testutils.cpp

/*!
 * Copyright (c) 2022 Microsoft Corporation. All rights reserved.
 * Licensed under the MIT License. See LICENSE file in the project root for license information.
 */

#include <testutils.h>
#include <LightGBM/c_api.h>
#include <LightGBM/utils/random.h>

#include <gtest/gtest.h>
#include <string>
#include <thread>
#include <utility>

using LightGBM::Log;
using LightGBM::Random;

namespace LightGBM {

  /*!
  * Creates a Dataset from the internal repository examples.
  */
  int TestUtils::LoadDatasetFromExamples(const char* filename, const char* config, DatasetHandle* out) {
    std::string fullPath("../examples/");
    fullPath += filename;
    Log::Info("Debug sample data path: %s", fullPath.c_str());
    return LGBM_DatasetCreateFromFile(
      fullPath.c_str(),
      config,
      nullptr,
      out);
  }

  /*!
  * Creates fake data in the passed vectors.
  */
  void TestUtils::CreateRandomDenseData(
    int32_t nrows,
    int32_t ncols,
    int32_t nclasses,
    std::vector<double>* features,
    std::vector<float>* labels,
    std::vector<float>* weights,
    std::vector<double>* init_scores,
    std::vector<int32_t>* groups) {
    Random rand(42);
    features->reserve(nrows * ncols);

    for (int32_t row = 0; row < nrows; row++) {
      for (int32_t col = 0; col < ncols; col++) {
        features->push_back(rand.NextFloat());
      }
    }

    CreateRandomMetadata(nrows, nclasses, labels, weights, init_scores, groups);
  }

  /*!
  * Creates fake data in the passed vectors.
  */
  void TestUtils::CreateRandomSparseData(
    int32_t nrows,
    int32_t ncols,
    int32_t nclasses,
    float sparse_percent,
    std::vector<int32_t>* indptr,
    std::vector<int32_t>* indices,
    std::vector<double>* values,
    std::vector<float>* labels,
    std::vector<float>* weights,
    std::vector<double>* init_scores,
    std::vector<int32_t>* groups) {
    Random rand(42);
    indptr->reserve(static_cast<int32_t>(nrows + 1));
    indices->reserve(static_cast<int32_t>(sparse_percent * nrows * ncols));
    values->reserve(static_cast<int32_t>(sparse_percent * nrows * ncols));

    indptr->push_back(0);
    for (int32_t row = 0; row < nrows; row++) {
      for (int32_t col = 0; col < ncols; col++) {
        float rnd = rand.NextFloat();
        if (rnd < sparse_percent) {
          indices->push_back(col);
          values->push_back(rand.NextFloat());
        }
      }
      indptr->push_back(static_cast<int32_t>(indices->size() - 1));
    }

    CreateRandomMetadata(nrows, nclasses, labels, weights, init_scores, groups);
  }

  /*!
  * Creates fake data in the passed vectors.
  */
  void TestUtils::CreateRandomMetadata(int32_t nrows,
    int32_t nclasses,
    std::vector<float>* labels,
    std::vector<float>* weights,
    std::vector<double>* init_scores,
    std::vector<int32_t>* groups) {
    Random rand(42);
    labels->reserve(nrows);
    if (weights) {
      weights->reserve(nrows);
    }
    if (init_scores) {
      init_scores->reserve(nrows * nclasses);
    }
    if (groups) {
      groups->reserve(nrows);
    }

    int32_t group = 0;

    for (int32_t row = 0; row < nrows; row++) {
      labels->push_back(rand.NextFloat());
      if (weights) {
        weights->push_back(rand.NextFloat());
      }
      if (init_scores) {
        for (int32_t i = 0; i < nclasses; i++) {
          init_scores->push_back(rand.NextFloat());
        }
      }
      if (groups) {
        if (rand.NextFloat() > 0.95) {
          group++;
        }
        groups->push_back(group);
      }
    }
  }

  void TestUtils::StreamDenseDataset(DatasetHandle dataset_handle,
    int32_t nrows,
    int32_t ncols,
    int32_t nclasses,
    int32_t batch_count,
    const std::vector<double>* features,
    const std::vector<float>* labels,
    const std::vector<float>* weights,
    const std::vector<double>* init_scores,
    const std::vector<int32_t>* groups) {
    int result = LGBM_DatasetSetWaitForManualFinish(dataset_handle, 1);
    EXPECT_EQ(0, result) << "LGBM_DatasetSetWaitForManualFinish result code: " << result;

    Log::Info("     Begin StreamDenseDataset");
    if ((nrows % batch_count) != 0) {
      Log::Fatal("This utility method only handles nrows that are a multiple of batch_count");
    }

    const double* features_ptr = features->data();
    const float* labels_ptr = labels->data();
    const float* weights_ptr = nullptr;
    if (weights) {
      weights_ptr = weights->data();
    }

    // Since init_scores are in a column format, but need to be pushed as rows, we have to extract each batch
    std::vector<double> init_score_batch;
    const double* init_scores_ptr = nullptr;
    if (init_scores) {
      init_score_batch.reserve(nclasses * batch_count);
      init_scores_ptr = init_score_batch.data();
    }

    const int32_t* groups_ptr = nullptr;
    if (groups) {
      groups_ptr = groups->data();
    }

    auto start_time = std::chrono::steady_clock::now();

    for (int32_t i = 0; i < nrows; i += batch_count) {
      if (init_scores) {
        init_scores_ptr = CreateInitScoreBatch(&init_score_batch, i, nrows, nclasses, batch_count, init_scores);
      }

      result = LGBM_DatasetPushRowsWithMetadata(dataset_handle,
                                                features_ptr,
                                                1,
                                                batch_count,
                                                ncols,
                                                i,
                                                labels_ptr,
                                                weights_ptr,
                                                init_scores_ptr,
                                                groups_ptr,
                                                0);
      EXPECT_EQ(0, result) << "LGBM_DatasetPushRowsWithMetadata result code: " << result;
      if (result != 0) {
        FAIL() << "LGBM_DatasetPushRowsWithMetadata failed";  // This forces an immediate failure, which EXPECT_EQ does not
      }

      features_ptr += batch_count * ncols;
      labels_ptr += batch_count;
      if (weights_ptr) {
        weights_ptr += batch_count;
      }
      if (groups_ptr) {
        groups_ptr += batch_count;
      }
    }

    auto cur_time = std::chrono::steady_clock::now();
    Log::Info(" Time: %d", cur_time - start_time);
  }

  void TestUtils::StreamSparseDataset(DatasetHandle dataset_handle,
                                      int32_t nrows,
                                      int32_t nclasses,
                                      int32_t batch_count,
                                      const std::vector<int32_t>* indptr,
                                      const std::vector<int32_t>* indices,
                                      const std::vector<double>* values,
                                      const std::vector<float>* labels,
                                      const std::vector<float>* weights,
                                      const std::vector<double>* init_scores,
                                      const std::vector<int32_t>* groups) {
    int result = LGBM_DatasetSetWaitForManualFinish(dataset_handle, 1);
    EXPECT_EQ(0, result) << "LGBM_DatasetSetWaitForManualFinish result code: " << result;

    Log::Info("     Begin StreamSparseDataset");
    if ((nrows % batch_count) != 0) {
      Log::Fatal("This utility method only handles nrows that are a multiple of batch_count");
    }

    const int32_t* indptr_ptr = indptr->data();
    const int32_t* indices_ptr = indices->data();
    const double* values_ptr = values->data();
    const float* labels_ptr = labels->data();
    const float* weights_ptr = nullptr;
    if (weights) {
      weights_ptr = weights->data();
    }

    const int32_t* groups_ptr = nullptr;
    if (groups) {
      groups_ptr = groups->data();
    }

    auto start_time = std::chrono::steady_clock::now();

    // Use multiple threads to test concurrency
    int thread_count = 2;
    if (nrows == batch_count) {
      thread_count = 1;  // If pushing all rows in 1 batch, we cannot have multiple threads
    }
    std::vector<std::thread> threads;
    threads.reserve(thread_count);
    for (int32_t t = 0; t < thread_count; ++t) {
      std::thread th(TestUtils::PushSparseBatch,
                     dataset_handle,
                     nrows,
                     nclasses,
                     batch_count,
                     indptr,
                     indptr_ptr,
                     indices_ptr,
                     values_ptr,
                     labels_ptr,
                     weights_ptr,
                     init_scores,
                     groups_ptr,
                     thread_count,
                     t);
      threads.push_back(move(th));
    }

    for (auto& t : threads) t.join();

    auto cur_time = std::chrono::steady_clock::now();
    Log::Info(" Time: %d", cur_time - start_time);
  }

  /*!
   * Pushes data from 1 thread into a Dataset based on thread_id and nrows.
   * e.g. with 100 rows, thread 0 will push rows 0-49, and thread 2 will push rows 50-99.
   * Note that rows are still pushed in microbatches within their range. 
   */
  void TestUtils::PushSparseBatch(DatasetHandle dataset_handle,
                                  int32_t nrows,
                                  int32_t nclasses,
                                  int32_t batch_count,
                                  const std::vector<int32_t>* indptr,
                                  const int32_t* indptr_ptr,
                                  const int32_t* indices_ptr,
                                  const double* values_ptr,
                                  const float* labels_ptr,
                                  const float* weights_ptr,
                                  const std::vector<double>* init_scores,
                                  const int32_t* groups_ptr,
                                  int32_t thread_count,
                                  int32_t thread_id) {
    int32_t threadChunkSize = nrows / thread_count;
    int32_t startIndex = threadChunkSize * thread_id;
    int32_t stopIndex = startIndex + threadChunkSize;

    indptr_ptr += threadChunkSize * thread_id;
    labels_ptr += threadChunkSize * thread_id;
    if (weights_ptr) {
      weights_ptr += threadChunkSize * thread_id;
    }
    if (groups_ptr) {
      groups_ptr += threadChunkSize * thread_id;
    }

    for (int32_t i = startIndex; i < stopIndex; i += batch_count) {
      // Since init_scores are in a column format, but need to be pushed as rows, we have to extract each batch
      std::vector<double> init_score_batch;
      const double* init_scores_ptr = nullptr;
      if (init_scores) {
        init_score_batch.reserve(nclasses * batch_count);
        init_scores_ptr = CreateInitScoreBatch(&init_score_batch, i, nrows, nclasses, batch_count, init_scores);
      }

      int32_t nelem = indptr->at(i + batch_count - 1) - indptr->at(i);

      int result = LGBM_DatasetPushRowsByCSRWithMetadata(dataset_handle,
                                                         indptr_ptr,
                                                         2,
                                                         indices_ptr,
                                                         values_ptr,
                                                         1,
                                                         batch_count + 1,
                                                         nelem,
                                                         i,
                                                         labels_ptr,
                                                         weights_ptr,
                                                         init_scores_ptr,
                                                         groups_ptr,
                                                         thread_id);
      EXPECT_EQ(0, result) << "LGBM_DatasetPushRowsByCSRWithMetadata result code: " << result;
      if (result != 0) {
        FAIL() << "LGBM_DatasetPushRowsByCSRWithMetadata failed";  // This forces an immediate failure, which EXPECT_EQ does not
      }

      indptr_ptr += batch_count;
      labels_ptr += batch_count;
      if (weights_ptr) {
        weights_ptr += batch_count;
      }
      if (groups_ptr) {
        groups_ptr += batch_count;
      }
    }
  }


  void TestUtils::AssertMetadata(const Metadata* metadata,
    const std::vector<float>* ref_labels,
    const std::vector<float>* ref_weights,
    const std::vector<double>* ref_init_scores,
    const std::vector<int32_t>* ref_groups) {
    const float* labels = metadata->label();
    auto nTotal = static_cast<int32_t>(ref_labels->size());
    for (auto i = 0; i < nTotal; i++) {
      EXPECT_EQ(ref_labels->at(i), labels[i]) << "Inserted data: " << ref_labels->at(i) << " at " << i;
      if (ref_labels->at(i) != labels[i]) {
        FAIL() << "Mismatched labels";  // This forces an immediate failure, which EXPECT_EQ does not
      }
    }

    const float* weights = metadata->weights();
    if (weights) {
      if (!ref_weights) {
        FAIL() << "Expected null weights";
      }
      for (auto i = 0; i < nTotal; i++) {
        EXPECT_EQ(ref_weights->at(i), weights[i]) << "Inserted data: " << ref_weights->at(i);
        if (ref_weights->at(i) != weights[i]) {
          FAIL() << "Mismatched weights";  // This forces an immediate failure, which EXPECT_EQ does not
        }
      }
    } else if (ref_weights) {
      FAIL() << "Expected non-null weights";
    }

    const double* init_scores = metadata->init_score();
    if (init_scores) {
      if (!ref_init_scores) {
        FAIL() << "Expected null init_scores";
      }
      for (size_t i = 0; i < ref_init_scores->size(); i++) {
        EXPECT_EQ(ref_init_scores->at(i), init_scores[i]) << "Inserted data: " << ref_init_scores->at(i) << " Index: " << i;
        if (ref_init_scores->at(i) != init_scores[i]) {
          FAIL() << "Mismatched init_scores";  // This forces an immediate failure, which EXPECT_EQ does not
        }
      }
    } else if (ref_init_scores) {
      FAIL() << "Expected non-null init_scores";
    }

    const int32_t* query_boundaries = metadata->query_boundaries();
    if (query_boundaries) {
      if (!ref_groups) {
        FAIL() << "Expected null query_boundaries";
      }
      // Calculate expected boundaries
      std::vector<int32_t> ref_query_boundaries;
      ref_query_boundaries.push_back(0);
      int group_val = ref_groups->at(0);
      for (auto i = 1; i < nTotal; i++) {
        if (ref_groups->at(i) != group_val) {
          ref_query_boundaries.push_back(i);
          group_val = ref_groups->at(i);
        }
      }
      ref_query_boundaries.push_back(nTotal);

      for (size_t i = 0; i < ref_query_boundaries.size(); i++) {
        EXPECT_EQ(ref_query_boundaries[i], query_boundaries[i]) << "Inserted data: " << ref_query_boundaries[i];
        if (ref_query_boundaries[i] != query_boundaries[i]) {
          FAIL() << "Mismatched query_boundaries";  // This forces an immediate failure, which EXPECT_EQ does not
        }
      }
    } else if (ref_groups) {
      FAIL() << "Expected non-null query_boundaries";
    }
  }

  const double* TestUtils::CreateInitScoreBatch(std::vector<double>* init_score_batch,
    int32_t index,
    int32_t nrows,
    int32_t nclasses,
    int32_t batch_count,
    const std::vector<double>* original_init_scores) {
    // Extract a set of rows from the column-based format (still maintaining column based format)
    init_score_batch->clear();
    for (int32_t c = 0; c < nclasses; c++) {
      for (int32_t row = index; row < index + batch_count; row++) {
        init_score_batch->push_back(original_init_scores->at(row + nrows * c));
      }
    }
    return init_score_batch->data();
  }

}  // namespace LightGBM