How to Visualise a JPT

See also

Visualising JPTs — the full illustrated tutorial with rendered plots for every example on this page.

pyjpt ships with two visualisation backends selectable via an engine parameter:

• 'matplotlib' — static PNG / SVG / PDF output

• 'plotly' — interactive HTML figures; static export via kaleido

Install the backend(s) you need:

pip install pyjpt[matplotlib]
pip install pyjpt[plotly]          # includes kaleido for static export
sudo apt-get install graphviz      # required for the tree plot

Plotting the Tree Structure

plot() renders the decision tree as a Graphviz SVG. The distribution mini-plots inside each leaf are drawn by the selected engine:

from IPython.display import SVG, display

path = model.plot(
    title='Iris JPT',
    filename='iris',
    directory='/tmp',
    engine='matplotlib',   # or 'plotly'
    view=False,
)
display(SVG(path))

Pass plotvars to restrict leaf plots to a subset of variables, and nodefill / leaffill for custom node colours:

model.plot(
    filename='iris_petals',
    directory='/tmp',
    engine='matplotlib',
    plotvars=[varnames['petal length (cm)'], varnames['species']],
    nodefill='#d0e8ff',
    leaffill='#ffe8d0',
    view=False,
)

Plotting Individual Distributions

Every distribution object exposes dist.plot(engine=...). Pass it a distribution retrieved from a leaf or from a posterior query:

# Numeric CDF — Matplotlib returns a Figure
fig = numeric_dist.plot(
    engine='matplotlib',
    title='Petal length',
    xlabel='petal length (cm)',
    fname='petal_length',
    directory='/tmp',
)

# Symbolic bar chart — Plotly returns a go.Figure
fig = symbolic_dist.plot(
    engine='plotly',
    title='Species posterior',
    horizontal=True,
    color='rgb(15, 21, 110)',
)
fig.show()

Accepted options for the Plotly engine:

Parameter	Description
color	CSS colour, rgb(r,g,b), or hex string
horizontal	Horizontal bars for symbolic/integer dists
max_values	Show only the top-N values
alphabet	Sort bars alphabetically instead of by probability
fill	Plotly fill style, e.g. 'tozeroy'
dim	Gaussian 2-D: 2 = heatmap, 3 = surface

Saving Figures

Matplotlib — pass fname and directory to any .plot() call; the PNG is written automatically.

Plotly — interactive HTML:

dist.plot(engine='plotly', fname='out.html', directory='/tmp')

Plotly — static image (PNG, SVG, JPEG, WebP via kaleido):

dist.plot(engine='plotly', fname='out.svg', directory='/tmp')
# or directly on the returned figure:
fig.write_image('/tmp/out.png', scale=2)

Custom Rendering Engine

Subclass DistributionRendering and pass the instance as engine to any plot() call:

from jpt.plotting.engines.rendering import DistributionRendering

class MyRenderer(DistributionRendering):
    def plot_numeric(self, dist, **kwargs):
        ...   # custom rendering logic

model.plot(engine=MyRenderer(), directory='/tmp', view=False)