BUG: using exoplanet package - NotImplementedError: ValuedVar should not be present in the final graph!

[vdeloupy]

Describe the issue:

Hello guys!

While trying to run the exact example from exoplanet package documentation (cf Transit, Occultations and Eclipses), I am encountering the following issue:

NotImplementedError: ValuedVar should not be present in the final graph!

This arises after I tried to upgrade my old PyMC3/theano/exoplanet script to newer PyMC/PyTensor/exoplanet[pymc] versions. I must admit I have no real idea where this issue arises from as the exoplanet tutorial script is quite simple and direct.
I also opened an issue on exoplanet-dev issue

Thanks for your help!

Reproduceable code example:

import pymc_ext as pmx
import pymc as pm
import exoplanet as xo
import matplotlib.pyplot as plt 
import numpy as np

random = np.random.default_rng(123)
num_transits = 4
t = np.arange(0, 35, 0.02)
yerr = 5e-4

with pm.Model():
    # The baseline flux
    mean = pm.Normal("mean", mu=0.0, sigma=1.0)

    # Often the best parameterization is actually in terms of the
    # times of two reference transits rather than t0 and period
    t0 = pm.Normal("t0", mu=4.35, sigma=1.0)
    t1 = pm.Normal("t1", mu=33.2, sigma=1.0)
    period = pm.Deterministic("period", (t1 - t0) / num_transits)

    # The Kipping (2013) parameterization for quadratic limb darkening
    # paramters
    u = xo.distributions.quad_limb_dark("u", initval=np.array([0.3, 0.2]))

    # The radius ratio and impact parameter; these parameters can
    # introduce pretty serious covariances and are ripe for
    # reparameterization
    log_r = pm.Normal("log_r", mu=np.log(0.04), sigma=2.0)
    r = pm.Deterministic("r", pt.exp(log_r))
    b = xo.distributions.impact_parameter("b", r, initval=0.35)

    # Set up a Keplerian orbit for the planets
    orbit = xo.orbits.KeplerianOrbit(period=period, t0=t0, b=b)

    # Compute the model light curve; note that we index using `[:, 0]`
    # since `get_light_curve` returns an object with the shape
    # `(n_times, n_planets)`
    light_curve = (
        xo.LimbDarkLightCurve(u[0], u[1]).get_light_curve(
            orbit=orbit, r=r, t=t
        )[:, 0]
        + mean
    )

    # Here we track the value of the model light curve for plotting
    # purposes
    pm.Deterministic("__light_curve", light_curve)

    # ================================================== #
    # Simulate data from the model for testing           #
    # You should remove the following lines in your code #
    # ================================================== #
    y = pmx.eval_in_model(light_curve)
    y += yerr * random.normal(size=len(y))
    # =============== end simulated data =============== #

    # The likelihood function assuming known Gaussian uncertainty
    pm.Normal("obs", mu=light_curve, sigma=yerr, observed=y)

    trace = pm.sample(
        tune=1000, draws=1000, cores=2, chains=2, init="adapt_full"
    )

# Plot the results
q16, q50, q84 = np.percentile(
    trace.posterior["__light_curve"].values, [16, 50, 84], axis=(0, 1)
)
plt.plot(t, y, ".k", ms=2, label="data")
plt.plot(t, q50)
plt.fill_between(t, q16, q84, alpha=0.3, label="posterior")
plt.xlim(0.0, 35)
plt.legend(fontsize=12, loc=3)
plt.xlabel("time [days]")
plt.ylabel("relative flux")

# Compute the convergence stats
az.summary(trace, var_names=["^(?!__).*"], filter_vars="regex")

Error message:

Traceback (most recent call last):
  File "/mnt/c/Users/vince/OneDrive/Bureau/DC20/ASTEP/package/ProSTEP3/prostep3/src/prostep3/scripts/redo_fit_prose3_vincent.py", line 838, in <module>
    main(args)
  File "/mnt/c/Users/vince/OneDrive/Bureau/DC20/ASTEP/package/ProSTEP3/prostep3/src/prostep3/scripts/redo_fit_prose3_vincent.py", line 720, in main
    opt,summary,trace=fit_georgina(data,predictions)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/c/Users/vince/OneDrive/Bureau/DC20/ASTEP/package/ProSTEP3/prostep3/src/prostep3/scripts/redo_fit_prose3_vincent.py", line 255, in fit_georgina
    opt = pm.find_MAP(start=model.initial_point())
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/jaskier/miniconda3/envs/prostep3-dev/lib/python3.11/site-packages/pymc/tuning/starting.py", line 134, in find_MAP
    model.check_start_vals(start)
  File "/home/jaskier/miniconda3/envs/prostep3-dev/lib/python3.11/site-packages/pymc/model/core.py", line 1766, in check_start_vals
    initial_eval = self.point_logps(point=elem, **kwargs)
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/jaskier/miniconda3/envs/prostep3-dev/lib/python3.11/site-packages/pymc/model/core.py", line 1798, in point_logps
    factor_logps_fn = [pt.sum(factor) for factor in self.logp(factors, sum=False)]
                                                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/jaskier/miniconda3/envs/prostep3-dev/lib/python3.11/site-packages/pymc/model/core.py", line 696, in logp
    rv_logps = transformed_conditional_logp(
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/jaskier/miniconda3/envs/prostep3-dev/lib/python3.11/site-packages/pymc/logprob/basic.py", line 595, in transformed_conditional_logp
    temp_logp_terms = conditional_logp(
                      ^^^^^^^^^^^^^^^^^
  File "/home/jaskier/miniconda3/envs/prostep3-dev/lib/python3.11/site-packages/pymc/logprob/basic.py", line 479, in conditional_logp
    fgraph = construct_ir_fgraph(rv_values, ir_rewriter=ir_rewriter)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/jaskier/miniconda3/envs/prostep3-dev/lib/python3.11/site-packages/pymc/logprob/rewriting.py", line 249, in construct_ir_fgraph
    replacements = tuple((rv, valued_rv(rv, value)) for rv, value in ir_rv_values.items())
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/jaskier/miniconda3/envs/prostep3-dev/lib/python3.11/site-packages/pymc/logprob/rewriting.py", line 249, in <genexpr>
    replacements = tuple((rv, valued_rv(rv, value)) for rv, value in ir_rv_values.items())
                              ^^^^^^^^^^^^^^^^^^^^
  File "/home/jaskier/miniconda3/envs/prostep3-dev/lib/python3.11/site-packages/pytensor/graph/op.py", line 304, in __call__
    compute_test_value(node)
  File "/home/jaskier/miniconda3/envs/prostep3-dev/lib/python3.11/site-packages/pytensor/graph/op.py", line 125, in compute_test_value
    thunk()
  File "/home/jaskier/miniconda3/envs/prostep3-dev/lib/python3.11/site-packages/pytensor/graph/op.py", line 531, in rval
    r = p(n, [x[0] for x in i], o)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/jaskier/miniconda3/envs/prostep3-dev/lib/python3.11/site-packages/pymc/logprob/abstract.py", line 245, in perform
    raise NotImplementedError("ValuedVar should not be present in the final graph!")
NotImplementedError: ValuedVar should not be present in the final graph!

PyMC version information:

• Version of exoplanet: 0.6.0
• Version of PyMC: 5.19.1
• Version of PyTensor: 2.26.4
• Operating system: Linux ZBook 5.15.167.4-microsoft-standard-WSL2 1 SMP Tue Nov 5 00:21:55 UTC 2024 x86_64 x86_64 x86_64 GNU/Linux
• Python version & installation method (pip, conda, etc.): Python 3.11.11, everything installed with pip inside a conda environment (conda create -n name python=3.11, package by conda-forge)

Context for the issue:

I am working on adapting a data reduction pipeline for exoplanetary transits detection using a 40-cm newtonian telescope.

[welcome]

]
🎉 Welcome to PyMC! 🎉 We're really excited to have your input into the project! 💖

If you haven't done so already, please make sure you check out our Contributing Guidelines and Code of Conduct.

[ricardoV94]

Try setting pytensor.config.compute_test_value="off" We are deprecating that functionality and not everything in PyMC handles it.

[vdeloupy]

That worked! Thanks @ricardoV94
I'm closing the issue.