, useGpu);
    }
  }
}

TEST(Layer, LstmLayer) {
  TestConfig config;
  config.layerConfig.set_type("lstmemory");
  config.layerConfig.set_size(4);
1128
  config.layerConfig.set_active_type("tanh");
Z
zhangjinchao01 已提交
1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386
  config.layerConfig.set_active_state_type("sigmoid");
  config.layerConfig.set_active_gate_type("sigmoid");
  config.biasSize = 28;

  config.inputDefs.push_back(
      {INPUT_SEQUENCE_DATA, "layer_0", /* dim= */ 16, /* paraSize= */ 64});
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    for (auto reversed : {false, true}) {
      config.layerConfig.set_reversed(reversed);
      config.testState = !reversed;
      testLayerGrad(config, "lstmemory", 100, /* trans= */ false, useGpu);
    }
  }
  for (auto useGpu : {true}) {
    config.testBatchState = true;
    config.layerConfig.set_reversed(false);
    testLayerGrad(config, "lstmemory", 10, /* trans= */ false, useGpu);
  }
}

TEST(Layer, MDLstmLayer) {
  TestConfig config;
  config.layerConfig.set_type("mdlstmemory");
  config.layerConfig.set_size(4);
  config.layerConfig.set_active_type("sigmoid");
  config.layerConfig.set_active_state_type("sigmoid");
  config.layerConfig.set_active_gate_type("sigmoid");
  config.biasSize = 4 * 9;

  config.inputDefs.push_back(
      {INPUT_SEQUENCE_MDIM_DATA, "layer_0", 4 * 5, 4 * 4 * 5});
  config.layerConfig.add_inputs();
  config.layerConfig.add_directions(true);
  config.layerConfig.add_directions(true);

  for (auto useGpu : {false, true}) {
    for (int i = 0; i < 2; i++) {
      for (int j = 0; j < 2; j++) {
        config.layerConfig.set_directions(0, bool(i));
        config.layerConfig.set_directions(1, bool(j));
        testLayerGrad(config, "mdlstmemory", 100, false, useGpu);
      }
    }
  }
}

TEST(Layer, ParameterReluLayer) {
  auto testParameterReluLayer = [&](size_t inputSize, size_t channels) {
    TestConfig config;
    config.layerConfig.set_type("prelu");
    config.inputDefs.push_back({INPUT_DATA, "layer_0", inputSize, channels});
    config.layerConfig.add_inputs();
    config.layerConfig.set_size(inputSize);
    config.layerConfig.set_partial_sum(inputSize /
                                       channels);  // size of feature map
    for (auto useGpu : {false, true}) {
      testLayerGrad(config, "prelu", 100, false, useGpu);
    }
  };

  testParameterReluLayer(192, 1);
  testParameterReluLayer(192, 3);
  testParameterReluLayer(192, 192);
}

TEST(Layer, ResizeLayer) {
  TestConfig config;
  config.biasSize = 0;
  config.layerConfig.set_type("resize");
  config.layerConfig.set_size(64);

  config.inputDefs.push_back({INPUT_DATA, "layer_0", 16, 0});
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "resize", 100, false, useGpu);
  }
}

TEST(Layer, NCELayer) {
  TestConfig config;
  size_t numClasses = 4;
  config.layerConfig.set_type("nce");
  config.layerConfig.set_size(1);
  config.layerConfig.set_active_type("sigmoid");
  config.layerConfig.set_num_classes(numClasses);
  config.biasSize = numClasses;

  config.inputDefs.push_back(
      {INPUT_DATA, "layer_0", /* dim= */ 16, /* paraSize= */ 16 * numClasses});
  config.inputDefs.push_back(
      {INPUT_LABEL, "label", /* dim= */ numClasses, /* paraSize= */ 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  for (auto withWeight : {false, true}) {
    if (withWeight) {
      config.inputDefs.push_back(
          {INPUT_DATA_TARGET, "weight", /* dim= */ 1, /* paraSize= */ 0});
      config.layerConfig.add_inputs();
    }

    for (auto isIdLabel : {false, true}) {
      config.inputDefs[1] = {
          isIdLabel ? INPUT_LABEL : INPUT_SPARSE_NON_VALUE_DATA, "label",
          /* dim= */ numClasses,
          /* paraSize= */ 0};

      for (auto withDist : {false, true}) {
        config.layerConfig.clear_neg_sampling_dist();
        if (withDist) {
          double sum = 0;
          for (size_t i = 0; i < numClasses; ++i) {
            real p = rand();  // NOLINT use rand_r
            config.layerConfig.add_neg_sampling_dist(p);
            sum += p;
          }
          for (size_t i = 0; i < numClasses; ++i) {
            real p = config.layerConfig.neg_sampling_dist(i) / sum;
            config.layerConfig.set_neg_sampling_dist(i, p);
          }
        }
        LOG(INFO) << "NCELayer "
                  << " isIdLabel=" << isIdLabel << " withWeight=" << withWeight
                  << " withDist=" << withDist;
        // Not support GPU now
        testLayerGrad(config, "nce", 100, /* trans= */ false,
                      /* useGpu */ false);
      }
    }
  }
}

TEST(Layer, GatedRecurrentLayer) {
  TestConfig config;
  config.layerConfig.set_type("gated_recurrent");
  config.layerConfig.set_size(4);
  config.layerConfig.set_active_type("sigmoid");
  config.layerConfig.set_active_gate_type("sigmoid");
  config.biasSize = 12;

  config.inputDefs.push_back(
      {INPUT_SEQUENCE_DATA, "layer_0", /* dim= */ 12, /* paraSize= */ 48});
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    for (auto reversed : {false, true}) {
      config.layerConfig.set_reversed(reversed);
      config.testState = !reversed;
      testLayerGrad(config, "gated_recurrent", 100, /* trans= */ false, useGpu);
    }
  }
}

TEST(Layer, GruStepLayer) {
  TestConfig config;
  config.layerConfig.set_type("gru_step");
  config.layerConfig.set_size(4);
  config.layerConfig.set_active_type("sigmoid");
  config.layerConfig.set_active_gate_type("sigmoid");
  config.biasSize = 12;

  config.inputDefs.push_back(
      {INPUT_DATA, "layer_0", /* dim= */ 12, /* paraSize= */ 48});
  config.inputDefs.push_back(
      {INPUT_DATA, "layer_1", /* dim= */ 4, /* paraSize= */ 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "gruStep", 100, /* trans= */ false, useGpu);
  }
}

TEST(Layer, LstmStepLayer) {
  TestConfig config;
  config.layerConfig.set_type("lstm_step");
  config.layerConfig.set_size(4);
  config.layerConfig.set_active_type("sigmoid");
  config.layerConfig.set_active_state_type("sigmoid");
  config.layerConfig.set_active_gate_type("sigmoid");
  config.biasSize = 12;
  config.testAccumulate = false;

  config.inputDefs.push_back(
      {INPUT_DATA, "layer_0", /* dim= */ 16, /* paraSize= */ 0});
  config.inputDefs.push_back(
      {INPUT_DATA, "layer_1", /* dim= */ 4, /* paraSize= */ 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "lstmStep", 100, /* trans= */ false, useGpu);
  }
}

void testBatchNormLayer(const string& type, bool trans, bool useGpu) {
  TestConfig config;
  const int CHANNELS = 10;
  const int IMG_SIZE = 16;
  config.layerConfig.set_type(type);
  config.layerConfig.set_size(CHANNELS * IMG_SIZE * IMG_SIZE);
  config.layerConfig.set_active_type("sigmoid");
  config.biasSize = CHANNELS;
  config.inputDefs.push_back({INPUT_DATA, "layer_0",
                              /* dim= */ IMG_SIZE * IMG_SIZE * CHANNELS,
                              /* paraSize= */ CHANNELS});

  config.inputDefs.push_back({INPUT_DATA, "layer_1_running_mean", 1, CHANNELS});
  config.inputDefs.back().isStatic = true;
  config.inputDefs.push_back({INPUT_DATA, "layer_2_running_var", 1, CHANNELS});
  config.inputDefs.back().isStatic = true;

  LayerInputConfig* input = config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  ImageConfig* img_conf = input->mutable_image_conf();
  img_conf->set_channels(CHANNELS);
  img_conf->set_img_size(IMG_SIZE);

  testLayerGrad(config, "batch_norm", 64, /* trans= */ trans, useGpu,
                /* useWeight */ true);
}

TEST(Layer, BatchNormalizationLayer) {
  testBatchNormLayer("batch_norm", false, false);
#ifndef PADDLE_ONLY_CPU
  testBatchNormLayer("batch_norm", false, true);
  if (hl_get_cudnn_lib_version() >= int(4000)) {
    testBatchNormLayer("cudnn_batch_norm", false, true);
  }
#endif
}

TEST(Operator, conv) {
  TestConfig config;
  const int NUM_FILTERS = 16;
  const int FILTER_SIZE = 2;
  const int FILTER_SIZE_Y = 3;
  const int CHANNELS = 3;
  const int IMAGE_SIZE = 16;
  OperatorConfig& operatorConf = *config.layerConfig.add_operator_confs();
  operatorConf.set_type("conv");
  ConvConfig* conv = operatorConf.mutable_conv_conf();
  operatorConf.set_num_filters(NUM_FILTERS);
  conv->set_filter_size(FILTER_SIZE);
  conv->set_filter_size_y(FILTER_SIZE_Y);
  conv->set_channels(CHANNELS);
  conv->set_padding(0);
  conv->set_padding_y(1);
  conv->set_stride(2);
  conv->set_stride_y(2);
  conv->set_groups(1);
  conv->set_filter_channels(conv->channels() / conv->groups());
  conv->set_img_size(IMAGE_SIZE);
1387 1388 1389 1390 1391
  int output_x =
      outputSize(conv->img_size(), conv->filter_size(), conv->padding(),
                 conv->stride(), /* caffeMode */ true);
  conv->set_output_x(output_x);
  config.layerConfig.set_size(output_x * output_x *
Z
zhangjinchao01 已提交
1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450
                              config.layerConfig.num_filters());
  config.layerConfig.set_size(conv->output_x() * conv->output_x() *
                              NUM_FILTERS);

  config.inputDefs.push_back(
      {INPUT_DATA, "layer_0", IMAGE_SIZE * IMAGE_SIZE * CHANNELS, 0});
  config.inputDefs.push_back(
      {INPUT_DATA, "layer_1",
       FILTER_SIZE * FILTER_SIZE_Y * CHANNELS * NUM_FILTERS, 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  testOperatorGrad(config, operatorConf, 100, /*useGpu*/ true, false);
}

TEST(Layer, FeatureMapExpandLayer) {
  TestConfig config;
  config.layerConfig.set_type("featmap_expand");
  const int CHANNELS = 10;
  const int INPUT_SIZE = 100;
  config.layerConfig.set_size(INPUT_SIZE * CHANNELS);
  config.layerConfig.set_num_filters(CHANNELS);
  config.inputDefs.push_back({INPUT_SEQUENCE_DATA, "layer_0",
                              /* dim= */ INPUT_SIZE, /* paraSize= */ 0});
  config.layerConfig.add_inputs();
  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "featmap_expand",
                  /*batch_size*/ 100, /* trans= */ false, useGpu,
                  /* useWeight */ true);
  }
}

TEST(Layer, MultiplexLayer) {
  TestConfig config;
  const int LAYER_SIZE = 100;
  config.layerConfig.set_type("multiplex");
  config.layerConfig.set_size(LAYER_SIZE);

  config.inputDefs.push_back({INPUT_LABEL, "layer_0", 2, 0});
  config.inputDefs.push_back(
      {INPUT_DATA, "layer_1", /* dim= */ LAYER_SIZE, /* paraSize= */ 0});
  config.inputDefs.push_back(
      {INPUT_DATA, "layer_2", /* dim= */ LAYER_SIZE, /* paraSize= */ 0});
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();
  config.layerConfig.add_inputs();

  for (auto useGpu : {false, true}) {
    testLayerGrad(config, "multiplex", 512, /* trans= */ false, useGpu);
  }
}

int main(int argc, char** argv) {
  testing::InitGoogleTest(&argc, argv);
  initMain(argc, argv);
  FLAGS_thread_local_rand_use_global_seed = true;
  srand(1);
  return RUN_ALL_TESTS();
}