diff --git a/README.md b/README.md index 6ddd7c1..a3a4f6a 100644 --- a/README.md +++ b/README.md @@ -8,9 +8,21 @@ Bayesian Network Classifier with libtorch from scratch Before compiling BayesNet. +### Miniconda + +To be able to run Python Classifiers such as STree, ODTE, SVC, etc. it is needed to install Miniconda. To do so, download the installer from [Miniconda](https://docs.conda.io/en/latest/miniconda.html) and run it. It is recommended to install it in the home folder. + +In Linux sometimes the library libstdc++ is mistaken from the miniconda installation and produces the next message when running the b_xxxx executables: + +```bash +libstdc++.so.6: version `GLIBCXX_3.4.32' not found (required by b_xxxx) +``` + +The solution is to erase the libstdc++ library from the miniconda installation: + ### MPI -In Linux just install openmpi & openmpi-devel packages. Only cmake can't find openmpi install (like in Oracle Linux) set the following variable: +In Linux just install openmpi & openmpi-devel packages. Only if cmake can't find openmpi installation (like in Oracle Linux) set the following variable: ```bash export MPI_HOME="/usr/lib64/openmpi" diff --git a/src/Platform/GridSearch.cc b/src/Platform/GridSearch.cc index 76c4b4c..3e9ae3d 100644 --- a/src/Platform/GridSearch.cc +++ b/src/Platform/GridSearch.cc @@ -1,4 +1,5 @@ #include +#include #include #include "GridSearch.h" #include "Models.h" @@ -27,10 +28,15 @@ namespace platform { oss << std::put_time(timeinfo, "%H:%M:%S"); return oss.str(); } + std::string get_color_rank(int rank) + { + auto colors = { Colors::WHITE(), Colors::RED(), Colors::GREEN(), Colors::BLUE(), Colors::MAGENTA(), Colors::CYAN() }; + return *(colors.begin() + rank % colors.size()); + } GridSearch::GridSearch(struct ConfigGrid& config) : config(config) { } - json GridSearch::getResults() + json GridSearch::loadResults() { std::ifstream file(Paths::grid_output(config.model)); if (file.is_open()) { @@ -38,7 +44,7 @@ namespace platform { } return json(); } - vector GridSearch::processDatasets(Datasets& datasets) + std::vector GridSearch::filterDatasets(Datasets& datasets) const { // Load datasets auto datasets_names = datasets.getNames(); @@ -48,7 +54,7 @@ namespace platform { throw std::invalid_argument("Dataset " + config.continue_from + " not found"); } // Remove datasets already processed - vector< string >::iterator it = datasets_names.begin(); + std::vector::iterator it = datasets_names.begin(); while (it != datasets_names.end()) { if (*it != config.continue_from) { it = datasets_names.erase(it); @@ -71,54 +77,32 @@ namespace platform { } return datasets_names; } - void showProgressComb(const int num, const int n_folds, const int total, const std::string& color) - { - int spaces = int(log(total) / log(10)) + 1; - int magic = n_folds * 3 + 22 + 2 * spaces; - std::string prefix = num == 1 ? "" : string(magic, '\b') + string(magic + 1, ' ') + string(magic + 1, '\b'); - std::cout << prefix << color << "(" << setw(spaces) << num << "/" << setw(spaces) << total << ") " << Colors::RESET() << flush; - } - void showProgressFold(int fold, const std::string& color, const std::string& phase) - { - std::string prefix = phase == "a" ? "" : "\b\b\b\b"; - std::cout << prefix << color << fold << Colors::RESET() << "(" << color << phase << Colors::RESET() << ")" << flush; - } - std::string getColor(bayesnet::status_t status) - { - switch (status) { - case bayesnet::NORMAL: - return Colors::GREEN(); - case bayesnet::WARNING: - return Colors::YELLOW(); - case bayesnet::ERROR: - return Colors::RED(); - default: - return Colors::RESET(); - } - } - json GridSearch::build_tasks_mpi() + json GridSearch::build_tasks_mpi(int rank) { auto tasks = json::array(); auto grid = GridData(Paths::grid_input(config.model)); auto datasets = Datasets(false, Paths::datasets()); - auto datasets_names = processDatasets(datasets); - for (const auto& dataset : datasets_names) { + auto all_datasets = datasets.getNames(); + auto datasets_names = filterDatasets(datasets); + for (int idx_dataset = 0; idx_dataset < datasets_names.size(); ++idx_dataset) { + auto dataset = datasets_names[idx_dataset]; for (const auto& seed : config.seeds) { auto combinations = grid.getGrid(dataset); for (int n_fold = 0; n_fold < config.n_folds; n_fold++) { json task = { { "dataset", dataset }, + { "idx_dataset", idx_dataset}, { "seed", seed }, - { "fold", n_fold} + { "fold", n_fold}, }; tasks.push_back(task); } } } - // It's important to shuffle the array so heavy datasets are spread across the Workers + // Shuffle the array so heavy datasets are spread across the workers std::mt19937 g{ 271 }; // Use fixed seed to obtain the same shuffle std::shuffle(tasks.begin(), tasks.end(), g); - std::cout << "Tasks size: " << tasks.size() << std::endl; + std::cout << get_color_rank(rank) << "* Number of tasks: " << tasks.size() << std::endl; std::cout << "|"; for (int i = 0; i < tasks.size(); ++i) { std::cout << (i + 1) % 10; @@ -126,38 +110,19 @@ namespace platform { std::cout << "|" << std::endl << "|" << std::flush; return tasks; } - std::pair GridSearch::part_range_mpi(int n_tasks, int nprocs, int rank) + void process_task_mpi_consumer(struct ConfigGrid& config, struct ConfigMPI& config_mpi, json& tasks, int n_task, Datasets& datasets, Task_Result* result) { - int assigned = 0; - int remainder = n_tasks % nprocs; - int start = 0; - if (rank < remainder) { - assigned = n_tasks / nprocs + 1; - } else { - assigned = n_tasks / nprocs; - start = remainder; - } - start += rank * assigned; - int end = start + assigned; - if (rank == nprocs - 1) { - end = n_tasks; - } - return { start, end }; - } - std::string get_color_rank(int rank) - { - auto colors = { Colors::RED(), Colors::GREEN(), Colors::BLUE(), Colors::MAGENTA(), Colors::CYAN() }; - return *(colors.begin() + rank % colors.size()); - } - void GridSearch::process_task_mpi(struct ConfigMPI& config_mpi, json& task, Datasets& datasets, json& results) - { - // Process the task and store the result in the results json + // initialize Timer timer; timer.start(); - auto grid = GridData(Paths::grid_input(config.model)); + json task = tasks[n_task]; + auto model = config.model; + auto grid = GridData(Paths::grid_input(model)); auto dataset = task["dataset"].get(); + auto idx_dataset = task["idx_dataset"].get(); auto seed = task["seed"].get(); auto n_fold = task["fold"].get(); + bool stratified = config.stratified; // Generate the hyperparamters combinations auto combinations = grid.getGrid(dataset); auto [X, y] = datasets.getTensors(dataset); @@ -168,7 +133,7 @@ namespace platform { // Start working on task // Fold* fold; - if (config.stratified) + if (stratified) fold = new StratifiedKFold(config.n_folds, y, seed); else fold = new KFold(config.n_folds, y.size(0), seed); @@ -179,10 +144,11 @@ namespace platform { auto y_train = y.index({ train_t }); auto X_test = X.index({ "...", test_t }); auto y_test = y.index({ test_t }); - auto num = 0; double best_fold_score = 0.0; + int best_idx_combination = -1; json best_fold_hyper; - for (const auto& hyperparam_line : combinations) { + for (int idx_combination = 0; idx_combination < combinations.size(); ++idx_combination) { + auto hyperparam_line = combinations[idx_combination]; auto hyperparameters = platform::HyperParameters(datasets.getNames(), hyperparam_line); Fold* nested_fold; if (config.stratified) @@ -213,6 +179,7 @@ namespace platform { score /= config.nested; if (score > best_fold_score) { best_fold_score = score; + best_idx_combination = idx_combination; best_fold_hyper = hyperparam_line; } } @@ -225,43 +192,172 @@ namespace platform { clf->setHyperparameters(best_fold_hyper); clf->fit(X_train, y_train, features, className, states); best_fold_score = clf->score(X_test, y_test); - // Save results - results[dataset][std::to_string(n_fold)]["score"] = best_fold_score; - results[dataset][std::to_string(n_fold)]["hyperparameters"] = best_fold_hyper; - results[dataset][std::to_string(n_fold)]["seed"] = seed; - results[dataset][std::to_string(n_fold)]["duration"] = timer.getDuration(); + // Return the result + result->idx_dataset = task["idx_dataset"].get(); + result->idx_combination = best_idx_combination; + result->score = best_fold_score; + result->n_fold = n_fold; + result->time = timer.getDuration(); + // Update progress bar std::cout << get_color_rank(config_mpi.rank) << "*" << std::flush; } - void GridSearch::go_mpi(struct ConfigMPI& config_mpi) + json store_result(std::vector& names, Task_Result& result, json& results) + { + json json_result = { + { "score", result.score }, + { "combination", result.idx_combination }, + { "fold", result.n_fold }, + { "time", result.time }, + { "dataset", result.idx_dataset } + }; + auto name = names[result.idx_dataset]; + if (!results.contains(name)) { + results[name] = json::array(); + } + results[name].push_back(json_result); + return results; + } + json producer(std::vector& names, json& tasks, struct ConfigMPI& config_mpi, MPI_Datatype& MPI_Result) + { + Task_Result result; + json results; + int num_tasks = tasks.size(); + + // + // 2a.1 Producer will loop to send all the tasks to the consumers and receive the results + // + for (int i = 0; i < num_tasks; ++i) { + MPI_Status status; + MPI_Recv(&result, 1, MPI_Result, MPI_ANY_SOURCE, MPI_ANY_TAG, MPI_COMM_WORLD, &status); + if (status.MPI_TAG == TAG_RESULT) { + //Store result + store_result(names, result, results); + } + MPI_Send(&i, 1, MPI_INT, status.MPI_SOURCE, TAG_TASK, MPI_COMM_WORLD); + } + // + // 2a.2 Producer will send the end message to all the consumers + // + for (int i = 0; i < config_mpi.n_procs - 1; ++i) { + MPI_Status status; + MPI_Recv(&result, 1, MPI_Result, MPI_ANY_SOURCE, MPI_ANY_TAG, MPI_COMM_WORLD, &status); + if (status.MPI_TAG == TAG_RESULT) { + //Store result + store_result(names, result, results); + } + MPI_Send(&i, 1, MPI_INT, status.MPI_SOURCE, TAG_END, MPI_COMM_WORLD); + } + return results; + } + void select_best_results_folds(json& results, json& all_results, std::string& model) + { + Timer timer; + auto grid = GridData(Paths::grid_input(model)); + // + // Select the best result of the computed outer folds + // + for (const auto& result : all_results.items()) { + // each result has the results of all the outer folds as each one were a different task + double best_score = 0.0; + json best; + for (const auto& result_fold : result.value()) { + double score = result_fold["score"].get(); + if (score > best_score) { + best_score = score; + best = result_fold; + } + } + auto dataset = result.key(); + auto combinations = grid.getGrid(dataset); + json json_best = { + { "score", best_score }, + { "hyperparameters", combinations[best["combination"].get()] }, + { "date", get_date() + " " + get_time() }, + { "grid", grid.getInputGrid(dataset) }, + { "duration", timer.translate2String(best["time"].get()) } + }; + results[dataset] = json_best; + } + } + void consumer(Datasets& datasets, json& tasks, struct ConfigGrid& config, struct ConfigMPI& config_mpi, MPI_Datatype& MPI_Result) + { + Task_Result result; + // + // 2b.1 Consumers announce to the producer that they are ready to receive a task + // + MPI_Send(&result, 1, MPI_Result, config_mpi.manager, TAG_QUERY, MPI_COMM_WORLD); + int task; + while (true) { + MPI_Status status; + // + // 2b.2 Consumers receive the task from the producer and process it + // + MPI_Recv(&task, 1, MPI_INT, config_mpi.manager, MPI_ANY_TAG, MPI_COMM_WORLD, &status); + if (status.MPI_TAG == TAG_END) { + break; + } + process_task_mpi_consumer(config, config_mpi, tasks, task, datasets, &result); + // + // 2b.3 Consumers send the result to the producer + // + MPI_Send(&result, 1, MPI_Result, config_mpi.manager, TAG_RESULT, MPI_COMM_WORLD); + } + } + void GridSearch::go(struct ConfigMPI& config_mpi) { /* - * Each task is a json object with the following structure: - * { - * "dataset": "dataset_name", - * "seed": # of seed to use, - * "model": "model_name", - * "Fold": # of fold to process - * } - * - * The overall process consists in these steps: - * 1. Manager will broadcast the tasks to all the processes - * 1.1 Broadcast the number of tasks - * 1.2 Broadcast the length of the following string - * 1.2 Broadcast the tasks as a char* string - * 2. Workers will receive the tasks and start the process - * 2.1 A method will tell each worker the range of tasks to process - * 2.2 Each worker will process the tasks and generate the best score for each task - * 3. Manager gather the scores from all the workers and find out the best hyperparameters for each dataset - * 3.1 Obtain the maximum size of the results message of all the workers - * 3.2 Gather all the results from the workers into the manager - * 3.3 Compile the results from all the workers - * 3.4 Filter the best hyperparameters for each dataset - */ - char* msg; + * Each task is a json object with the following structure: + * { + * "dataset": "dataset_name", + * "idx_dataset": idx_dataset, // used to identify the dataset in the results + * // this index is relative to the used datasets in the actual run not to the whole datasets + * "seed": # of seed to use, + * "Fold": # of fold to process + * } + * + * The overall process consists in these steps: + * 0. Create the MPI result type & tasks + * 0.1 Create the MPI result type + * 0.2 Manager creates the tasks + * 1. Manager will broadcast the tasks to all the processes + * 1.1 Broadcast the number of tasks + * 1.2 Broadcast the length of the following string + * 1.2 Broadcast the tasks as a char* string + * 2a. Producer delivers the tasks to the consumers + * 2a.1 Producer will loop to send all the tasks to the consumers and receive the results + * 2a.2 Producer will send the end message to all the consumers + * 2b. Consumers process the tasks and send the results to the producer + * 2b.1 Consumers announce to the producer that they are ready to receive a task + * 2b.2 Consumers receive the task from the producer and process it + * 2b.3 Consumers send the result to the producer + * 3. Manager select the bests sccores for each dataset + * 3.1 Loop thru all the results obtained from each outer fold (task) and select the best + * 3.2 Save the results + */ + // + // 0.1 Create the MPI result type + // + Task_Result result; int tasks_size; + MPI_Datatype MPI_Result; + MPI_Datatype type[5] = { MPI_UNSIGNED, MPI_UNSIGNED, MPI_INT, MPI_DOUBLE, MPI_DOUBLE }; + int blocklen[5] = { 1, 1, 1, 1, 1 }; + MPI_Aint disp[5]; + disp[0] = offsetof(Task_Result, idx_dataset); + disp[1] = offsetof(Task_Result, idx_combination); + disp[2] = offsetof(Task_Result, n_fold); + disp[3] = offsetof(Task_Result, score); + disp[4] = offsetof(Task_Result, time); + MPI_Type_create_struct(5, blocklen, disp, type, &MPI_Result); + MPI_Type_commit(&MPI_Result); + // + // 0.2 Manager creates the tasks + // + char* msg; + json tasks; if (config_mpi.rank == config_mpi.manager) { timer.start(); - auto tasks = build_tasks_mpi(); + tasks = build_tasks_mpi(config_mpi.rank); auto tasks_str = tasks.dump(); tasks_size = tasks_str.size(); msg = new char[tasks_size + 1]; @@ -275,289 +371,35 @@ namespace platform { msg = new char[tasks_size + 1]; } MPI_Bcast(msg, tasks_size + 1, MPI_CHAR, config_mpi.manager, MPI_COMM_WORLD); - json tasks = json::parse(msg); + tasks = json::parse(msg); delete[] msg; - // - // 2. All Workers will receive the tasks and start the process - // - int num_tasks = tasks.size(); - // 2.1 A method will tell each worker the range of tasks to process - auto [start, end] = part_range_mpi(num_tasks, config_mpi.n_procs, config_mpi.rank); - // 2.2 Each worker will process the tasks and return the best scores obtained auto datasets = Datasets(config.discretize, Paths::datasets()); - json results; - for (int i = start; i < end; ++i) { - // Process task - process_task_mpi(config_mpi, tasks[i], datasets, results); - } - int size = results.dump().size() + 1; - int max_size = 0; - // - // 3. Manager gather the scores from all the workers and find out the best hyperparameters for each dataset - // - //3.1 Obtain the maximum size of the results message of all the workers - MPI_Allreduce(&size, &max_size, 1, MPI_INT, MPI_MAX, MPI_COMM_WORLD); - // Assign the memory to the message and initialize it to 0s - char* total = NULL; - msg = new char[max_size]; - strncpy(msg, results.dump().c_str(), size); if (config_mpi.rank == config_mpi.manager) { - total = new char[max_size * config_mpi.n_procs]; - } - // 3.2 Gather all the results from the workers into the manager - MPI_Gather(msg, max_size, MPI_CHAR, total, max_size, MPI_CHAR, config_mpi.manager, MPI_COMM_WORLD); - delete[] msg; - if (config_mpi.rank == config_mpi.manager) { - std::cout << Colors::RESET() << "|" << std::endl; - json total_results; - json best_results; - // 3.3 Compile the results from all the workers - for (int i = 0; i < config_mpi.n_procs; ++i) { - json partial_results = json::parse(total + i * max_size); - for (auto& [dataset, folds] : partial_results.items()) { - for (auto& [fold, result] : folds.items()) { - total_results[dataset][fold] = result; - } - } - } - delete[] total; - // 3.4 Filter the best hyperparameters for each dataset - auto grid = GridData(Paths::grid_input(config.model)); - for (auto& [dataset, folds] : total_results.items()) { - double best_score = 0.0; - double duration = 0.0; - json best_hyper; - for (auto& [fold, result] : folds.items()) { - duration += result["duration"].get(); - if (result["score"] > best_score) { - best_score = result["score"]; - best_hyper = result["hyperparameters"]; - } - } - auto timer = Timer(); - json result = { - { "score", best_score }, - { "hyperparameters", best_hyper }, - { "date", get_date() + " " + get_time() }, - { "grid", grid.getInputGrid(dataset) }, - { "duration", timer.translate2String(duration) } - }; - best_results[dataset] = result; - } - save(best_results); - } - } - void GridSearch::go() - { - timer.start(); - auto grid_type = config.nested == 0 ? "Single" : "Nested"; - auto datasets = Datasets(config.discretize, Paths::datasets()); - auto datasets_names = processDatasets(datasets); - json results = initializeResults(); - std::cout << "***************** Starting " << grid_type << " Gridsearch *****************" << std::endl; - std::cout << "input file=" << Paths::grid_input(config.model) << std::endl; - auto grid = GridData(Paths::grid_input(config.model)); - Timer timer_dataset; - double bestScore = 0; - json bestHyperparameters; - for (const auto& dataset : datasets_names) { - if (!config.quiet) - std::cout << "- " << setw(20) << left << dataset << " " << right << flush; - auto combinations = grid.getGrid(dataset); - timer_dataset.start(); - if (config.nested == 0) - // for dataset // for hyperparameters // for seed // for fold - tie(bestScore, bestHyperparameters) = processFileSingle(dataset, datasets, combinations); - else - // for dataset // for seed // for fold // for hyperparameters // for nested fold - tie(bestScore, bestHyperparameters) = processFileNested(dataset, datasets, combinations); - if (!config.quiet) { - std::cout << "end." << " Score: " << Colors::IBLUE() << setw(9) << setprecision(7) << fixed - << bestScore << Colors::BLUE() << " [" << bestHyperparameters.dump() << "]" - << Colors::RESET() << ::endl; - } - json result = { - { "score", bestScore }, - { "hyperparameters", bestHyperparameters }, - { "date", get_date() + " " + get_time() }, - { "grid", grid.getInputGrid(dataset) }, - { "duration", timer_dataset.getDurationString() } - }; - results[dataset] = result; - // Save partial results + // + // 2a. Producer delivers the tasks to the consumers + // + auto datasets_names = filterDatasets(datasets); + json all_results = producer(datasets_names, tasks, config_mpi, MPI_Result); + std::cout << get_color_rank(config_mpi.rank) << "|" << std::endl; + // + // 3. Manager select the bests sccores for each dataset + // + auto results = initializeResults(); + select_best_results_folds(results, all_results, config.model); + // + // 3.2 Save the results + // save(results); + } else { + // + // 2b. Consumers process the tasks and send the results to the producer + // + consumer(datasets, tasks, config, config_mpi, MPI_Result); } - // Save final results - save(results); - std::cout << "***************** Ending " << grid_type << " Gridsearch *******************" << std::endl; - } - pair GridSearch::processFileSingle(std::string fileName, Datasets& datasets, vector& combinations) - { - int num = 0; - double bestScore = 0.0; - json bestHyperparameters; - auto totalComb = combinations.size(); - for (const auto& hyperparam_line : combinations) { - if (!config.quiet) - showProgressComb(++num, config.n_folds, totalComb, Colors::CYAN()); - auto hyperparameters = platform::HyperParameters(datasets.getNames(), hyperparam_line); - // Get dataset - auto [X, y] = datasets.getTensors(fileName); - auto states = datasets.getStates(fileName); - auto features = datasets.getFeatures(fileName); - auto className = datasets.getClassName(fileName); - double totalScore = 0.0; - int numItems = 0; - for (const auto& seed : config.seeds) { - if (!config.quiet) - std::cout << "(" << seed << ") doing Fold: " << flush; - Fold* fold; - if (config.stratified) - fold = new StratifiedKFold(config.n_folds, y, seed); - else - fold = new KFold(config.n_folds, y.size(0), seed); - for (int nfold = 0; nfold < config.n_folds; nfold++) { - auto clf = Models::instance()->create(config.model); - auto valid = clf->getValidHyperparameters(); - hyperparameters.check(valid, fileName); - clf->setHyperparameters(hyperparameters.get(fileName)); - auto [train, test] = fold->getFold(nfold); - auto train_t = torch::tensor(train); - auto test_t = torch::tensor(test); - auto X_train = X.index({ "...", train_t }); - auto y_train = y.index({ train_t }); - auto X_test = X.index({ "...", test_t }); - auto y_test = y.index({ test_t }); - // Train model - if (!config.quiet) - showProgressFold(nfold + 1, getColor(clf->getStatus()), "a"); - clf->fit(X_train, y_train, features, className, states); - // Test model - if (!config.quiet) - showProgressFold(nfold + 1, getColor(clf->getStatus()), "b"); - totalScore += clf->score(X_test, y_test); - numItems++; - if (!config.quiet) - std::cout << "\b\b\b, \b" << flush; - } - delete fold; - } - double score = numItems == 0 ? 0.0 : totalScore / numItems; - if (score > bestScore) { - bestScore = score; - bestHyperparameters = hyperparam_line; - } - } - return { bestScore, bestHyperparameters }; - } - pair GridSearch::processFileNested(std::string fileName, Datasets& datasets, vector& combinations) - { - // Get dataset - auto [X, y] = datasets.getTensors(fileName); - auto states = datasets.getStates(fileName); - auto features = datasets.getFeatures(fileName); - auto className = datasets.getClassName(fileName); - int spcs_combinations = int(log(combinations.size()) / log(10)) + 1; - double goatScore = 0.0; - json goatHyperparameters; - // for dataset // for seed // for fold // for hyperparameters // for nested fold - for (const auto& seed : config.seeds) { - Fold* fold; - if (config.stratified) - fold = new StratifiedKFold(config.n_folds, y, seed); - else - fold = new KFold(config.n_folds, y.size(0), seed); - double bestScore = 0.0; - json bestHyperparameters; - std::cout << "(" << seed << ") doing Fold: " << flush; - for (int nfold = 0; nfold < config.n_folds; nfold++) { - if (!config.quiet) - std::cout << Colors::GREEN() << nfold + 1 << " " << flush; - // First level fold - auto [train, test] = fold->getFold(nfold); - auto train_t = torch::tensor(train); - auto test_t = torch::tensor(test); - auto X_train = X.index({ "...", train_t }); - auto y_train = y.index({ train_t }); - auto X_test = X.index({ "...", test_t }); - auto y_test = y.index({ test_t }); - auto num = 0; - json result_fold; - double hypScore = 0.0; - double bestHypScore = 0.0; - json bestHypHyperparameters; - for (const auto& hyperparam_line : combinations) { - std::cout << "[" << setw(spcs_combinations) << ++num << "/" << setw(spcs_combinations) - << combinations.size() << "] " << std::flush; - Fold* nested_fold; - if (config.stratified) - nested_fold = new StratifiedKFold(config.nested, y_train, seed); - else - nested_fold = new KFold(config.nested, y_train.size(0), seed); - for (int n_nested_fold = 0; n_nested_fold < config.nested; n_nested_fold++) { - // Nested level fold - auto [train_nested, test_nested] = nested_fold->getFold(n_nested_fold); - auto train_nested_t = torch::tensor(train_nested); - auto test_nested_t = torch::tensor(test_nested); - auto X_nexted_train = X_train.index({ "...", train_nested_t }); - auto y_nested_train = y_train.index({ train_nested_t }); - auto X_nested_test = X_train.index({ "...", test_nested_t }); - auto y_nested_test = y_train.index({ test_nested_t }); - // Build Classifier with selected hyperparameters - auto hyperparameters = platform::HyperParameters(datasets.getNames(), hyperparam_line); - auto clf = Models::instance()->create(config.model); - auto valid = clf->getValidHyperparameters(); - hyperparameters.check(valid, fileName); - clf->setHyperparameters(hyperparameters.get(fileName)); - // Train model - if (!config.quiet) - showProgressFold(n_nested_fold + 1, getColor(clf->getStatus()), "a"); - clf->fit(X_nexted_train, y_nested_train, features, className, states); - // Test model - if (!config.quiet) - showProgressFold(n_nested_fold + 1, getColor(clf->getStatus()), "b"); - hypScore += clf->score(X_nested_test, y_nested_test); - if (!config.quiet) - std::cout << "\b\b\b, \b" << flush; - } - int magic = 3 * config.nested + 2 * spcs_combinations + 4; - std::cout << string(magic, '\b') << string(magic, ' ') << string(magic, '\b') << flush; - delete nested_fold; - hypScore /= config.nested; - if (hypScore > bestHypScore) { - bestHypScore = hypScore; - bestHypHyperparameters = hyperparam_line; - } - } - // Build Classifier with selected hyperparameters - auto clf = Models::instance()->create(config.model); - clf->setHyperparameters(bestHypHyperparameters); - // Train model - if (!config.quiet) - showProgressFold(nfold + 1, getColor(clf->getStatus()), "a"); - clf->fit(X_train, y_train, features, className, states); - // Test model - if (!config.quiet) - showProgressFold(nfold + 1, getColor(clf->getStatus()), "b"); - double score = clf->score(X_test, y_test); - if (!config.quiet) - std::cout << string(2 * config.nested - 1, '\b') << "," << string(2 * config.nested, ' ') << string(2 * config.nested - 1, '\b') << flush; - if (score > bestScore) { - bestScore = score; - bestHyperparameters = bestHypHyperparameters; - } - } - if (bestScore > goatScore) { - goatScore = bestScore; - goatHyperparameters = bestHyperparameters; - } - delete fold; - } - return { goatScore, goatHyperparameters }; } json GridSearch::initializeResults() { - // Load previous results + // Load previous results if continue is set json results; if (config.continue_from != NO_CONTINUE()) { if (!config.quiet) diff --git a/src/Platform/GridSearch.h b/src/Platform/GridSearch.h index c00b2ee..ec1b3cb 100644 --- a/src/Platform/GridSearch.h +++ b/src/Platform/GridSearch.h @@ -30,24 +30,30 @@ namespace platform { int n_procs; int manager; }; + typedef struct { + uint idx_dataset; + uint idx_combination; + int n_fold; + double score; + double time; + } Task_Result; + const int TAG_QUERY = 1; + const int TAG_RESULT = 2; + const int TAG_TASK = 3; + const int TAG_END = 4; class GridSearch { public: explicit GridSearch(struct ConfigGrid& config); - void go(); - void go_mpi(struct ConfigMPI& config_mpi); + void go(struct ConfigMPI& config_mpi); ~GridSearch() = default; - json getResults(); + json loadResults(); static inline std::string NO_CONTINUE() { return "NO_CONTINUE"; } private: void save(json& results); json initializeResults(); - vector processDatasets(Datasets& datasets); - pair processFileSingle(std::string fileName, Datasets& datasets, std::vector& combinations); - pair processFileNested(std::string fileName, Datasets& datasets, std::vector& combinations); + std::vector filterDatasets(Datasets& datasets) const; struct ConfigGrid config; - pair part_range_mpi(int n_tasks, int nprocs, int rank); - json build_tasks_mpi(); - void process_task_mpi(struct ConfigMPI& config_mpi, json& task, Datasets& datasets, json& results); + json build_tasks_mpi(int rank); Timer timer; // used to measure the time of the whole process }; } /* namespace platform */ diff --git a/src/Platform/b_grid.cc b/src/Platform/b_grid.cc index d870353..6e9796d 100644 --- a/src/Platform/b_grid.cc +++ b/src/Platform/b_grid.cc @@ -32,13 +32,25 @@ void manageArguments(argparse::ArgumentParser& program) group.add_argument("--report").help("Report the computed hyperparameters").default_value(false).implicit_value(true); group.add_argument("--compute").help("Perform computation of the grid output hyperparameters").default_value(false).implicit_value(true); program.add_argument("--discretize").help("Discretize input datasets").default_value((bool)stoi(env.get("discretize"))).implicit_value(true); - program.add_argument("--mpi").help("Use MPI computing grid").default_value(false).implicit_value(true); program.add_argument("--stratified").help("If Stratified KFold is to be done").default_value((bool)stoi(env.get("stratified"))).implicit_value(true); program.add_argument("--quiet").help("Don't display detailed progress").default_value(false).implicit_value(true); program.add_argument("--continue").help("Continue computing from that dataset").default_value(platform::GridSearch::NO_CONTINUE()); program.add_argument("--only").help("Used with continue to compute that dataset only").default_value(false).implicit_value(true); program.add_argument("--exclude").default_value("[]").help("Datasets to exclude in json format, e.g. [\"dataset1\", \"dataset2\"]"); - program.add_argument("--nested").help("Do a double/nested cross validation with n folds").default_value(0).scan<'i', int>(); + program.add_argument("--nested").help("Set the double/nested cross validation number of folds").default_value(5).scan<'i', int>().action([](const std::string& value) { + try { + auto k = stoi(value); + if (k < 2) { + throw std::runtime_error("Number of nested folds must be greater than 1"); + } + return k; + } + catch (const runtime_error& err) { + throw std::runtime_error(err.what()); + } + catch (...) { + throw std::runtime_error("Number of nested folds must be an integer"); + }}); program.add_argument("--score").help("Score used in gridsearch").default_value("accuracy"); program.add_argument("-f", "--folds").help("Number of folds").default_value(stoi(env.get("n_folds"))).scan<'i', int>().action([](const std::string& value) { try { @@ -108,8 +120,8 @@ void list_results(json& results, std::string& model) + " Nested: " + (results["nested"].get() == 0 ? "False" : to_string(results["nested"].get())) ); std::cout << std::string(MAXL, '*') << std::endl; - int spaces = 0; - int hyperparameters_spaces = 0; + int spaces = 7; + int hyperparameters_spaces = 15; for (const auto& item : results["results"].items()) { auto key = item.key(); auto value = item.value(); @@ -128,11 +140,10 @@ void list_results(json& results, std::string& model) int index = 0; for (const auto& item : results["results"].items()) { auto color = odd ? Colors::CYAN() : Colors::BLUE(); - auto key = item.key(); auto value = item.value(); std::cout << color; std::cout << std::setw(3) << std::right << index++ << " "; - std::cout << left << setw(spaces) << key << " " << value["date"].get() + std::cout << left << setw(spaces) << item.key() << " " << value["date"].get() << " " << setw(8) << right << value["duration"].get() << " " << setw(8) << setprecision(6) << fixed << right << value["score"].get() << " " << value["hyperparameters"].dump() << std::endl; odd = !odd; @@ -171,11 +182,6 @@ int main(int argc, char** argv) } auto excluded = program.get("exclude"); config.excluded = json::parse(excluded); - if (program.get("mpi")) { - if (!compute || config.nested == 0) { - throw std::runtime_error("Cannot use --mpi without --compute or without --nested"); - } - } } catch (const exception& err) { cerr << err.what() << std::endl; @@ -195,26 +201,24 @@ int main(int argc, char** argv) list_dump(config.model); } else { if (compute) { - if (program.get("mpi")) { - struct platform::ConfigMPI mpi_config; - mpi_config.manager = 0; // which process is the manager - MPI_Init(&argc, &argv); - MPI_Comm_rank(MPI_COMM_WORLD, &mpi_config.rank); - MPI_Comm_size(MPI_COMM_WORLD, &mpi_config.n_procs); - grid_search.go_mpi(mpi_config); - if (mpi_config.rank == mpi_config.manager) { - auto results = grid_search.getResults(); - list_results(results, config.model); - std::cout << "Process took " << timer.getDurationString() << std::endl; - } - MPI_Finalize(); - } else { - grid_search.go(); + struct platform::ConfigMPI mpi_config; + mpi_config.manager = 0; // which process is the manager + MPI_Init(&argc, &argv); + MPI_Comm_rank(MPI_COMM_WORLD, &mpi_config.rank); + MPI_Comm_size(MPI_COMM_WORLD, &mpi_config.n_procs); + if (mpi_config.n_procs < 2) { + throw std::runtime_error("Cannot use --compute with less than 2 mpi processes, try mpirun -np 2 ..."); + } + grid_search.go(mpi_config); + if (mpi_config.rank == mpi_config.manager) { + auto results = grid_search.loadResults(); + list_results(results, config.model); std::cout << "Process took " << timer.getDurationString() << std::endl; } + MPI_Finalize(); } else { // List results - auto results = grid_search.getResults(); + auto results = grid_search.loadResults(); if (results.empty()) { std::cout << "** No results found" << std::endl; } else {