Posterior predictive plot gallery¶

A compact tour of plot_predictive's renderings and uncertainty displays, one option per section. For the full plotting API, see the Plotting tutorial.

For demonstration purposes we inject a pre-computed trace with posterior predictive samples into the model, so this notebook runs in seconds without sampling. In practice you would obtain the trace from model.sample().

In [1]:

from pathlib import Path

import arviz as az
import pandas as pd

import hssm

%config InlineBackend.figure_format = 'retina'

fixtures = Path("../../tests/fixtures")
cav_data = pd.read_csv(fixtures / "cavanagh_theta_test.csv", index_col=None)
model = hssm.HSSM(
    data=cav_data,
    include=[{"name": "v", "formula": "v ~ theta + (1|participant_id)"}],
)
model._inference_obj = az.from_netcdf(fixtures / "cavanagh_idata.nc")

from pathlib import Path import arviz as az import pandas as pd import hssm %config InlineBackend.figure_format = 'retina' fixtures = Path("../../tests/fixtures") cav_data = pd.read_csv(fixtures / "cavanagh_theta_test.csv", index_col=None) model = hssm.HSSM( data=cav_data, include=[{"name": "v", "formula": "v ~ theta + (1|participant_id)"}], ) model._inference_obj = az.from_netcdf(fixtures / "cavanagh_idata.nc")

Model initialized successfully.

The default: graded uncertainty bands¶

Graded 50% + 94% equal-tailed bands around the predictive mean, step histograms, legend, despined axes, x-range fit to the data, signed-RT axis annotation.

In [2]:

model.plot_predictive();

model.plot_predictive();

uncertainty="samples" — the model-cartoon idiom¶

Every retained posterior draw as a translucent curve under an opaque mean. More honest about multimodality; noisier at facet scale.

In [3]:

model.plot_predictive(uncertainty="samples");

model.plot_predictive(uncertainty="samples");

uncertainty="both"¶

In [4]:

model.plot_predictive(uncertainty="both");

model.plot_predictive(uncertainty="both");

Custom graded bands¶

hdi= takes a list, one entry per band — masses, "90%" strings, or explicit quantile tuples.

In [5]:

model.plot_predictive(hdi=[0.5, 0.8, 0.95]);

model.plot_predictive(hdi=[0.5, 0.8, 0.95]);

KDE rendering¶

kind="kde" swaps the binned histogram for Gaussian kernel density estimates, evaluated per posterior draw on a shared grid — the mean, bands, and sample curves work identically. On the signed-RT axis the KDE is fit separately per choice side, weighted by response proportion, so no density is smoothed across the gap at zero. Note the inherent tradeoff: kernels still smooth toward zero on each side, so kind="hist" remains the more faithful view right at the non-decision-time gap. bw_method= tunes the bandwidth (scipy's Scott rule by default).

In [6]:

model.plot_predictive(kind="kde");

model.plot_predictive(kind="kde");

In [7]:

model.plot_predictive(kind="kde", uncertainty="samples");

model.plot_predictive(kind="kde", uncertainty="samples");

The classic line-plot look¶

step=False restores the frequency-polygon rendering the plot used before — a connected line through the bin densities — now drawn through the bin centers (the old version anchored at left bin edges, shifting every curve by half a bin) and fully composable with bands and sample curves. Together with the step default and kind="kde", this rounds out the three renderings: step histogram, frequency polygon, and kernel density.

In [8]:

model.plot_predictive(step=False);

model.plot_predictive(step=False);

Mean only¶

uncertainty=None draws just the predictive mean against the observed data — the behavior of earlier HSSM versions.

In [9]:

model.plot_predictive(uncertainty=None);

model.plot_predictive(uncertainty=None);

Facets — the workhorse view¶

Bands propagate to grids unchanged; the legend is grid-level.

In [10]:

model.plot_predictive(col="participant_id", row="conf");

model.plot_predictive(col="participant_id", row="conf");

Styling knobs¶

Per-source colors (dict now genuinely supported), band opacity, mean opacity.

In [11]:

model.plot_predictive(
    colors={"predicted": "#2a9d8f", "observed": "black"},
    alpha_uncertainty=0.4,
    hdi=[0.94],
);

model.plot_predictive( colors={"predicted": "#2a9d8f", "observed": "black"}, alpha_uncertainty=0.4, hdi=[0.94], );

Legacy spellings still work¶

hdi=0.9, hdi="90%", and quantile tuples hdi=(0.05, 0.95) behave as before (single band); prior-predictive titles itself correctly now.

In [12]:

model.plot_predictive(hdi=(0.05, 0.95));

model.plot_predictive(hdi=(0.05, 0.95));