lucianopaz
commented
2 years ago By any chance, could you share your model code with us? We've had reports similar to this on Mac M1 chips, but haven't been able to reproduce the error.
Closed mathtimm closed 1 year ago
lucianopaz
commented
2 years ago By any chance, could you share your model code with us? We've had reports similar to this on Mac M1 chips, but haven't been able to reproduce the error.
mathtimm
commented
2 years ago Yes of course, here it is. It's for 13 distinct light curves obtained with various instruments and in multiple filters.
` with pm.Model() as model:
# Stellar parameters
# -----------------
r_s = pm.Normal('r_s',0.326568 ,0.00988657)
m_s = pm.Normal('m_s',0.30713 ,0.0202867)
# Orbital parameters
# -----------------
t0= pm.Normal('t0',2459384.5731,0.005)
p = pm.Normal('P',16.336545,0.001)
b = pm.Uniform("b",0.,1.,testval=0.4)
# Keplerian orbit
# ---------------
orbit = xo.orbits.KeplerianOrbit(period=p, t0=t0, r_star=r_s, b=b, m_star=m_s)
pm.Deterministic("a",orbit.a)
pm.Deterministic ('i',orbit.incl * 180/np.pi)
pm.Deterministic('a/r_s', orbit.a / orbit.r_star)
depths =[]
names=[]
# Loop over the instruments
# -------------------------
#Loop over the filters
for i, (filt,ldc) in enumerate(filters.items()):
# The limb darkening
u = xo.QuadLimbDark(f"u_{filt}",testval=np.array(ldc))
star = xo.LimbDarkLightCurve(u)
# Radius
depth = pm.Uniform(f"depth_{filt}", 1e-3, 9e-3,testval=4e-3)
ror = pm.Deterministic(f"ror_{filt}",star.get_ror_from_approx_transit_depth(depth, b))
r_p = pm.Deterministic(f"r_p_{filt}", ror * r_s) # In solar radius
r = pm.Deterministic(f'r_{filt}',r_p*1/earth2sun)
#Loop over all the light curves
for n, (name, (x, y, yerr, texp, X,f)) in enumerate(data[filt].items()):
names.append(name)
if filt == 'zp':
alpha = pm.Normal(f'alpha_{name}',0.86,0.001)
y_p = y*(1+alpha) - alpha
else:
y_p=y
# starry light-curve
light_curves = star.get_light_curve(orbit=orbit, r=r_p, t=x, texp=texp/24/60/60)
transit = pm.Deterministic(f"transit_{name}", pm.math.sum(light_curves, axis=-1))
# Systematics and final model
c=np.linalg.lstsq(X.T,y,rcond=None)[0]
w = pm.Flat(f'w_{name}',shape=len(X),testval=np.array(c))
systematics = pm.Deterministic(f'systematics_{name}',w@X.copy())
mu = pm.Deterministic(f"mu_{name}", transit+systematics)
# Likelihood function
sigma = pm.Uniform(f"sigma_{name}",0.5,4,testval=f) # Adjustment of the errors
pm.Normal(f"obs_{name}", mu=mu, sd=yerr*sigma, observed=y_p)
# Maximum a posteriori
# --------------------
opt = pmx.optimize(start=model.test_point)`I created a new clean environment and installed exoplanet and exoplanet-ext with pip, the problem seems fixed now. The only difference in terms of version is that I am running the 3.11.5 of PyMC3, but I tried upgrading the package in the other environment and the bug still happens. I'm not sure what is happening but it might not be related to PyMC3 but rather to exoplanet and the compatibility with PyMC3 and Theano-PyMC ?
lucianopaz
commented
2 years ago From the other reports that we had, it seems related to BLAS, scipy and the cholesky decomposition. We haven’t identified the exact configuration options that trigger the error though, but we think that at its core, what is failing is related to this:
from scipy.linalg import lapack
import cloudpickle
cloudpickle.dumps(lapack.cpotrf)
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
Input In [39], in <cell line: 5>()
1 from scipy.linalg import lapack
2 import cloudpickle
----> 5 cloudpickle.dumps(lapack.cpotrf)
File ~/anaconda3/envs/test_env/lib/python3.10/site-packages/cloudpickle/cloudpickle_fast.py:73, in dumps(obj, protocol, buffer_callback)
69 with io.BytesIO() as file:
70 cp = CloudPickler(
71 file, protocol=protocol, buffer_callback=buffer_callback
72 )
---> 73 cp.dump(obj)
74 return file.getvalue()
File ~/anaconda3/envs/test_env/lib/python3.10/site-packages/cloudpickle/cloudpickle_fast.py:633, in CloudPickler.dump(self, obj)
631 def dump(self, obj):
632 try:
--> 633 return Pickler.dump(self, obj)
634 except RuntimeError as e:
635 if "recursion" in e.args[0]:
TypeError: cannot pickle 'fortran' objectIt looks like this problem might be solved once numpy/numpy#21767 is fixed.
I am getting a TypeError when I try to sample with multiple cores, the issue does not appear if cores=1. I am working on the modeling of exoplanet transits using the code exoplanet that makes use of PyMC3.
Complete error traceback
```python `--------------------------------------------------------------------------- TypeError Traceback (most recent call last) Input In [10], inCould this be a version issue ? I am having trouble finding the right combination of pymc3, pymc3-ext and exoplanet.
Thanks a lot in advance !
Versions and main components