Explore a DataFrame and Chart It

In this tutorial you show a pandas DataFrame in a notebook and work with it where it is. You read its columns, sort and filter it, page through a million rows, open it in a tab, and chart it. You write none of that code yourself. At the end, the chart becomes code you keep.

The notebook is tutorials/04-data-analysis.ipynb from zasper-tutorials. It loads the Palmer penguins with pandas: 344 penguins of three species, from three islands. The repository’s README sets up its kernel, Python (zasper-tutorials), which has pandas and plotly in it. This page assumes Zasper is installed and open on that folder. A DataFrame of your own works the same way, on any Python kernel.

A DataFrame Is A Grid

Open the notebook and run its first code cell, under Load. It reads data/penguins.csv into penguins.

Now show the whole table. Add a cell below penguins.head(), and type the name of the DataFrame on a line of its own:

penguins

Run it:

Action	macOS	Windows / Linux
Run Cell and Select Next	`⇧⏎`	`Shift+Enter`
Run Cell	`⌃⏎`	`Ctrl+Enter`

Instead of a long printed table, you get a grid. Its top line says what it is: a DataFrame, with 344 rows × 8 columns.

It shows ten rows at a time. Next page and Previous page, at the bottom, page through them. The drop-down beside them shows 25 or 100 rows at once.

The penguins DataFrame as a grid: its top line reads DataFrame, 344 rows × 8 columns, each column's header shows its type and a small chart, and the foot reads Rows 1–10 of 344, ringed and labelled Paged, ten rows at a time. — Ten rows a page. The foot says which ten.

The grid needs the kernel that printed the frame. After a restart, the output is pandas’ own table again; run the cell to get the grid back. The saved notebook keeps pandas’ table, so it reads the same in JupyterLab and on GitHub.

Read A Column At A Glance

Each column’s header shows its type, a small chart of its values, and how much is missing. You can see them in the grid above. Body mass has a few gaps: body_mass_g is a float64, and its header says 1% missing. Hover the small chart for the numbers behind it.

Click a header for its menu. At the top are the actions: sort, filter, chart, hide and move the column. At the bottom is a Summary of the column: Rows, Missing, Distinct, Min, Max, Mean and Std dev. For body_mass_g, Missing reads 2 (0.6%): two of the 344 penguins have no body mass. Min and Max give the range: from 2,700 to 6,300 grams.

The menu of the body_mass_g header, over the grid. Its actions are Sort ascending, Sort descending, Filter this column…, Chart this column, Hide column, Move left and Move right. Below them the Summary heading is ringed, labelled A summary of body_mass_g, and lists Rows 344, Missing 2 (0.6%), Distinct 94, Min 2,700, Max 6,300, Mean 4,201.75 and Std dev 801.955. — The **Summary** at the bottom of a column’s menu.

A text column, such as species, lists its commonest values instead, under Summary · commonest.

Sort And Filter

In the same menu, choose Sort descending. The heaviest penguins come first. The sorting happens in the kernel, not in your browser.

To keep only some rows, add a filter. Press Filter rows, the funnel in the grid’s bar. Pick a column, a comparison and a value, and press Apply. Here, the penguins on Biscoe island: island, =, Biscoe. A text column also offers contains, does not contain and starts with, which ignore case.

The filter shows above the grid, and the count drops to 168. The sort still holds, so the heaviest penguins on Biscoe come first. Remove the filter any time with its cross.

The penguins grid filtered: a chip reading island = Biscoe with its cross ringed, the top line reading 168 of 344 rows × 8 columns, Gentoo penguins sorted by body mass from 6300 down, and the foot Rows 1–10 of 168 ringed, labelled 168 penguins on Biscoe. — The filter is a chip above the rows. Its cross removes it.

A Million Rows, Still Fast

That was 344 rows. Here’s a million. Add a cell below the grid that makes a DataFrame with a million numbers and their squares:

big = pd.DataFrame({"n": range(1_000_000)})
big["square"] = big.n ** 2
big

Run it. The grid appears straight away, because only the ten rows on screen come to the browser.

Sorting a million rows is just as quick. Click the square header and choose Sort descending. The kernel sorts, and sends back one page.

The big DataFrame as a grid: 1,000,000 rows × 2 columns, sorted by square, with 999998000001 in the first row and the foot reading Rows 1–10 of 1,000,000. The square header is ringed, labelled Largest first, sorted in the kernel. — One page of a million rows, sorted in the kernel.

Open It In A Tab

For a closer look, press Open in a tab, the last button in the grid’s bar. The table opens in a tab of its own, named Out[n] after the cell’s count, with the filters, sort and columns it had. There it scrolls, all the way through. It reads rows as they come into view, so the far end of a million rows is as quick to reach as the start.

The Variables panel, under the editor, lists everything the kernel holds, with each one’s type and size. Open it from its tab in the dock, or with Variables in the command palette. Any table there opens the same way: click its row.

The Variables tab of the dock under the editor. It lists big, DataFrame · 1,000,000 × 2 · n, square, and penguins, DataFrame · 344 × 8, with its column names. The penguins line is ringed and labelled Every variable in the kernel, with its type and size. — **Variables** lists what the kernel holds, and opens any table in a tab.

Chart It, And Keep The Chart As Code

Back in the notebook, any grid can become a chart. Press Show as a chart in the penguins grid’s bar. The chart uses the same rows as the grid, filter included. The kernel counts them, and the grid’s sort does not apply.

The buttons at the left of the chart pick its kind: Histogram, Bar, Line, Scatter or Box. Pick Scatter, choose the columns for X and Y, and set Colour to species. With the Biscoe filter on, two species are left: Adelie and Gentoo.

When the chart is right, Insert as code writes the Plotly code for it into a new cell below, with the filter. It does not run it. Run that cell, and the notebook draws the chart again by itself.

The penguins output shown as a scatter chart: its top line reads DataFrame, 168 of 344 rows × 8 columns, with the island = Biscoe chip below it, then the chart kinds and the X, Y and Colour choices. bill_depth_mm is plotted against bill_length_mm, coloured by species, Adelie in blue and Gentoo in orange. The foot reads bill_depth_mm against bill_length_mm, over 168 of 344 rows, and the Insert as code button at its right is ringed and labelled Insert as code. — **Insert as code** turns the chart on screen into a cell of Plotly code.

The chart itself is never saved in the notebook; the code is. Insert as code needs a name to write against, so it works when the cell ends in a bare name such as penguins. For an expression such as penguins.head(), the button is disabled and says why.

Recap

A displayed DataFrame is a grid
Headers show type, spread, and gaps
Sort and filter, computed in the kernel
A million rows page instantly
Open tables and variables in a tab
Chart it, then insert the chart as code

Next: Exploring Data has the rest of the grid and the charts for reference, then Keep a Run Going When You Close the Browser. To get a DataFrame out of a database instead, try Query Data with SQL.