Add CSV logging to GenericLogger

[glenn-jocher]

Enable CSV logging for Classify training.

Signed-off-by: Glenn Jocher glenn.jocher@ultralytics.com

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Enhanced logging capabilities in YOLOv5 with CSV logger additions and parameter updates.

📊 Key Changes

• Transitioned save_dir to use the Pathlib module, enhancing file path handling.
• Introduced a CSV file logger for metrics, improving persistent and structured data storage.
• Enabled logging of updated hyperparameters during WandB runs.
• Streamlined the metric logging function for clarity and consistency across different loggers.

🎯 Purpose & Impact

• The use of Path ensures better cross-platform compatibility for file paths. 🌍
• CSV logging allows users to easily track model performance over time in a commonly used format, which is useful for analysis and reporting. 📉
• The ability to update parameters in WandB provides better experiment tracking and reproducibility, especially when tuning hyperparameters. 🔧
• Simplifying the log_metrics method reduces code redundancy and enhances maintainability. 🛠️

Overall, these changes aim to make the logging more robust, user-friendly, and conducive to comprehensive performance tracking.

[glenn-jocher]

Looks good!

                  epoch,             train/loss,               val/loss,  metrics/accuracy_top1,  metrics/accuracy_top5,                   lr/0
                      0,                0.86282,                0.85036,                0.87287,                0.98497,                0.00067
                      1,                0.61298,                 0.6996,                0.92968,                0.99338,                0.00034