Histogram Generator
Divides a numeric column's range into equal-width bins and counts how many values fall into each, rendering the result as a bar chart — the standard way to visualize a single variable's distribution shape (skewed, symmetric, multi-modal, etc.).
Bin count is adjustable because it genuinely changes what the histogram reveals: too few bins can hide real structure (like two separate peaks blending into one), while too many can make the shape look noisy and hard to read.
Bin count is an analytical choice, not a setting
A histogram does not show your data — it shows your data grouped, and the grouping changes what you see. Too few bins smooths real structure away: two distinct peaks merge into one broad hump, and you conclude the distribution is unimodal when it is not. Too many bins turns sampling noise into apparent structure, producing a spiky chart where every wobble looks meaningful.
This is why the bin count is adjustable and why you should move it. If a feature survives across a range of bin widths it is probably real; if it appears at 40 bins and vanishes at 20, it was noise.
Starting points that are defensible
Two published rules give a reasonable first bin count, and both are starting points rather than answers.
Sturges' rule, ⌈log₂n⌉ + 1, is simple and works for roughly normal data of moderate size. It under-bins large datasets badly.
The Freedman–Diaconis rule sets bin width to 2·IQR/∛n. Because it uses the interquartile range it adapts to spread and resists outliers, which makes it the better default for skewed real-world data.
What the shape is telling you
Right-skewed — a peak on the left with a tail to the right — is the most common shape in practice: durations, incomes, file sizes, counts. The mean sits above the median, so report the median.
Two peaks almost always mean two populations mixed together. Find the variable that separates them and plot each group; the combined histogram is describing something that does not exist as a single thing.
A spike at one value, often zero, usually means a sentinel: a default, a "not measured", or a floor effect. Treat it as a data-quality finding rather than a feature of the distribution.
Gaps in an otherwise continuous range suggest rounding, a measurement limit, or values that were filtered out upstream.