Skip to content

One-line dashboards and the command line

ts.auto: a dashboard from the data alone

import tabsight as ts

orders = ts.load("examples/superstore_sample.csv")
dash = ts.auto(orders, "Superstore")
print(dash)
for line in dash.decisions:
    print(line)
Output
<Dashboard 'Superstore': 10 tiles, filters ['Order_Date', 'Category', 'Region', 'State']>
measures: Sales (currency, sum), Profit (currency, sum), Quantity (count, sum), Discount (percent, mean)
date: Order_Date (KPIs compare with the previous year)
map: State holds places
filters: Order_Date, Category, Region, State

# not run
dash.to_html("superstore.html")

ts.auto reads the column profile and follows fixed rules, so the same data always gives the same dashboard:

  1. KPI cards for up to four measures, money first. When the dates cover more than a year, each card compares with the previous year.
  2. A line over time of the main measure, by day, month, quarter or year to suit the time span.
  3. A map, when a column holds countries, regions or cities.
  4. Bars (or a donut, for a few parts of a total) of the main measure by the categories with a handful of values. A hierarchy's top level comes first (Region before State).
  5. Filters on the date and the smallest categories, and an insights panel.

dash.decisions lists what was chosen and why. Every tile is tried before it is added, and the dashboard is checked against the size limit. If the dashboard's own steps aren't enough, it drops filters and then tiles until it fits. A table with no measures still works: tiles count rows (as Records), and distinct IDs get a KPI.

The result is an ordinary Dashboard, so you can keep going:

dash.add("scatter", x="Sales", y="Profit", color="Category", w=12)
spec = (
    dash.to_dict()
)  # or dash.save("superstore.json"); reuse with Dashboard.load(...).bind(new_df)
print(len(dash.tiles), "tiles;", len(spec["tiles"]), "in the saved spec")
Output
11 tiles; 11 in the saved spec

Options: max_tiles= (default 10), insights=False, and theme=, locale=, currency= as for any dashboard.

The command line

Installing tabsight adds a tabsight command (or run python -m tabsight):

tabsight auto sales.csv                      # writes sales.html next to it
tabsight auto sales.csv -o report.html --theme dark --save sales.json
tabsight build sales.json november.csv -o november.html   # the same dashboard, new data
tabsight chart sales.csv bar --x Region --y Sales -o bar.png   # .html, .svg or .png
tabsight serve sales.csv                     # server mode at http://127.0.0.1:8050/
tabsight profile sales.csv                   # what each column is, and why
tabsight check sales.csv                     # data-quality report (exit code 1 on errors)
tabsight clean sales.csv -o clean.csv        # a cleaned copy; the input is never changed
tabsight recommend sales.csv --x Region --y Sales

Files are read with ts.load: CSV, TSV, Excel (--sheet), Parquet and JSON. Output goes where -o says, or next to the input. tabsight refuses to write over the input file. Errors print one line and exit with code 2.

tabsight serve runs a dashboard in server mode with Python's own web server, on this computer only (127.0.0.1). The data stays in Python, so large files and every chart kind stay exact. It only answers requests addressed to 127.0.0.1 or localhost, so a web page elsewhere can't reach it by pointing a domain at your computer. --host 0.0.0.0 lets other computers connect and prints a warning, since anyone who can reach the address can see the data. Add the name they will use with --allow-host my-laptop.local.

PNG pictures

# not run
fig = ts.chart(orders, "bar", x="Region", y="Sales")
fig.to_png("bar.png")  # sharp on high-density screens (scale=2)
fig.to_png("small.png", width=600, scale=1)

to_png draws the same picture as fig.to_svg() and turns it into pixels with resvg. It needs neither a browser nor Node.js: pip install "tabsight[png]" installs dukpy and resvg-py.