text_classification_press.cpp 10.3 KB
Newer Older
W
wangguibao 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
// Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
//
// 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 <sys/stat.h>
#include <sys/types.h>
#include <unistd.h>

#include <atomic>
#include <fstream>
#include <thread>  // NOLINT
22
#include "sdk-cpp/builtin_format.pb.h"
W
wangguibao 已提交
23 24 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 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 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207
#include "sdk-cpp/include/common.h"
#include "sdk-cpp/include/predictor_sdk.h"
#include "sdk-cpp/text_classification.pb.h"

using baidu::paddle_serving::sdk_cpp::Predictor;
using baidu::paddle_serving::sdk_cpp::PredictorApi;
using baidu::paddle_serving::predictor::text_classification::TextReqInstance;
using baidu::paddle_serving::predictor::text_classification::TextResInstance;
using baidu::paddle_serving::predictor::text_classification::Request;
using baidu::paddle_serving::predictor::text_classification::Response;

const char *g_test_file = "./data/text_classification/test_set.txt";
DEFINE_int32(batch_size, 50, "Set the batch size of test file.");
DEFINE_int32(concurrency, 1, "Set the max concurrent number of requests");

std::vector<std::vector<int64_t>> g_test_input;
std::vector<int> g_test_label;
std::vector<int> g_correct;
std::vector<std::vector<uint64_t>> g_round_time;

std::atomic<int> g_concurrency(0);

// Text Classification Data Feed
//
// Input format:
// ([termid list], truth_label)
// Where 'termid list' is a variant length id list, `truth label` is a single
// number (0 or 1)
//
const int MAX_LINE_SIZE = 1024 * 1024;
std::vector<std::vector<int>> g_pred_labels;
const float g_decision_boundary = 0.500;

class DataFeed {
 public:
  virtual ~DataFeed() {}
  virtual void init(std::vector<std::vector<int64_t>> *test_input,
                    std::vector<int> *test_label);
  std::vector<std::vector<int64_t>> *get_test_input() { return _test_input; }
  std::vector<int> *get_labels() { return _test_label; }
  uint32_t sample_id() { return _current_sample_id; }
  void set_sample_id(uint32_t sample_id) { _current_sample_id = sample_id; }

 private:
  std::vector<std::vector<int64_t>> *_test_input;
  std::vector<int> *_test_label;
  uint32_t _current_sample_id;
  int _batch_size;
};

void DataFeed::init(std::vector<std::vector<int64_t>> *test_input,
                    std::vector<int> *test_label) {
  _test_input = test_input;
  _test_label = test_label;
}

bool set_file(const char *filename) {
  std::ifstream ifs(filename);
  char *line = new char[MAX_LINE_SIZE];
  int len = 0;
  char *sequence_begin_ptr = NULL;
  char *sequence_end_ptr = NULL;
  char *id_begin_ptr = NULL;
  char *id_end_ptr = NULL;
  char *label_ptr = NULL;
  int label = -1;
  int id = -1;
  while (!ifs.eof()) {
    std::vector<int64_t> vec;
    ifs.getline(line, MAX_LINE_SIZE);
    len = strlen(line);
    if (line[0] != '(' || line[len - 1] != ')') {
      continue;
    }
    line[len - 1] = '\0';

    sequence_begin_ptr = strchr(line, '(') + 1;
    if (*sequence_begin_ptr != '[') {
      continue;
    }

    sequence_end_ptr = strchr(sequence_begin_ptr, ']');
    if (sequence_end_ptr == NULL) {
      continue;
    }
    *sequence_end_ptr = '\0';

    id_begin_ptr = sequence_begin_ptr;
    while (id_begin_ptr != NULL) {
      id_begin_ptr++;
      id_end_ptr = strchr(id_begin_ptr, ',');
      if (id_end_ptr != NULL) {
        *id_end_ptr = '\0';
      }
      id = atoi(id_begin_ptr);
      id_begin_ptr = id_end_ptr;
      vec.push_back(id);
    }

    label_ptr = strchr(sequence_end_ptr + 1, ',');
    if (label_ptr == NULL) {
      continue;
    }
    *label_ptr = '\0';

    label_ptr++;
    label = atoi(label_ptr);

    g_test_input.push_back(vec);
    g_test_label.push_back(label);
  }

  ifs.close();

  std::cout << "read record" << g_test_input.size() << std::endl;

  return 0;
}

int create_req(std::shared_ptr<DataFeed> data_feed, Request *req) {  // NOLINT
  std::vector<std::vector<int64_t>> *inputs = data_feed->get_test_input();
  uint32_t current_sample_id = data_feed->sample_id();
  int idx = 0;

  for (idx = 0;
       idx < FLAGS_batch_size && current_sample_id + idx < inputs->size();
       ++idx) {
    TextReqInstance *req_instance = req->add_instances();
    std::vector<int64_t> &sample = inputs->at(current_sample_id + idx);
    for (auto x : sample) {
      req_instance->add_ids(x);
    }
  }

  if (idx < FLAGS_batch_size) {
    return -1;
  }

  data_feed->set_sample_id(current_sample_id + FLAGS_batch_size);
  return 0;
}

void extract_res(const Request &req, const Response &res, int thread_id) {
  uint32_t sample_size = res.predictions_size();
  std::string err_string;
  for (uint32_t si = 0; si < sample_size; ++si) {
    const TextResInstance &res_instance = res.predictions(si);

    if (res_instance.class_1_prob() < g_decision_boundary) {
      g_pred_labels[thread_id].push_back(0);
    } else if (res_instance.class_1_prob() >= g_decision_boundary) {
      g_pred_labels[thread_id].push_back(1);
    }
  }
}

void thread_worker(PredictorApi *api, int thread_id) {
  std::shared_ptr<DataFeed> local_feed(new DataFeed());
  local_feed->init(&g_test_input, &g_test_label);

  Request req;
  Response res;

  api->thrd_initialize();

  while (true) {
    api->thrd_clear();

    req.Clear();
    res.Clear();

    Predictor *predictor = api->fetch_predictor("text_classification");
    if (!predictor) {
      LOG(ERROR) << "Failed fetch predictor: text_classification";
      return;
    }

    if (create_req(local_feed, &req) != 0) {
      break;
    }

    while (g_concurrency.load() >= FLAGS_concurrency) {
    }
    g_concurrency++;
#if 1
W
wangguibao 已提交
208
    LOG(INFO) << "Current concurrency " << g_concurrency.load();
W
wangguibao 已提交
209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247
#endif

    timeval start;
    timeval end;

    gettimeofday(&start, NULL);
    if (predictor->inference(&req, &res) != 0) {
      LOG(ERROR) << "failed call predictor with req:" << req.ShortDebugString();
      return;
    }
    gettimeofday(&end, NULL);

    g_round_time[thread_id].push_back(end.tv_sec * 1000 + end.tv_usec / 1000 -
                                      start.tv_sec * 1000 -
                                      start.tv_usec / 1000);

    extract_res(req, res, thread_id);

    g_concurrency--;
#if 1
    LOG(INFO) << "Done. Current concurrency " << g_concurrency.load();
#endif
  }  // while (true)

  std::vector<int> *truth_label = local_feed->get_labels();
  for (int i = 0; i < g_pred_labels[thread_id].size(); ++i) {
    if (g_pred_labels[thread_id][i] == truth_label->at(i)) {
      ++g_correct[thread_id];
    }
  }

  api->thrd_finalize();
}

int main(int argc, char **argv) {
  google::ParseCommandLineFlags(&argc, &argv, true);

  PredictorApi api;

W
wangguibao 已提交
248
// initialize logger instance
W
wangguibao 已提交
249 250 251 252 253 254 255 256 257 258 259 260 261 262 263
#ifdef BCLOUD
  logging::LoggingSettings settings;
  settings.logging_dest = logging::LOG_TO_FILE;

  std::string filename(argv[0]);
  filename = filename.substr(filename.find_last_of('/') + 1);
  settings.log_file = (std::string("./log/") + filename + ".log").c_str();
  settings.delete_old = logging::DELETE_OLD_LOG_FILE;
  logging::InitLogging(settings);

  logging::ComlogSinkOptions cso;
  cso.process_name = filename;
  cso.enable_wf_device = true;
  logging::ComlogSink::GetInstance()->Setup(&cso);
#else
W
wangguibao 已提交
264 265 266 267 268 269 270 271 272 273 274 275 276 277 278
  struct stat st_buf;
  int ret = 0;
  if ((ret = stat("./log", &st_buf)) != 0) {
    mkdir("./log", 0777);
    ret = stat("./log", &st_buf);
    if (ret != 0) {
      LOG(WARNING) << "Log path ./log not exist, and create fail";
      return -1;
    }
  }

  FLAGS_log_dir = "./log";
  google::InitGoogleLogging(strdup(argv[0]));
  FLAGS_logbufsecs = 0;
  FLAGS_logbuflevel = -1;
W
wangguibao 已提交
279
#endif
W
wangguibao 已提交
280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343

  g_pred_labels.resize(FLAGS_concurrency);
  g_correct.resize(FLAGS_concurrency);
  g_round_time.resize(FLAGS_concurrency);

  set_file(g_test_file);
  if (api.create("./conf", "predictors.prototxt") != 0) {
    LOG(ERROR) << "Failed create predictors api!";
    return -1;
  }

  uint64_t elapse_ms = 0;

  timeval start;
  gettimeofday(&start, NULL);

  std::vector<std::thread *> worker_threads;
  int i = 0;
  for (; i < FLAGS_concurrency; ++i) {
    worker_threads.push_back(new std::thread(thread_worker, &api, i));
  }

  for (i = 0; i < FLAGS_concurrency; ++i) {
    worker_threads[i]->join();
    delete worker_threads[i];
  }

  timeval end;
  gettimeofday(&end, NULL);

  api.destroy();

  elapse_ms = (end.tv_sec * 1000 + end.tv_usec / 1000) -
              (start.tv_sec * 1000 + start.tv_usec / 1000);

  uint64_t count = 0;
  uint64_t correct = 0;

  for (int i = 0; i < FLAGS_concurrency; ++i) {
    count += g_pred_labels[i].size();
    correct += g_correct[i];
  }

  std::vector<uint64_t> round_times;
  for (auto x : g_round_time) {
    round_times.insert(round_times.end(), x.begin(), x.end());
  }

  std::sort(round_times.begin(), round_times.end());

  int percent_pos_50 = round_times.size() * 0.5;
  int percent_pos_80 = round_times.size() * 0.8;
  int percent_pos_90 = round_times.size() * 0.9;
  int percent_pos_99 = round_times.size() * 0.99;
  int percent_pos_999 = round_times.size() * 0.999;

  uint64_t total_ms = 0;
  for (auto x : round_times) {
    total_ms += x;
  }

  LOG(INFO) << "Total requests: " << round_times.size();
  LOG(INFO) << "Max concurrency: " << FLAGS_concurrency;
  LOG(INFO) << "Elapse ms (wall-time): " << elapse_ms;
344 345 346

  double qps = 0.0;
  if (elapse_ms != 0) {
W
wangguibao 已提交
347
    qps = (static_cast<double>(count) / (elapse_ms / FLAGS_concurrency)) * 1000;
348
  }
W
wangguibao 已提交
349 350 351 352 353

  LOG(INFO) << "QPS: " << qps / FLAGS_batch_size << "/s";
  LOG(INFO) << "Accuracy " << static_cast<double>(correct) / count;

  LOG(INFO) << "Latency statistics: ";
354
  if (round_times.size() != 0) {
W
wangguibao 已提交
355 356
    LOG(INFO) << "Average ms: "
              << static_cast<float>(total_ms) / round_times.size();
357 358 359 360 361 362 363 364
    LOG(INFO) << "50 percent ms: " << round_times[percent_pos_50];
    LOG(INFO) << "80 percent ms: " << round_times[percent_pos_80];
    LOG(INFO) << "90 percent ms: " << round_times[percent_pos_90];
    LOG(INFO) << "99 percent ms: " << round_times[percent_pos_99];
    LOG(INFO) << "99.9 percent ms: " << round_times[percent_pos_999];
  } else {
    LOG(INFO) << "N/A";
  }
W
wangguibao 已提交
365 366 367 368 369

  return 0;
}

/* vim: set expandtab ts=4 sw=4 sts=4 tw=100: */