Resampling Time Series Data

In the real world, data does not always come in the format we want. Sometimes it is too detailed. Sometimes it is too noisy.

This is where resampling becomes extremely important.

Real-World Problem First

Imagine you work for a ride-sharing company. You are given minute-level ride demand data.

Your manager asks:

“Can you give me a clear monthly trend of demand?”

If you show minute-level data, it looks chaotic and unreadable.

What you really need is:

• Daily averages
• Weekly totals
• Monthly trends

This transformation is called resampling.

What Is Resampling?

Resampling means changing the time frequency of a time series.

Examples:

• Hourly → Daily
• Daily → Weekly
• Daily → Monthly

Resampling always involves two steps:

• Grouping data by time
• Applying an aggregation (mean, sum, max, etc.)

Let’s Create High-Frequency Data

We will simulate daily sales data for one full year.

This data contains:

• Growth over time
• Weekly seasonality
• Random fluctuations

Python: Creating Daily Data

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

np.random.seed(42)

dates = pd.date_range(start="2023-01-01", periods=365, freq="D")
trend = np.linspace(50, 120, 365)
weekly_seasonality = 10 * np.sin(2 * np.pi * np.arange(365) / 7)
noise = np.random.normal(0, 5, 365)

sales = trend + weekly_seasonality + noise

df = pd.DataFrame({"sales": sales}, index=dates)

Now let’s visualize the raw daily data.

What you should notice:

• Very noisy signal
• Hard to see long-term trend
• Too detailed for decision making

Why Daily Data Is Hard to Read

At daily level:

• Random events dominate
• Business patterns are hidden
• Executives get confused

So we resample.

Resampling to Weekly Data

First, we aggregate daily sales into weekly averages.

This removes short-term noise while keeping meaningful patterns.

Python: Weekly Resampling

weekly_sales = df.resample("W").mean()

Here is how the weekly data looks:

What changed?

• Noise reduced
• Trend becomes clearer
• Still responsive to changes

Weekly resampling is common in operations and logistics.

Resampling to Monthly Data

Now let’s go one level higher: monthly averages.

This is what executives usually want.

Python: Monthly Resampling

monthly_sales = df.resample("M").mean()

Here is the monthly view:

What you should clearly see now:

• Very smooth curve
• Clear upward growth
• No short-term distractions

This is perfect for:

• Business reports
• Strategy planning
• Long-term forecasting

Choosing the Right Aggregation

Resampling is not just about frequency — aggregation matters.

Always match aggregation with business meaning.

Common Mistakes to Avoid

• Resampling before cleaning missing values
• Using sum when mean is required
• Losing seasonality by over-aggregating

Once you resample too aggressively, information is lost forever.

Key Takeaways

• Resampling changes time frequency
• It simplifies noisy data
• Different stakeholders need different resolutions
• Always choose aggregation carefully

Next Lesson

In the next lesson, we’ll study rolling statistics — how moving averages and rolling windows reveal hidden patterns.