Configure the Heatmap
Formula Used
Daily aggregation: A(d) = Σxᵢ, mean(xᵢ), min(xᵢ), max(xᵢ), or n.
Week position: w(d) = floor((d − startMonday) / 7).
Weekday row: r(d) = ISOWeekday(d) − 1.
Min-max: z = (x − xmin) / (xmax − xmin).
Z-score: z = (x − μ) / σ.
Signed log: z = sign(x) × ln(1 + |x|).
Percentage: z = 100x / Σx.
Interpolation: x(t) = xa + (xb − xa)(t − a)/(b − a).
How to Use This Calculator
- Paste one date and numeric value on each line.
- Select optional date boundaries for focused analysis.
- Choose duplicate aggregation and missing date handling.
- Apply clipping, transformation, filtering, and color settings.
- Press Generate Heatmap to calculate and display results.
- Inspect metrics, hover over cells, or export outputs.
Understanding Calendar Heatmaps
Calendar heatmaps arrange daily values across weeks and weekdays. Their grid reveals seasonality without requiring long time-series axes. Dense periods, gaps, and unusual spikes become easier to compare.
Machine learning teams often track errors, drift, latency, or predictions. Daily aggregation converts noisy events into stable operational signals. The calendar layout then exposes recurring behavior across working cycles.
Each cell represents one date and its processed numeric value. Columns represent consecutive weeks, while rows represent seven weekdays. Color intensity maps magnitude through the selected visualization scale.
Data Processing Choices
Duplicate dates must be combined before the calendar grid forms. Sum works for volumes, while averages suit rates or scores. Minimum and maximum options highlight daily boundaries and extreme behavior.
Missing dates can influence interpretation more than visible values. Zero filling assumes no activity occurred during absent periods. Blank cells preserve uncertainty and prevent false numerical conclusions.
Interpolation estimates gaps between known observations using nearby values. Carry-forward repeats the latest value until another value appears. Both methods require careful judgment because they introduce synthetic data.
Normalization and Filtering
Normalization supports fair comparisons across differently scaled datasets. Min-max scaling compresses values between zero and one. Z-scores express each value through standard deviation distance.
Log transformation reduces skew from occasional large observations. Percentage scaling shows every day relative to total activity. Clipping limits extreme values before normalization and protects visual contrast.
Weekday filters help isolate business activity or weekend behavior. They also prevent irrelevant days from weakening visible patterns. Date boundaries restrict analysis to a focused experimental window.
Interpreting Results
Summary statistics provide context that colors alone cannot communicate. Mean and deviation describe central tendency and daily variability. Peak dates identify unusual events requiring deeper model investigation.
Use consistent aggregation when comparing multiple experiments or deployments. Keep missing-value rules documented beside every exported visualization. Review raw records whenever a pattern appears operationally important.
Calendar heatmaps support monitoring, diagnosis, reporting, and stakeholder communication. They compress lengthy timelines into an understandable analytical surface. Thoughtful settings produce reliable insights for machine learning decisions.
Practical Visualization Guidance
Color choices should match analytical questions carefully. Sequential palettes suit ordered, nonnegative values and counts. Diverging palettes emphasize meaningful centers, such as zero differences.
Hover details improve auditability during formal reviews. Visible dates connect chart cells with source records. Exports preserve findings for reports, audits, and model documentation.
Begin with raw values before complex transformations. Compare several settings and note every interpretation change. Reliable conclusions emerge when preprocessing choices remain transparent throughout.
Frequently Asked Questions
1. What data format should I enter?
Enter one ISO date and one numeric value per line. Use commas, tabs, semicolons, or pipes as separators. A header row may be included.
2. How are duplicate dates processed?
The selected aggregation combines every value sharing one date. Choose sum, average, minimum, maximum, or row count. The result reports how many duplicate rows were combined.
3. Which missing method should I choose?
Keep blanks when missing dates represent uncertainty. Use zero only when absence truly means no activity. Interpolation and carry-forward create estimated values, so document those choices.
4. When is min-max scaling useful?
Min-max scaling helps compare patterns within a fixed range. It maps the smallest value to zero. It maps the largest value to one.
5. What does Z-score scaling show?
Z-scores express distance from the mean using standard deviations. Positive values exceed the mean. Negative values fall below it.
6. Why use clipping limits?
Clipping reduces the visual influence of extreme observations. It can reveal ordinary variation hidden by large spikes. Keep original values available for separate investigation.
7. Can the calculator display negative values?
Yes, negative values remain valid throughout aggregation and charting. Signed logarithms preserve their direction. Diverging palettes work well when zero has meaningful importance.
8. Why do some cells remain blank?
Blank cells may represent missing dates or excluded weekdays. Interpolation also leaves leading and trailing gaps unchanged. Hover details identify dates for visible cells.
9. How should I compare several heatmaps?
Use identical ranges, aggregation methods, filters, and color scales. Record every transformation beside the exported chart. Consistent preprocessing makes later comparisons more trustworthy and explainable.