Group

What is group

Group turns each row into a point with up to four dimensions: Name, XAxis, YAxis, and ZAxis. This lets vizb chart the data meaningfully instead of as a flat list.

It always works in two stages:

1. Build a label for the row.
   • For CSV/JSON, you pick columns with --group. Vizb joins their per-row values using the same separators declared in --group-pattern.
   • For benchmarks, the benchmark name is the label already (e.g. BenchmarkSort/1024/QuickSort).
2. Split the label into dimensions with --group-pattern (separator-based) or --group-regex (named captures). This stage is identical for both inputs.

The two flows differ only in stage 1 — where the label comes from. The same pattern and regex syntax serves both.

Why group

Group creates multi-dimensional charts. Without it, every row is an isolated bar or point. With it, vizb can:

• Separate charts by logical category (e.g. one chart per algorithm or region).
• Populate axes with meaningful values (e.g. input size on the X-axis).
• Create series for variants (e.g. different data distributions on the Y-axis).

Without group	With group
Single flat chart with all rows	Multiple charts grouped by name
No axis structure	X-axis and Y-axis with meaningful categories
Hard to compare across dimensions	Easy side-by-side comparison

The Dimensions

Group extracts up to four dimensions from each label:

Dimension	Role	Description
Name (name / n)	Groups rows into separate charts	Rows sharing the same name appear in the same chart
XAxis (xAxis / x)	Sets the X-axis categories	Each unique value becomes a tick on the X-axis
YAxis (yAxis / y)	Creates series / variants	Each unique value becomes a separate series or bar group
ZAxis (zAxis / z)	Adds a depth axis → 3D chart	Each unique value becomes a stacked depth layer in a 3D chart

Note

At least one axis (x or y) is required for group to take effect. The name dimension is optional. When omitted, all rows go into a single chart. The z dimension requires both x and y to be present.

Group CSV/JSON data

CSV and JSON rows have no benchmark name. You tell vizb which columns carry the structure. The --group / -g flag selects columns. The --group-pattern / -p flag assigns each column to a dimension. By default each column maps to one dimension (-p y,x). Use bracket slots [...] when a single column’s cell value encodes multiple dimensions (dates, slash paths, underscore IDs, etc.).

Note

Not sure whether to group or to use --select? See Group vs Select for the two models, auto-group / auto-value inference, and a decision matrix. Solo --select value, mixed, and multi-stat modes are covered in Select.

Matching separators

For tabular data, -g and -p must use the same separators between column slots. If they disagree, vizb exits before parsing:

--group "<g>" and --group-pattern "<p>" separators do not match (expected "<from g>", got "<from p>")

The expected side is derived from how -g declares column boundaries. The got side is what -p declares. Fix whichever flag is wrong so both use the same separator sequence.

--group style	Matching -p	Row label (example)	Mismatch would report
Comma list (-g region,product,category)	Commas (-p x,y,z)	Asia,Widget,Electronics	-p x/y/z → expected ", ,", got "/ /"
Quoted string (-g "region product")	Spaces (-p "y x")	UAE Gadget	—
Single value (-g "region/product/category")	Same separators (-p x/y/z)	EU/Widget/Sports	-p x,y/z → expected "/ /", got ", /"
Structured (-g region,product/category)	Same shape (-p y,x/z)	Asia,Widget/Electronics	—

Comma-separated -g (the usual pflag form) requires comma-separated -p. For example, use -g region,product,category with -p x,y,z, not x/y/z. Every separator in -g must appear in -p at the same positions.

When -g and -p define a different number of slots (common with bracket patterns), vizb reports a separate dimension-count error:

--group and --group-pattern dimension count do not match: N column(s) in "<g>", M slot(s) in --group-pattern "<p>"

Take this sales.csv:

region	product	category	sales	latency
Asia	Widget	Electronics	100	12
UAE	Gadget	Home	80	15
EU	Gizmo	Sports	60	11

vizb sales.csv --group region,product -p y,x -o sales.html

For the first row, vizb joins the selected columns into the label Asia,Widget, then the pattern y,x splits it: y → Asia (YAxis), x → Widget (XAxis).

Space-separated columns — pass one quoted -g value and use spaces in -p:

vizb sales.csv -g "region product" -p "y x" -o sales.html

Structured grouping — mix separators inside -g (commas between pflag values, / or _ inside a value) to declare a hierarchy. -p must mirror the same shape:

vizb sales.csv -g region,product/category -p y,x/z -o sales.html
vizb sales.csv -g "region/product/category" -p x/y/z -o sales.html

Bracket slots — split a column’s cell value

When one column holds an encoded string (not separate columns), wrap the inner pattern in [...]. Each top-level slot lines up 1:1 with a -g column:

Slot form	Meaning
x, y, n, z	Whole cell value → that dimension
x{label}	Whole cell → dimension, axis titled label (overrides column name)
[inner]	Split the cell value with inner separators → multiple dimensions
n{year}-y{months} inside [...]	Split value and set per-dimension axis titles

Top-level separators between slots still match -g (commas for -g a,b, spaces for -g "a b"). Benchmarks do not use brackets. Keep flat -p n/x/y for bench.txt.

Tip

Use single quotes around -p when it contains [...] bracket slots:

vizb sales.csv -g date,category -p '[n{year}-y{months}-x{dates}],z' -o sales.html

Bracket slots can attach {label} to each inner dimension — see Axis labels.

Date + category — split 2022-2-30 into year/month/day, keep category whole:

vizb sales.csv -g date,category -p "[x-y-n],z" -o sales.html

date	category	Pattern	Result
2022-2-30	Widget	[x-y-n] splits on -, z ← category	x=2022, y=2, n=30, z=Widget

More bracket-slot examples (ISO dates, slash paths, underscore IDs, mixed separators)

ISO dates (YYYY-MM-DD) — use a bracket slot when the date is one column. {year}, {months}, and {dates} only set axis titles. Splitting is driven by - between dimension tokens. The same prefix/middle/trailing skip rules from Separators apply inside brackets too — e.g. [n{year}-x{months}-] drops the day, [n{year}--x{dates}] skips the month.

vizb sales.csv -g date,category -p '[n{year}-y{months}-x{dates}],z' -o sales.html

date	Pattern	Result
2024-01-01	[n{year}-y{months}-x{dates}]	n=2024, y=01, x=01

Slash-encoded benchmark column — same layout as Go bench names, stored in CSV:

vizb results.csv -g benchmark -p "[n/x/y]" -o out.html

benchmark	Pattern	Result
Sort/1024/QuickSort	[n/x/y]	n=Sort, x=1024, y=QuickSort

Underscore-delimited ID:

vizb runs.csv -g id -p "[n_y_x]" -o out.html

id	Result
lib_func_case	n=lib, y=func, x=case

Space-separated label in one column:

vizb data.csv -g "label region" -p "[x n y],z" -o out.html

label	region	Result
alpha beta gamma	EU	x=alpha, n=beta, y=gamma, z=EU

Prefix skip inside a bracket (leading / drops the first segment of the cell value):

vizb data.csv -g name -p "[/n/y]" -o out.html

name	Result
BenchmarkTest/JSON/Marshal	n=JSON, y=Marshal

Mix value-split with whole columns — date parts + region on Z:

vizb sales.csv -g date,region -p "[n-y-x],z" -o sales.html

date	region	Result
2022-2-30	Asia	n=2022, y=2, x=30, z=Asia

Note

Bracket splitting is literal. Vizb does not parse dates or semver semantically. Values must match the separator layout in -p. For irregular strings without stable separators (e.g. Hashing64MD5), use --group-regex instead.

The columns you don’t select decide the chart values:

• Unselected numeric columns each become their own chart. Here sales and latency each become a separate chart (one Stat per column).
• Non-numeric unselected columns are ignored. A notes text column would simply be dropped.
• No --group at all → dimensions stay empty. No rows are merged. Each row becomes one point, and every numeric column renders as its own chart. This is useful for a quick chart of a wide table.

Note

Each --group column must match a header exactly. A missing column is fatal: group column 'X' not found; available: [...].

Aggregation

When group is active (--group or auto-group), multiple rows can share the same (name, x, y, z) key. Vizb sums their values into a single point. A 200k-row export collapses into a few thousand chart points. You’ll see it in the CLI output when duplicates were merged:

🧮 Aggregating 50000 rows by columns: region, product, month (name: region, x: product, y: month)...
✅ Aggregated into 1200 grouped data points

If every row already has a unique group key (e.g. auto-group picked a high-cardinality ID column), vizb still runs the sum step but nothing collapses — the closing line instead reports ✅ N grouped rows — all unique (no duplicates to sum).

Caution

Aggregation is summing, and applies to CSV/JSON only. Benchmark parsers are never aggregated. Repeated runs share a key but the UI averages them instead of summing.

See Tabular Data (CSV & JSON) for the full parser rules.

Group benchmarks

Benchmark names already encode structure. A name like BenchmarkSort/1024/QuickSort carries the function, the input size, and the variant. There are no columns to pick. So --group-pattern / --group-regex do all the extraction. Set axis titles with {label} suffixes in -p (see Axis labels).

go test -bench . | vizb -p n/x/y -o output.html

Benchmark Name	Name	XAxis	YAxis
BenchmarkSort/1024/QuickSort	Sort	1024	QuickSort
BenchmarkSort/2048/MergeSort	Sort	2048	MergeSort

Every metric the benchmark reports (time, memory, allocations, …) becomes its own chart automatically. You only group the name.

Pattern Syntax

The --group-pattern flag splits the label using the separators you declare between dimension tokens. It assigns each segment to a dimension. It applies to both flows — the benchmark name or the CSV/JSON label built from --group.

Separators

Any single character works between dimension tokens. Vizb reads them from your -p string:

Separator	Typical use
/	Go/Rust/JS benchmark names (n/x/y) and slash-structured -g (region/product)
_	Underscore-delimited bench names (n_y_x) and structured -g specs
,	Comma-separated --group lists (y,x with -g region,product)
(space)	Quoted --group specs ("y x" with -g "region product")
#, ., etc.	Any single character works when -g and -p agree (e.g. -g "a#b#c" -p "x#y#z")

Three skip patterns drop segments without assigning them to a dimension:

Pattern	Example	Effect
Leading separator	/n/y	First value segment is unused (prefix skip)
Consecutive separators	n--x or x//y	Middle segment is unused
Trailing separator	n-x-	Last value segment is unused

Each repeated separator character adds one more split point and one empty dimension slot. A single trailing - after the last token is enough to drop the final segment. You do not need ---.

Works in flat benchmark patterns and inside bracket slots (e.g. [n{year}-x{month}-] on a CSV date column).

Components

Component	Shorthand	Meaning
name	n	Chart group name
xAxis	x	X-axis values
yAxis	y	Y-axis series
zAxis	z	Z-axis depth layer (3D, requires x + y)

The default pattern is x — a single segment mapped to the X-axis.

Examples

Slash-separated n/x/y:

vizb bench.txt --group-pattern "n/x/y" -o output.html

Label	Name	XAxis	YAxis
BenchmarkSort/1024/QuickSort	Sort	1024	QuickSort
BenchmarkSort/2048/MergeSort	Sort	2048	MergeSort
BenchmarkSort/4096/QuickSort	Sort	4096	QuickSort

Underscore-separated n_y_x:

vizb bench.txt --group-pattern "n_y_x" -o output.html

Label	Name	YAxis	XAxis
BenchmarkHash_SHA256_1KB	Hash	SHA256	1KB
BenchmarkHash_MD5_4KB	Hash	MD5	4KB

2D with n/y (no X-axis grouping):

vizb bench.txt --group-pattern "n/y" -o output.html

Label	Name	YAxis
BenchmarkJSON/Marshal	JSON	Marshal
BenchmarkJSON/Unmarshal	JSON	Unmarshal

Prefix skip with /n/y (leading slash discards the first segment):

vizb bench.txt --group-pattern "/n/y" -o output.html

Label	Name	YAxis
BenchmarkTest/JSON/Marshal	JSON	Marshal
BenchmarkTest/XML/Parse	XML	Parse

Middle skip with n//y (consecutive separators discard the middle segment):

vizb bench.txt --group-pattern "n//y" -o output.html

Label	Name	YAxis
Sort/1024/QuickSort	Sort	QuickSort

Trailing skip with n-x- (trailing separator discards the last segment):

vizb bench.txt --group-pattern "n-x-" -o output.html

Label	Name	XAxis
2024-01-01	2024	01

4D with n/x/y/z (renders a 3D chart):

vizb bench.txt --group-pattern "n/x/y/z" -o output.html

Label	Name	XAxis	YAxis	ZAxis
BenchmarkSort/1024/QuickSort/warm	Sort	1024	QuickSort	warm
BenchmarkSort/2048/MergeSort/cold	Sort	2048	MergeSort	cold

Note

A pattern must contain x or y (pattern must contain xAxis (x) or yAxis (y)). A z segment requires both x and y. Vizb rejects z otherwise with zAxis (z) requires both xAxis (x) and yAxis (y). See 3D Charts.

Regex Syntax

The --group-regex flag uses Go-style named capture groups. It runs on the label — a benchmark name, or the CSV/JSON label built from --group. Use it when the label does not follow a simple separator pattern. When set, regex takes precedence over --group-pattern.

Named Capture Groups

Group Name	Shorthand	Dimension
(?<name>...)	(?<n>...)	Chart group name
(?<xAxis>...)	(?<x>...)	X-axis values
(?<yAxis>...)	(?<y>...)	Y-axis series
(?<zAxis>...)	(?<z>...)	Z-axis depth layer (3D, requires x + y)

Examples

Label	Regex	Extracted
BenchmarkHashing64MD5	Hashing64(?<x>.*)	X: MD5
BenchmarkJSONByMarshal	(?<x>.*)By(?<y>.*)	X: JSON, Y: Marshal
BenchmarkDecode/text=digits/level=speed	(?<n>.*)/text=(?<x>.*)/level=(?<y>.*)	N: Decode, X: digits, Y: speed

vizb bench.txt --group-regex "(?<n>.*)/text=(?<x>.*)/level=(?<y>.*)" -o output.html

Note

The regex must capture x or y (regex '...' does not contain x (xAxis) or y (yAxis)). A captured z still requires both x and y.

Axis labels

Set chart axis titles in --group-pattern / -p by appending {title} immediately after a dimension token (no space):

Shorthand	Syntax	Chart title
n	n{Algorithm}	Algorithm
x	x{Product}	Product
y	y{Region}	Region
z	z{Category}	Category

Works on flat patterns, bracket slots, and benchmarks. The same syntax is used everywhere -p is used.

# CSV — titles independent of column names
vizb sales.csv -g region,product -p 'y{Region},x{Product}' -o sales.html

# Bracket slot — one column, multiple titled dimensions
vizb sales.csv -g date,category -p '[n{year}-y{months}-x{dates}],z{Category}' -o sales.html

# Benchmark — no --group required
vizb bench.txt -p 'n{Algorithm}/x{Size}/y{Variant}' -o out.html

Omit {label} on CSV/JSON and the matching --group column name becomes the axis title. When {label} is present, it always overrides that default.

Note

Titles are emitted as axisLabels in the JSON output and rendered on chart axes. The name title is preserved for the UI axis-swap feature.

Choosing a Method

Use --group-pattern when the label follows a consistent separator format — standard Go benchmarks (BenchmarkFunc/size/variant) or CSV/JSON columns that map cleanly to dimensions. Use --group-regex when the format is non-standard or you need fine-grained extraction, e.g. --group-regex "Hash(?<n>[A-Za-z]+)(?<x>\d+)" to pull a name and size out of HashingMD5Test1024 without a stable separator.

Tip

Either method splits the label the same way. For CSV/JSON, remember you still need --group to pick the columns first. Benchmarks need no --group to chart, since the name is already the label.