Introduction to Summaries¶

When you have a few chains and want to contrast them all with each other, you probably want a summary plot.

To show you how they work, let's make some sample data that all has the same average.

from chainconsumer import Chain, ChainConfig, ChainConsumer, PlotConfig, Truth, make_sample

# Here's what you might start with
df_1 = make_sample(num_dimensions=4, seed=1)
df_2 = make_sample(num_dimensions=5, seed=2)
print(df_1.head())

Out:

          A         B          C          D  log_posterior
0 -0.964563  6.021349  10.684439  14.940567      -3.426756
1  0.196227  5.706161  10.018915  14.806242      -3.016730
2 -0.632224  5.219669  11.663687  13.628753      -4.045577
3 -1.516314  6.174327   9.836924  15.385204      -4.761603
4 -0.276030  4.776105  10.724559  14.600003      -3.047953

Using distributions¶

# And now we give this to chainconsumer
c = ChainConsumer()
c.add_chain(Chain(samples=df_1, name="An Example Contour"))
c.add_chain(Chain(samples=df_2, name="A Different Contour"))
fig = c.plotter.plot_summary()

A, B, C, D, E

Using Errorbars¶

Note that because the errorbar kwarg is specific to this function it is not part of the PlotConfig class.

fig = c.plotter.plot_summary(errorbar=True)

A, B, C, D, E

The other features of ChainConsumer should work with summaries too.

For example, truth values should work just fine.

c.add_truth(Truth(location={"A": 0, "B": 1}, line_style=":", color="red"))
fig = c.plotter.plot_summary(errorbar=True, vertical_spacing_ratio=2.0)

A, B, C, D, E

And similarly, our overrides are generic and so effect this method too.

c.set_override(ChainConfig(bar_shade=False))
c.set_plot_config(PlotConfig(watermark="Preliminary", blind=["E"]))
fig = c.plotter.plot_summary()