routix 0.0.13

Define routines as structured algorithmic workflows

routix

routix is a lightweight Python toolkit for designing, executing, and analyzing structured algorithmic workflows. It is designed to systematically manage complex experiments and enhance reproducibility.

Key Features

• Subroutine-based Execution Control: Flexibly manage workflows with SubroutineController and support detailed logging by tracking context with MethodContextManager.
• Structured Flow Validation: Validate workflow definitions in advance using SubroutineFlowValidator.
• Dynamic Data Objects: Conveniently manage hierarchical data and configurations as dot-accessible objects with DynamicDataObject.
• Accurate Time Measurement: Accurately measure and manage the execution time of experiments and each subroutine with ElapsedTimer.
• Time Series Data Management: Collect and store various metrics generated during experiments as time series data using MetricTimeSeries and NamedTimeSeriesStore.
• Solution Management: Systematically manage solutions found during the optimization process and track the incumbent solution with SolutionManager.
• Experiment Reporting and Statistics: A modular system to record, analyze, and serialize experiment results.
• Extensible Runners: Abstract base classes to easily implement various experimental patterns (e.g., single, multiple, concurrent execution).
• File I/O Utilities: Simplified serialization of Python objects to YAML or JSON, and creation of timestamped directories for experiment outputs.

Subroutine Flow Data Format

routix executes workflows defined as a structured list of dictionaries. Each step is clearly specified with a method name and parameters.

- method: initialize
- method: repeat
  params:
    n_repeats: 3
    routine_data:
      - method: sample_method
        params:
          value: 42

For more details, see subroutine_flow_data.md.

Report System

routix provides a modular and extensible reporting system designed with the Single Responsibility Principle in mind.

• SubroutineReport: An immutable dataclass that records the results of a single subroutine execution, including elapsed time, objective value, and objective bound.
• SubroutineReportRecorder: Collects SubroutineReport instances and tracks method call counts during a workflow run.
• SubroutineReportStatistics: Computes summary statistics (e.g., total time, best objective, improvement ratio) from the reports collected by a recorder.
• SubroutineReportStatisticsSerializer: Serializes a collection of SubroutineReportStatistics objects into various formats, including dictionaries, CSV, JSON, and YAML, making it easy to export and analyze results from multiple runs.

Runner Base Classes

Build custom workflow runners by extending these abstract base classes to fit your experimental patterns.

• SingleInstanceRunner: The foundation for running a workflow on a single problem instance. Subclasses must implement get_controller() to set up the SubroutineController and post_run_process() to handle results.
• MultiInstanceRunner: Orchestrates running a set of instances sequentially. It uses a specified SingleInstanceRunner subclass for each instance.
• MultiInstanceConcurrentRunner: Extends MultiInstanceRunner to execute multiple instances in parallel using a process pool, significantly speeding up large-scale experiments.
• MultiScenarioRunner: Manages experiments across multiple scenarios, where each scenario may have a different configuration (e.g., a different subroutine flow). It uses a MultiInstanceRunner subclass to execute the instances within each scenario.

File I/O Utilities

routix includes helper functions to streamline file I/O for experiments.

• object_to_yaml / object_to_json: Serialize Python objects (including those with to_dict() methods and pathlib.Path objects) into clean and readable YAML or JSON files.
• init_timestamped_working_dir: Creates a unique, timestamped directory for storing experiment outputs, helping to keep results organized.

Installation

pip install routix

Testing

Unit tests for all major components of the project are included in the tests/ directory. You can run all tests using pytest.

pytest