pavement.summary

One call, an at-a-glance picture of a dataframe, Series, or sequence — a tally beside a distribution for every column. Hover any strip for its share, value, and count, or grab a row's grip handle to reorder.

tally: distinct   duplicate   missing

A dataframe

One row per column, under a top row for the frame as a whole: its 507 rows, and a tally that treats each whole row as the entity. Numeric columns get a pavement spark whose resolution adapts to the column — a frequency rug for 24 or fewer values, or for 16 or fewer distinct values (so the discrete rating shows its five levels and their counts as scaled lines rather than equal-mass bins), then 4, 8, or 16 equal-mass bins for larger continuous columns. Dates work too: signup_date is laid out on a time axis (hover for the dates). Durations too: session_duration shows timedeltas on a duration axis (e.g. 1d 02:00). Categorical columns get a proportion strip, and legacy_field shows what an almost-all-missing column looks like.

import pavement

pavement.summary(df)        # dict of lists, pandas/polars DataFrame, or any mapping

A single Series

Pass one sequence and you get a single row. A bare list has no name, so the label shows the value count instead. Numeric values give a spark:

pavement.summary(daily_signups)   # a bare list — label shows value count

…and categorical values give a proportion strip:

pavement.summary(survey)          # a list of strings

Drag to reorder

Every summary on this page is drag-to-reorder by default: a small grip handle sits at the left of each column name — grab it and drop the row where you want, and the others slide to make room, handy for sitting two columns next to each other to compare them. Only the grip drags, so the rest of the row keeps its normal cursor and stays text-selectable, and the top summary row stays pinned. It's the page's only JavaScript, scoped per table and purely visual (the new order isn't read back into Python). The grip only appears where that script runs, so notebooks and static exports simply show the plain table — which is why it's harmless to leave on. Pass draggable=False for a guaranteed script-free, static fragment.

pavement.summary(df)                   # rows drag to reorder by default
pavement.summary(df, draggable=False)  # a script-free, static table

A pandas DataFrame

The call is the same — pavement.summary(df) — for a real DataFrame. Numeric columns become sparks, the species and the boolean flag become proportion strips.

import pandas as pd
import pavement

df = pd.DataFrame({...})
pavement.summary(df)

Numpy datetime64 and timedelta64 arrays

Raw numpy arrays work too — no pandas or polars needed. datetime64 columns land on a time axis; timedelta64 columns show durations (e.g. 00:01 for one minute).

import numpy as np
import pavement

data = {
    "event_time": np.array([...], dtype="datetime64[s]"),
    "response_delay": np.array([...], dtype="timedelta64[s]"),
}
pavement.summary(data)

A frequency rug

A rug (bins=None) draws one line per distinct value. By default every line spans the full height, so a value that occurs once looks just like one that occurs a hundred times. With proportional_representation=True each line is instead scaled to how often that value occurs: the most common value reaches the full box and the rest reach proportionally less, so the shape of the distribution reads straight off the line lengths — without binning the data away.

The data below is a five-level rating with 97 fours (the most common), 94 threes, 60 fives, 40 twos, and 12 ones. Hover any line for its value, percentile, and count — or the space between two lines for that gap's range (an easier target than a thin line), just like the boxes of a binned pavement.

A floor (min_representation, default 5% of full) keeps a rare value's line visible rather than letting it collapse to a point; the lines sit on a shared baseline (the bottom, for a horizontal rug) and grow upward, so they read like little bars.

This is what summary reaches for automatically: a numeric column with 16 or fewer distinct values (like rating in the dataframe above) is drawn as a frequency rug rather than smeared into equal-mass bins, so its handful of levels and their counts read straight off the strip.

Expressive box edges

Each bin draws its long top and bottom edges only over itself, and only when one or more values fall strictly inside it. So the box closes around the bins where values are spread out, and opens into a gap wherever the bin's mass sits on a value line instead — letting spread and clumping read straight off the outline. All three below are 8-bin pavements; hover any bin or line for its value range, percentile, and count.

Forcing a complete box

The gapped outline is the default, but show_box=True forces the complete box — the two long edges unbroken across the whole value range — for a rug or a pavement alike, for when a solid outline is wanted regardless of where the values fall. The four below are the same 150 iris sepal_width measurements, drawn each way.

In a notebook

Everything above is just str(pavement.summary(...)) dropped into this page. In Jupyter you skip the str() — the last line of a cell, pavement.summary(df), renders the table inline on its own. And for pandas or polars, import pavement.pandas (or import pavement.polars) adds a .pave accessor: df.pave(), df.pave.spark("age"), df.pave.tally("plan") — or pavement.pandas.enable_repr() to make the summary every DataFrame's default preview. See pandas_polars_demo.html for a full walkthrough.