mlflow.cpp

mlflow.cpp is a C++ library which communicates with MLflow Tracking via its REST API. This library supports Windows10 and linux.

Requiremtents

• C++17 or later
• cpp-httplib
• nlohmann/json

Implemented APIs

Currently, only the following APIs are implemented.

• Create Experiment
• Get Experiment
• Get Experiment By Name
• Create Run
• Log Metric
• Log Batch
• Set Tag
• Log Param
• Update Run

Also, mlflow::start_run() and mlflow::end_run() are implemented as helper functions.

Example

#include<mlflow.hpp>
#include<thread>voidsend_data(mlflow::Client& cli) {
usingnamespacestd::chrono;
mlflow::Param param{"PARAM1", "PARAM_VALUE1"};
cli.log_param(param);
cli.set_tag({"TAG1", "TAG_VALUE1"});
cli.log_batch({{"METRIC_KEY", "-1.0", 0}}, {{"PARAM2", "PARAM_VALUE2"}},
{{"TAG2", "TAG_VALUE2"}});
std::this_thread::sleep_for(100ms);
cli.log_metric({"METRIC_KEY", "0.0", 1});
std::this_thread::sleep_for(100ms);
cli.log_metric({ "METRIC_KEY", "1.0", 2 });
cli.set_tag({ "mlflow.note.content", "HELLO!" });
}
voidexample_low_level(void) {
mlflow::Client cli("http://localhost:5000");
auto exp = cli.get_experiment_by_name("Default");
std::string exp_id = exp.experiment_id;
auto run = cli.create_run(exp_id);
std::string run_id = run.info.run_id;
cli.set_runid(run_id);
cli.set_run_name("example_low_level");
cli.set_user_name();
cli.set_source_name();
cli.set_source_type();
send_data(cli);
cli.update_run(mlflow::RunStatus::FINISHED);
}
voidexample_high_level(void) {
mlflow::Client cli("http://localhost:5000");
cli.start_run("example_high_level");
send_data(cli);
// called in the destructor// ret_ = cli.end_run();
}
intmain(void) {
example_low_level();
example_high_level();
return0;
}