More chart types¶

Beyond bars and lines, tabsight draws relationships, distributions, parts of a whole, flows and single values. Every one follows the same rules as the everyday kinds: numbers are aggregated exactly and formatted by their unit, labels are readable, and fig.explain() says what was decided.

import pandas as pd

import tabsight as ts

orders = ts.load("examples/superstore_sample.csv")

Relationships: scatter and bubble¶

With two number columns, a scatter draws one dot per row. The axes fit the data rather than start at zero, because position matters here, not length:

fig = ts.chart(orders, "scatter", x="Sales", y="Profit", color="Category")
print(fig.title, "|", [s["name"] for s in fig.option["series"]])

Output

Profit vs Sales by Category | ['Furniture', 'Office Supplies', 'Technology']

With a category on x and two measures, it draws one dot per category instead, aggregated exactly. A third measure makes it a bubble chart, where circle area grows with the value:

fig = ts.chart(orders, "bubble", x="Sub_Category", y=["Sales", "Profit", "Quantity"])
print(fig.title)

Output

Profit vs Sales by Sub Category

More than 5,000 points are sampled evenly, and the chart says so.

Distributions: histogram and box plot¶

fig = ts.chart(orders, "histogram", x="Sales")
print(fig.decisions[1])

Output

bar width from the spread of the middle half of the data (Freedman-Diaconis), rounded: 21 bars

The bar width is chosen from the spread of the middle half of the data and rounded, so the edges are numbers people use. bins=20 asks for about 20 bars and bin_width=100 fixes the width.

A box plot shows the middle half of the values as a box with the median inside. Whiskers reach the furthest values within 1.5 times the box height, and dots beyond them are outliers. Hover a box to see all five numbers.

fig = ts.chart(orders, "boxplot", x="Category", y="Profit")
low, q1, median, q3, high = fig.option["series"][0]["data"][0]["value"]
print(f"median {median:.2f}, middle half {q1:.2f} to {q3:.2f}")

Output

median 9.88, middle half -28.27 to 56.18

Lines over a number¶

When x is a number rather than a date or a category, it runs along a real number axis, so points sit at their true distances. This is how a sine wave becomes a curve rather than 126 category labels:

import numpy as np

wave = pd.DataFrame({"angle": np.arange(0, 6.3, 0.05)})
wave["sine"] = np.sin(wave["angle"])
fig = ts.chart(wave, "line", x="angle", y="sine")
print(
    fig.title,
    "|",
    fig.option["xAxis"]["type"],
    "axis,",
    len(fig.option["series"][0]["data"]),
    "points",
)

Output

sine by angle | value axis, 126 points

Lines with more than 2,000 points are thinned with LTTB, which keeps peaks and dips, and the chart says how many points it shows.

Parts of a whole: treemap, sunburst, waterfall, funnel¶

A treemap or sunburst takes one or more levels in x. Both need a total that adds up, so an average is refused with a reason:

fig = ts.chart(orders, "treemap", x=["Category", "Sub_Category"], y="Sales")
print([(d["name"], d["note"]) for d in fig.option["series"][0]["data"]])
try:
    ts.chart(orders, "treemap", x="Category", y="Discount")
except ts.warnings.InvalidSpecError as error:
    print(error)

Output

[('Technology', '40.4% of the total'), ('Office Supplies', '31.1% of the total'), ('Furniture', '28.6% of the total')]
a treemap shows parts of a whole, but Discount is averaged, which doesn't add up; use bars

A waterfall walks from one total to another. Each bar starts where the last ended, blue adds and orange takes away, and the last bar is the total from zero. Steps keep the order they appear in the data:

bridge = pd.DataFrame(
    {
        "Step": ["Revenue", "Costs", "Marketing", "Grants", "Tax"],
        "Amount": [1000, -420, -180, 60, -45],
    }
)
fig = ts.chart(bridge, "waterfall", x="Step", y="Amount")
print([(d["name"], d["value"][4]) for d in fig.option["series"][0]["data"]])

Output

[('Revenue', 1000.0), ('Costs', 580.0), ('Marketing', 400.0), ('Grants', 460.0), ('Tax', 415.0), ('Total', 415.0)]

A funnel's tooltip shows how much of the previous and of the first stage each one keeps:

stages = pd.DataFrame({"Stage": ["Visited", "Signed up", "Paid"], "Users": [12000, 3400, 420]})
fig = ts.chart(stages, "funnel", x="Stage", y="Users")
print([d.get("note") for d in fig.option["series"][0]["data"]])

Output

[None, '28% of the previous stage; 28% of the first', '12% of the previous stage; 4% of the first']

Ranking: Pareto¶

Bars largest first, with a line that adds them up to 100% and a dashed guide at 80%:

fig = ts.chart(orders, "pareto", x="Sub_Category", y="Sales")
print(fig.decisions[-2])

Output

bars largest first; the line adds them up: 10 of 13 make 80% of the total

Flows and grids: Sankey and heatmaps¶

A Sankey diagram draws flows from a source column to a target column, with band width showing the value. A heatmap puts x across, color down and colours each cell by the measure. Negative values get a diverging scale. A calendar heatmap gives each day a square:

fig = ts.chart(orders, "sankey", source="Segment", target="Category", y="Sales")
print(len(fig.option["series"][0]["links"]), "flows")
fig = ts.chart(orders, "heatmap", x="Order_Date", color="Category", y="Profit", freq="year")
print(fig.decisions[-2])

Output

9 flows
columns are Order_Date, rows are Category; darker means bigger

Against a goal: bullet and gauge; two measures: combo¶

reps = pd.DataFrame({"Rep": ["Ann", "Bo"], "Sales": [120.0, 80.0], "Target": [100.0, 100.0]})
fig = ts.chart(reps, "bullet", x="Rep", y="Sales", goal="Target")
print([d["note"] for d in fig.option["series"][1]["data"]])
fig = ts.chart(orders, "combo", x="Region", y=["Sales", "Quantity"])
print(next(d for d in fig.decisions if "two axes" in d))

Output

['goal $100.00 (120% of it)', 'goal $100.00 (80% of it)']
two axes because the units differ (count, currency): bars on the left, lines on the right

A goal can be one number for everything or a column with one per row. A combo draws the first measure as bars and the others as lines, on a second axis when the units differ. One axis would flatten the smaller one.