Level 4 · Smooth the daily revenue chart

Why daily charts are noisy

A day’s revenue depends on a few dozen orders. One business customer buying a sofa set can double a Tuesday. Weekdays and weekends differ too. Plot every day and the line zigzags so much that the real movement, the slow build of the holiday season, hides behind single spikes. Readers then point at the tallest spike and tell a story about it.

From orders to days

order_date is a timestamp with a time of day, so grouping by it gives one group per order time. resample cuts a datetime index into calendar periods and fills days without orders with zero:

daily = lines.set_index("order_date").resample("D")["revenue"].sum()
len(daily)   # 365: one value per day of 2025

"D" is days, "W" weeks, "MS" months starting on the 1st.

A 7-day rolling mean

rolling(7) slides a window of seven rows along the Series. .mean() averages each window and labels it with its last day:

smooth = daily.rolling(7).mean()
smooth.head(7)   # six NaN, then the mean of Jan 1 to Jan 7

The first six values are missing because there aren’t seven days yet. Seven is the usual window for daily business data: every window holds each weekday exactly once, so the weekly rhythm cancels out and what’s left is the trend.

Reading the smoothed line

The smoothed value on December 4 is the average of November 28 to December 4: the week after Thanksgiving, with Black Friday, Cyber Monday and their BFCM25 coupon orders inside it. That’s the busiest week of 2025. The single busiest day, December 1, is part of it, but a single day is a weaker signal. Notice that the smoothed line lags: it peaks a few days after the rush starts, because each point looks back over a week.

Common mistakes

• Rolling over rows that aren’t days. If days with no orders are missing, rolling(7) spans more than a week. Resample first so there is exactly one row per day.
• Centring by accident, or not at all. rolling(7, center=True) labels each window with its middle day, which removes the lag but uses future days. Say which one your chart shows.
• Smoothing away the thing you were asked about. If the question is “what happened on Black Friday?”, show the daily bars. Smoothing answers questions about trends.