Summary of "Handling Date and Time Variables | Day 34 | 100 Days of Machine Learning"

Handling date & time variables (Day 34, 100 Days of ML)

Main idea

Date and time columns are highly informative. Convert columns stored as strings/objects to pandas datetime (pd.to_datetime) so you can extract many useful features (year, month, day, weekday, quarter, time parts, differences, etc.). The original video is a code-focused walkthrough demonstrating how to convert columns, extract features, and compute differences; the accompanying notebook is intended as a future reference.

Key concepts & lessons

Always convert string/object date/time columns to pandas datetime using pd.to_datetime before extracting components.
Use the .dt accessor to extract year, month, day, weekday, quarter, and time parts (hour/minute/second).
Many useful features are “hidden” inside a single datetime value (week number, day of week, month name, quarter, is-weekend, week-of-year, semester, day-of-year, etc.). Extract these as new columns to improve ML models.
Compute timedeltas by subtracting two datetime values; extract days, hours, minutes, seconds, or total seconds from the resulting timedelta.
Timezone work is uncommon for many problems but can be handled with appropriate libraries (pytz/dateutil); most tasks only need basic extraction.

Step-by-step methodology

Inspect dtypes
- Check whether a column is object/string or already datetime:
  - df.info() or df['date'].dtype
Convert string/object to datetime
- Example:
  - df['date'] = pd.to_datetime(df['date'])
  - df['time'] = pd.to_datetime(df['time']) (if you have a separate time column)
Extract calendar / date features (use .dt)
- Examples:
  - Year: df['year'] = df['date'].dt.year
  - Month (numeric): df['month'] = df['date'].dt.month
  - Month name: df['month_name'] = df['date'].dt.month_name()
  - Day of month: df['day'] = df['date'].dt.day
  - Day of week (numeric): df['dayofweek'] = df['date'].dt.dayofweek (Mon=0..Sun=6)
  - Day name: df['day_name'] = df['date'].dt.day_name()
  - Is weekend: df['is_weekend'] = df['date'].dt.dayofweek.isin([5, 6])
  - Week number: df['week'] = df['date'].dt.isocalendar().week (.dt.week is deprecated)
  - Day of year: df['day_of_year'] = df['date'].dt.dayofyear
  - Quarter: df['quarter'] = df['date'].dt.quarter
  - Semester (example logic): map quarters 1–2 -> semester 1, quarters 3–4 -> semester 2
Compute differences between dates (timedelta)
- Example diffs:
  - diff = pd.to_datetime('today') - df['date'] or df['date2'] - df['date1']
  - Days: diff.dt.days
  - Total seconds: diff.dt.total_seconds()
  - Convert to hours/minutes: diff.dt.total_seconds() / 3600 (hours) or / 60 (minutes)
- If you only want the day portion: diff.dt.days. For component-level access, use attributes of the timedelta or convert as needed.
Work with time-only parts
- After conversion:
  - Hour: df['hour'] = df['time'].dt.hour
  - Minute: df['minute'] = df['time'].dt.minute
  - Second: df['second'] = df['time'].dt.second
  - Time object only: df['time_only'] = df['time'].dt.time
- To compute time differences in seconds/minutes, use diff.dt.total_seconds() and divide as appropriate.

Notes & tips

Many practical use cases (expense tracking, chat/message timestamps, e-commerce order timestamps) benefit from these extracted features.
Keep the notebook and helper functions as a reference and reuse them in future projects that involve date/time.
Timezone handling is rare in basic tasks; use timezone-aware conversions only when needed.