Logging vs. Printing

When you're hacking away in a Jupyter Notebook, print() is your best friend. But when you move to production scripts, pipelines, or APIs, print() becomes a liability.

The Limitation of Print

1. No Context: print("Error") tells you nothing about when it happened or where.
2. No Levels: You can't easily turn off "debug" prints while keeping "error" prints.
3. No Destinations: print only goes to standard output (console). You can't easily send it to a file, Slack, or a monitoring tool.

Enter the logging Module

Python's built-in logging module solves all these problems.

Basic Usage

import logging

# Configure the logger
logging.basicConfig(
    level=logging.INFO,
    format='%(asctime)s - %(levelname)s - %(message)s'
)

logging.debug("This is for detailed debugging info") # Won't show if level is INFO
logging.info("Training started")
logging.warning("Disk space low")
logging.error("Database connection failed")
logging.critical("System is down!")

Output:

2026-01-05 10:00:00,123 - INFO - Training started
2026-01-05 10:00:05,456 - WARNING - Disk space low

Why it's better for Data Science

1. Control Verbosity

You can leave logging.debug("Shape: " + str(df.shape)) in your code. When running in production, set the level to WARNING, and your logs stay clean. When debugging a crash, specific the level to DEBUG via an environment variable, and see all the details without changing code.

2. File Logging

You can easily write logs to a file to audit long-running training jobs.

logging.basicConfig(filename='training.log', level=logging.INFO)

Now you can close your terminal, come back tomorrow, and check training.log to see exactly what happened step-by-step.

3. Structured Logging

For advanced use cases, you can output logs in JSON format, which can be ingested by tools like Datadog or Splunk for fancy dashboards.

Best Practices

• Use __name__: Instantiate loggers with logger = logging.getLogger(__name__). This allows you to filter logs by module.
• Don't print exceptions: Use logger.exception("Message") inside except blocks to automatically include the full stack trace.

Stop printing. Start logging.