17 Change Over Time
Line charts and their variants β showing change over a sequence.
(ns plotje-book.change-over-time
(:require
;; Tablecloth -- dataset manipulation
[tablecloth.api :as tc]
;; Kindly -- notebook rendering protocol
[scicloj.kindly.v4.kind :as kind]
;; Plotje -- composable plotting
[scicloj.plotje.api :as pj]))Line
Connected line through data points.
(def wave {:x (range 30)
:y (map #(Math/sin (* % 0.3)) (range 30))})(-> wave
(pj/lay-line :x :y))Grouped Lines
Color separates multiple series. Real datasets often start in wide form β each series in its own column. One way is to reshape to long form β one row per observation, with the series label in a column β with tc/pivot->longer, then map the label column to :color. See Datasets for more on the wide-to-long reshape.
(def waves-wide
(tc/dataset
{:x (range 30)
:sin (map #(Math/sin (* % 0.3)) (range 30))
:cos (map #(Math/cos (* % 0.3)) (range 30))}))(def waves
(tc/pivot->longer waves-wide [:sin :cos]
{:target-columns :function
:value-column-name :y}))(-> waves
(pj/lay-line :x :y {:color :function}))A vector of columns written where one column goes does that reshape inside Plotje: the wide table draws one panel with a line, a color and a legend entry for each column β see Series.
(-> waves-wide
(pj/lay-line :x [:sin :cos]))Thick Line
Constant stroke width via :size.
(-> wave
(pj/lay-line :x :y {:size 4}))Line with Points
Points added to a grouped line plot, on the same panel.
(def growth
{:day [1 2 3 4 5 1 2 3 4 5]
:value [10 15 13 18 22 8 12 11 16 19]
:group [:a :a :a :a :a :b :b :b :b :b]})(-> growth
(pj/pose :day :value {:color :group})
pj/lay-line
pj/lay-point)Step
Horizontal-then-vertical connected points.
(-> {:x [1 2 3 4 5]
:y [2 4 1 5 3]}
(pj/lay-step :x :y)
pj/lay-point)Step by Group
Grouped step lines.
(-> growth
(pj/pose :day :value {:color :group})
pj/lay-step
pj/lay-point)Stacked Step
The groups pile up in the order the legend lists them, the first on top, with horizontal-then-vertical segments.
(-> {:x (concat (range 5) (range 5) (range 5))
:y (concat [1 2 3 4 5] [2 2 2 2 2] [3 1 2 1 2])
:group (concat (repeat 5 "A") (repeat 5 "B") (repeat 5 "C"))}
(pj/lay-step :x :y {:position :stack :color :group}))Area
Filled area under a line.
(-> {:x (range 30)
:y (map #(Math/sin (* % 0.3)) (range 30))}
(pj/lay-area :x :y))Stacked Area
The bands pile up in the order the legend lists them, the first on top.
(-> {:x (concat (range 10) (range 10) (range 10))
:y (concat [1 2 3 4 5 4 3 2 1 0]
[2 2 2 3 3 3 2 2 2 2]
[1 1 1 1 2 2 2 1 1 1])
:group (concat (repeat 10 "A") (repeat 10 "B") (repeat 10 "C"))}
(pj/lay-area :x :y {:position :stack :color :group}))Dates on the x-axis
Real time-series usually have an actual date column, not an integer step. Plotje detects temporal columns (java.util.Date via #inst, java.time.LocalDate, LocalDateTime, Instant) and picks calendar-aware tick labels automatically.
(def temp-pose
(-> {:date [#inst "2024-01-01" #inst "2024-02-01" #inst "2024-03-01"
#inst "2024-04-01" #inst "2024-05-01" #inst "2024-06-01"]
:temperature [3 5 9 14 19 23]}
(pj/lay-line :date :temperature)
pj/lay-point))temp-poseMultiple Series Over Time
Pass {:color :group} to get one line per category. Each line joins its points in order of the date on :x, whatever order the rows are in, so unsorted data draws the same lines.
(def months
[#inst "2024-01-01" #inst "2024-02-01" #inst "2024-03-01"
#inst "2024-04-01" #inst "2024-05-01" #inst "2024-06-01"])(-> {:date (concat months months)
:temperature [3 5 9 14 19 23
15 17 19 22 25 28]
:city (concat (repeat 6 "Zurich")
(repeat 6 "Athens"))}
(pj/lay-line :date :temperature {:color :city})
pj/lay-point)Area Over Dates
Filled area also works on a date axis β useful for cumulative metrics where the volume below the curve carries meaning.
(-> {:date [#inst "2024-01-01" #inst "2024-02-01" #inst "2024-03-01"
#inst "2024-04-01" #inst "2024-05-01" #inst "2024-06-01"]
:sales [10 25 30 22 35 40]}
(pj/lay-area :date :sales))See Inference Rules for details on how dates are detected and formatted.
Two Series on One Date Axis
Two series measured over different stretches of time read against each other only if their axes agree. Stacked by pj/arrange with {:cols 1} and no other option, each panel spans its own dates across the same width. A reading directly above another then stands for a different month.
(def zurich
{:date [#inst "2024-01-01" #inst "2024-02-01" #inst "2024-03-01"
#inst "2024-04-01" #inst "2024-05-01" #inst "2024-06-01"]
:temperature [3 5 9 14 19 23]})(def athens
{:date [#inst "2024-05-01" #inst "2024-06-01" #inst "2024-07-01"
#inst "2024-08-01" #inst "2024-09-01" #inst "2024-10-01"]
:temperature [25 28 31 32 27 22]})(def cities
[(-> zurich
(pj/lay-line :date :temperature)
pj/lay-point)
(-> athens
(pj/lay-line :date :temperature)
pj/lay-point)])(pj/arrange cities {:cols 1}):share-scales #{:x} pools the date column across the cells. The pooled domain spans the dates of both series, so each series occupies the part of the axis it covers, and the axis is labelled once, under the lower panel. A month is at the same place in both panels.
(pj/arrange cities {:cols 1 :share-scales #{:x}})An axis holds a date as a number of milliseconds, which is what lets the two date domains be pooled at all.
The two panels line up here because their y axes label at the same width. Where they do not β one series in single digits and one in millions β each panel reserves the room its own labels need, and the shared axis comes out spanning different widths. :align-panels, which pj/arrange takes among its options, reserves the same room on every cell; the Composition chapter covers that and :share-scales on any column.
A Distribution of the Dates
A time series shows what the readings were, and less about when they were taken. pj/marginal puts a distribution of the poseβs :x column in a thin panel above the plot, sharing its axis β and on a date column that distribution is how the observations are spread through time.
(def sightings
{:date [#inst "2021-03-14" #inst "2021-07-02" #inst "2021-11-28"
#inst "2022-02-09" #inst "2022-05-30" #inst "2022-06-11"
#inst "2022-06-25" #inst "2022-07-08" #inst "2022-09-17"
#inst "2023-01-22" #inst "2023-08-05" #inst "2024-02-19"]
:count [2 5 3 8 6 11 9 14 12 7 4 2]})(-> sightings
(pj/lay-point :date :count)
(pj/marginal :top :histogram))Smoothed Time Series
A LOESS smoother drawn over a noisy time series makes the underlying trend easier to see. pj/lay-smooth fits a numerical :y against dates on :x as it does against numbers.
(-> {:date [#inst "2024-01-01" #inst "2024-02-01" #inst "2024-03-01"
#inst "2024-04-01" #inst "2024-05-01" #inst "2024-06-01"
#inst "2024-07-01" #inst "2024-08-01" #inst "2024-09-01"
#inst "2024-10-01" #inst "2024-11-01" #inst "2024-12-01"]
:sales [10 14 12 18 22 19 25 28 24 30 27 33]}
(pj/pose :date :sales)
pj/lay-line
pj/lay-smooth)Zero-Line Baseline
Time series with positive and negative values often benefit from a horizontal reference at zero, drawn with pj/lay-rule-h. The rule is drawn at a value written on the layer, :y-intercept, rather than from a data column, and it is a layer like any other: it paints in the order it was added, and the y axis reaches its value even where the series does not.
(-> {:t (range 12)
:delta [-3 -1 -2 0 2 4 -1 3 5 -2 1 4]}
(pj/lay-line :t :delta)
pj/lay-point
(pj/lay-rule-h {:y-intercept 0 :color "#888"}))Whatβs Next
- Timelines β events, intervals, and schedules on a time axis (Gantt charts, Marey diagrams, annotated time series)
- Relationships β heatmaps, contours, and 2D density
- Polar Coordinates β radial charts for cyclical data
- Gallery β more chart variations with side-by-side code