test_Evaluator.cpp 8.2 KB
Newer Older
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17

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 <vector>
#include "ModelConfig.pb.h"
18
#include "paddle/testing/TestUtil.h"
Y
Yu Yang 已提交
19
#include "paddle/trainer/Trainer.h"
20 21 22 23

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

24 25 26
DECLARE_bool(use_gpu);
DECLARE_int32(gpu_id);
DECLARE_bool(thread_local_rand_use_global_seed);
27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49

enum InputType {
  INPUT_DATA,         // dense vector
  INPUT_LABEL,        // id
  INPUT_DATA_TARGET,  // dense vector, but no gradient
  INPUT_SEQUENCE_DATA,
  INPUT_SEQUENCE_LABEL,
  INPUT_SPARSE_NON_VALUE_DATA
};

struct InputDef {
  InputType inputType;
  string name;
  size_t dim;
};

struct TestConfig {
  EvaluatorConfig evaluatorConfig;
  std::vector<InputDef> inputDefs;
  bool testAccumulate;
  TestConfig() : testAccumulate(true) {}
};

50 51 52 53
void testEvaluator(TestConfig testConf,
                   string testEvaluatorName,
                   size_t batchSize,
                   bool useGpu) {
54 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
#ifdef PADDLE_ONLY_CPU
  if (useGpu) return;
#endif
  FLAGS_use_gpu = useGpu;
  testConf.evaluatorConfig.set_name(testEvaluatorName);
  LOG(INFO) << " evaluator_type=" << testConf.evaluatorConfig.type()
            << " useGpu=" << useGpu;

  std::vector<Argument> arguments;
  for (size_t i = 0; i < testConf.inputDefs.size(); ++i) {
    Argument data;
    size_t dim = testConf.inputDefs[i].dim;
    switch (testConf.inputDefs[i].inputType) {
      case INPUT_DATA:
      case INPUT_SEQUENCE_DATA:
      case INPUT_DATA_TARGET:
        data.value = Matrix::create(batchSize, dim, false, useGpu);
        data.value->randomizeUniform();

        // make sure output > 0 && output < 1
        data.value->add(-0.5);
        data.value->sigmoid(*data.value);
        break;
      case INPUT_LABEL:
      case INPUT_SEQUENCE_LABEL:
        data.ids = VectorT<int>::create(batchSize, useGpu);
        data.ids->rand(dim);  // now rand number can be 0 to inputDefs[i].dim.
        break;
      case INPUT_SPARSE_NON_VALUE_DATA:
83 84 85 86
        data.value = makeRandomSparseMatrix(batchSize,
                                            dim,
                                            /* withValue= */ false,
                                            useGpu);
87 88 89 90 91 92
        break;
      default:
        LOG(FATAL) << " unknown inputType ";
        return;
    }

93 94 95 96 97 98 99 100 101
    ICpuGpuVectorPtr sequenceStartPositions;
    if (testConf.inputDefs[i].inputType == INPUT_SEQUENCE_DATA ||
        testConf.inputDefs[i].inputType == INPUT_SEQUENCE_LABEL) {
      if (!sequenceStartPositions) {
        generateSequenceStartPositions(batchSize, sequenceStartPositions);
      }
      data.sequenceStartPositions = sequenceStartPositions;
    }

102 103 104 105 106
    arguments.push_back(data);
  }

  Evaluator* testEvaluator = Evaluator::create(testConf.evaluatorConfig);
  double totalScore = 0.0;
107
  testEvaluator->start();
108 109
  totalScore += testEvaluator->evalImp(arguments);
  testEvaluator->updateSamplesNum(arguments);
110
  testEvaluator->finish();
111 112
  LOG(INFO) << *testEvaluator;

113 114 115 116
  std::vector<std::string> names;
  testEvaluator->getNames(&names);
  paddle::Error err;
  for (auto& name : names) {
Y
Yu Yang 已提交
117
    auto value = testEvaluator->getValue(name, &err);
118 119 120 121 122 123 124
    ASSERT_TRUE(err.isOK());
    LOG(INFO) << name << " " << value;
    auto tp = testEvaluator->getType(name, &err);
    ASSERT_TRUE(err.isOK());
    ASSERT_EQ(testConf.evaluatorConfig.type(), tp);
  }

125 126
  double totalScore2 = 0.0;
  if (testConf.testAccumulate) {
127
    testEvaluator->start();
128
    totalScore2 += testEvaluator->evalImp(arguments);
129
    testEvaluator->finish();
130 131 132 133
    EXPECT_LE(fabs(totalScore - totalScore2), 1.0e-5);
  }
}

134 135 136
void testEvaluatorAll(TestConfig testConf,
                      string testEvaluatorName,
                      size_t batchSize) {
137 138 139 140 141 142 143
  testEvaluator(testConf, testEvaluatorName, batchSize, true);
  testEvaluator(testConf, testEvaluatorName, batchSize, false);
}

TEST(Evaluator, classification_error) {
  TestConfig config;
  config.evaluatorConfig.set_type("classification_error");
144
  config.evaluatorConfig.set_top_k(5);
145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161

  config.inputDefs.push_back({INPUT_DATA, "output", 50});
  config.inputDefs.push_back({INPUT_LABEL, "label", 50});
  testEvaluatorAll(config, "classification_error", 100);
  config.inputDefs.push_back({INPUT_DATA, "weight", 1});
  testEvaluatorAll(config, "classification_error_weight", 100);

  // multi binary labels
  config.inputDefs.clear();
  config.inputDefs.push_back({INPUT_DATA, "output", 100});
  config.inputDefs.push_back({INPUT_SPARSE_NON_VALUE_DATA, "label", 100});
  // Not support GPU
  testEvaluator(config, "classification_error_multi_binary_label", 50, false);

  config.evaluatorConfig.set_classification_threshold(0.4);
  config.inputDefs.push_back({INPUT_DATA, "weight", 1});
  // Not support GPU
162 163
  testEvaluator(
      config, "classification_error_weight_multi_binary_label", 50, false);
164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230
}

TEST(Evaluator, sum) {
  TestConfig config;
  config.evaluatorConfig.set_type("sum");

  // sum of output
  config.inputDefs.push_back({INPUT_DATA, "output", 10});
  testEvaluatorAll(config, "sum_output", 200);
  config.inputDefs.push_back({INPUT_DATA, "weight", 1});
  testEvaluatorAll(config, "sum_output_weight", 200);

  // sum of label
  config.inputDefs.clear();
  config.inputDefs.push_back({INPUT_LABEL, "label", 10});
  testEvaluatorAll(config, "sum_label", 200);
  config.inputDefs.push_back({INPUT_DATA, "weight", 1});
  testEvaluatorAll(config, "sum_label_weight", 200);
}

TEST(Evaluator, last_column_sum) {
  TestConfig config;
  config.evaluatorConfig.set_type("last-column-sum");

  config.inputDefs.push_back({INPUT_DATA, "output", 50});
  testEvaluatorAll(config, "last-column-sum", 200);
  config.inputDefs.push_back({INPUT_DATA, "weight", 1});
  testEvaluatorAll(config, "last-column-sum_weight", 200);
}

TEST(Evaluator, last_column_auc) {
  TestConfig config;
  config.evaluatorConfig.set_type("last-column-auc");

  config.inputDefs.push_back({INPUT_DATA, "output", 2});
  config.inputDefs.push_back({INPUT_LABEL, "label", 2});
  testEvaluatorAll(config, "last-column-auc", 500);
  config.inputDefs.push_back({INPUT_DATA, "weight", 1});
  testEvaluatorAll(config, "last-column-auc_weight", 200);
}

TEST(Evaluator, precision_recall) {
  TestConfig config;
  config.evaluatorConfig.set_type("precision_recall");

  config.inputDefs.push_back({INPUT_DATA, "output", 10});
  config.inputDefs.push_back({INPUT_LABEL, "label", 10});
  testEvaluatorAll(config, "precision_recall", 200);
  config.inputDefs.push_back({INPUT_DATA, "weight", 1});
  testEvaluatorAll(config, "precision_recall_weight", 200);

  LOG(INFO) << "positive_label = 5";
  config.evaluatorConfig.set_positive_label(5);
  testEvaluatorAll(config, "precision_recall_weight", 200);

  // multi binary labels
  config.inputDefs.clear();
  config.evaluatorConfig.set_positive_label(-1);
  config.inputDefs.push_back({INPUT_DATA, "output", 10});
  config.inputDefs.push_back({INPUT_SPARSE_NON_VALUE_DATA, "label", 10});
  // Not support GPU
  testEvaluator(config, "precision_recall_multi_binary_label", 100, false);

  LOG(INFO) << "classification_threshold = 0.4";
  config.evaluatorConfig.set_classification_threshold(0.4);
  config.inputDefs.push_back({INPUT_DATA, "weight", 1});
  // Not support GPU
231 232
  testEvaluator(
      config, "precision_recall_weight_multi_binary_label", 100, false);
233 234
}

235 236 237 238 239 240 241 242 243
TEST(Evaluator, ctc_error_evaluator) {
  TestConfig config;
  config.evaluatorConfig.set_type("ctc_edit_distance");

  config.inputDefs.push_back({INPUT_SEQUENCE_DATA, "output", 32});
  config.inputDefs.push_back({INPUT_SEQUENCE_LABEL, "label", 1});
  testEvaluatorAll(config, "ctc_error_evaluator", 100);
}

244 245 246 247 248 249 250
int main(int argc, char** argv) {
  initMain(argc, argv);
  FLAGS_thread_local_rand_use_global_seed = true;
  srand(1);
  testing::InitGoogleTest(&argc, argv);
  return RUN_ALL_TESTS();
}
新手
引导
客服 返回
顶部