Linear trend Changepoints Misaligned in Plot

[gdeleva]

Running the following code:

delta = Prior(
    "Laplace",
    mu=Prior("Normal", mu=np.array([-10,-20, -30,-40]), sigma=np.array([0.001,0.001, 0.001, 0.001]), dims=("changepoint")),
    b=Prior("HalfNormal", sigma=0.001),
    dims=("changepoint"),
)
priors = dict(delta=delta)

trend = LinearTrend(
    priors=priors,
    n_changepoints=4,
)
prior = trend.sample_prior(random_seed=rng)
curve = trend.sample_curve(prior)
_, axes = trend.plot_curve(curve, sample_kwargs={"rng": rng})
ax = axes[0]
ax.set(
    xlabel="Time",
    ylabel="Trend",
    title=f"Linear Trend with {trend.n_changepoints} Change Points",
)

I get the following plot

The 4 vertical lines are not aligned with the change points which are - I suppose - 0, 0.333, 0.667, 1.

[williambdean]

Would you be interested in making the change to have this behavior?

[gdeleva]

@williambdean Sure, I'll take care of it within the next few days.

On a related note, it appears that the final value of mu has no effect. I suspect this is because the change point occurs at max(t). If this behavior is intentional, then the documentation may be somewhat misleading — it suggests the last change point should occur at M × max(t) / (M + 1). Could you please confirm?

https://www.pymc-marketing.io/en/latest/api/generated/pymc_marketing.mmm.linear_trend.LinearTrend.html

[williambdean]

Your visual is probably best assumption

[gdeleva]

Thanks. Will look into it then.

[williambdean]

We can reproduce this with

import numpy as np

import matplotlib.pyplot as plt

from pymc_marketing.prior import Prior
from pymc_marketing.mmm import LinearTrend

seed = sum(map(ord, "Seeing the disconnect"))
rng = np.random.default_rng(seed)

trend = LinearTrend(n_changepoints=3)

prior = trend.sample_prior(random_seed=rng)
curve = trend.sample_curve(prior)
trend.plot_curve(curve, random_seed=rng)
plt.show()

[gdeleva]

@williambdean Thanks. Planning to work on this issue this week.

[williambdean]

Thank you @gdeleva !

[williambdean]

If we can provide any assistance, let us know. Feel free to open PR even if it is draft