ExperimentTracker API Reference

Overview

ExperimentTracker is the core class for managing experiments in ORruns.

from orruns import ExperimentTracker

tracker = ExperimentTracker(
    name="experiment_name",
    base_dir="./orruns_experiments"
)

Core Methods

Log Parameters

tracker.log_params({
    "solver": "cplex",
    "threads": 4,
    "time_limit": 600
})

Log Metrics

tracker.log_metrics({
    "objective": 123.45,
    "solve_time": 10.5,
    "gap": 0.01
})

Save Artifacts

# Save plots
tracker.log_artifact("convergence.png", plt.gcf())

# Save data files
tracker.log_artifact("solution.csv", df)

Context Manager

with ExperimentTracker("experiment_name") as tracker:
    tracker.log_params({"solver": "gurobi"})
    # Your experiment code
    tracker.log_metrics({"objective": result.value})

Class Methods

Delete Experiments

# Delete specific experiment
ExperimentTracker.delete_experiment("experiment_name")

# Delete specific run
ExperimentTracker.delete_experiment(
    "experiment_name",
    run_id="run_20240315_123456"
)

Clean Old Experiments

# Delete experiments older than 30 days
ExperimentTracker.delete_all_experiments()

Properties

experiment_name: Name of the experiment
run_id: Unique identifier for current run
base_dir: Base directory for experiment data

Directory Structure

base_dir/
└── experiment_name/
    └── run_id/
        ├── params/
        │   └── params.json
        ├── metrics/
        │   └── metrics.json
        └── artifacts/
            ├── figures/
            └── data/

Best Practices

Use descriptive experiment names
Log all relevant parameters
Save intermediate results
Use context manager for automatic cleanup
Organize artifacts by type

Error Handling

Common errors and solutions:

Error	Solution
`FileNotFoundError`	Check base directory exists
`PermissionError`	Check write permissions
`ValueError`	Verify parameter/metric format