Closed igrahek closed 1 month ago
digicosmos86
commented
4 months ago Hi @igrahek!
Thanks for looking into this. This seems to be a new error with a new version of PyMC. What is your PyMC version? Can you try downgrading this to 5.9.4 and let us know if this problem still persisits? Once the PyMC version changes, you might want to change the JAX version as well. We recommend anything before 0.4.16.
Thanks! Paul
igrahek
commented
4 months ago Thanks for the quick response @digicosmos86! My PyMC version is 5.9.2. This is what was installed with HSSM as default yesterday.
You were not able to reproduce the error?
I downgraded jax to 0.4.14 but this didn't solve the error. This code still doesn't run:
cav_data = hssm.load_data("cavanagh_theta")
model_safe = hssm.HSSM(
data=cav_data,
hierarchical=True,
model="ddm",
loglik_kind="analytical"
)
model_safe.sample(sampler="mcmc")Here's my whole conda env:
(pyHSSM) [igrahek@login009 ~]$ conda list
# packages in environment at /users/igrahek/.conda/envs/pyHSSM:
#
# Name Version Build Channel
_libgcc_mutex 0.1 conda_forge conda-forge
_openmp_mutex 4.5 2_gnu conda-forge
anyio 4.2.0 <pip>
argon2-cffi 23.1.0 <pip>
argon2-cffi-bindings 21.2.0 <pip>
arrow 1.3.0 <pip>
arviz 0.14.0 <pip>
asttokens 2.4.1 <pip>
async-lru 2.0.4 <pip>
attrs 23.2.0 <pip>
Babel 2.14.0 <pip>
bambi 0.12.0 <pip>
beautifulsoup4 4.12.3 <pip>
bleach 6.1.0 <pip>
bzip2 1.0.8 h7f98852_4 conda-forge
ca-certificates 2022.9.24 ha878542_0 conda-forge
cachetools 5.3.2 <pip>
certifi 2024.2.2 <pip>
cffi 1.16.0 <pip>
cftime 1.6.3 <pip>
charset-normalizer 3.3.2 <pip>
cloudpickle 3.0.0 <pip>
coloredlogs 15.0.1 <pip>
comm 0.2.1 <pip>
cons 0.4.6 <pip>
contourpy 1.2.0 <pip>
cycler 0.12.1 <pip>
Cython 3.0.8 <pip>
debugpy 1.8.1 <pip>
decorator 5.1.1 <pip>
defusedxml 0.7.1 <pip>
etuples 0.3.9 <pip>
executing 2.0.1 <pip>
fastjsonschema 2.19.1 <pip>
fastprogress 1.0.3 <pip>
filelock 3.13.1 <pip>
flatbuffers 23.5.26 <pip>
fonttools 4.48.1 <pip>
formulae 0.5.1 <pip>
fqdn 1.5.1 <pip>
fsspec 2024.2.0 <pip>
graphviz 0.20.1 <pip>
h11 0.14.0 <pip>
hddm-wfpt 0.1.1 <pip>
HSSM 0.2.0 <pip>
httpcore 1.0.3 <pip>
httpx 0.26.0 <pip>
huggingface-hub 0.15.1 <pip>
humanfriendly 10.0 <pip>
idna 3.6 <pip>
ipykernel 6.29.2 <pip>
ipython 8.21.0 <pip>
isoduration 20.11.0 <pip>
jax 0.4.14 <pip>
jaxlib 0.4.14 <pip>
jedi 0.19.1 <pip>
Jinja2 3.1.3 <pip>
joblib 1.3.2 <pip>
json5 0.9.14 <pip>
jsonpointer 2.4 <pip>
jsonschema 4.21.1 <pip>
jsonschema-specifications 2023.12.1 <pip>
jupyter-events 0.9.0 <pip>
jupyter-lsp 2.2.2 <pip>
jupyter_client 8.6.0 <pip>
jupyter_core 5.7.1 <pip>
jupyter_server 2.12.5 <pip>
jupyter_server_terminals 0.5.2 <pip>
jupyterlab 4.1.1 <pip>
jupyterlab_pygments 0.3.0 <pip>
jupyterlab_server 2.25.3 <pip>
kiwisolver 1.4.5 <pip>
ld_impl_linux-64 2.39 hc81fddc_0 conda-forge
libffi 3.4.2 h7f98852_5 conda-forge
libgcc-ng 12.2.0 h65d4601_19 conda-forge
libgomp 12.2.0 h65d4601_19 conda-forge
libnsl 2.0.0 h7f98852_0 conda-forge
libsqlite 3.40.0 h753d276_0 conda-forge
libuuid 2.32.1 h7f98852_1000 conda-forge
libzlib 1.2.13 h166bdaf_4 conda-forge
logical-unification 0.4.6 <pip>
MarkupSafe 2.1.5 <pip>
matplotlib 3.8.3 <pip>
matplotlib-inline 0.1.6 <pip>
miniKanren 1.0.3 <pip>
mistune 3.0.2 <pip>
ml-dtypes 0.3.2 <pip>
mpmath 1.3.0 <pip>
multipledispatch 1.0.0 <pip>
nbclient 0.9.0 <pip>
nbconvert 7.16.0 <pip>
nbformat 5.9.2 <pip>
ncurses 6.3 h27087fc_1 conda-forge
nest-asyncio 1.6.0 <pip>
netCDF4 1.6.5 <pip>
notebook 7.1.0 <pip>
notebook_shim 0.2.4 <pip>
numpy 1.25.2 <pip>
numpyro 0.12.1 <pip>
onnx 1.15.0 <pip>
onnxruntime 1.17.0 <pip>
openssl 3.0.7 h166bdaf_0 conda-forge
opt-einsum 3.3.0 <pip>
overrides 7.7.0 <pip>
packaging 23.2 <pip>
pandas 2.2.0 <pip>
pandocfilters 1.5.1 <pip>
parso 0.8.3 <pip>
pexpect 4.9.0 <pip>
pillow 10.2.0 <pip>
pip 22.3.1 pyhd8ed1ab_0 conda-forge
platformdirs 4.2.0 <pip>
prometheus_client 0.20.0 <pip>
prompt-toolkit 3.0.43 <pip>
protobuf 4.25.2 <pip>
psutil 5.9.8 <pip>
ptyprocess 0.7.0 <pip>
pure-eval 0.2.2 <pip>
pycparser 2.21 <pip>
Pygments 2.17.2 <pip>
pymc 5.9.2 <pip>
pyparsing 3.1.1 <pip>
pytensor 2.17.3 <pip>
python 3.11.0 ha86cf86_0_cpython conda-forge
python-dateutil 2.8.2 <pip>
python-json-logger 2.0.7 <pip>
pytz 2024.1 <pip>
PyYAML 6.0.1 <pip>
pyzmq 25.1.2 <pip>
readline 8.1.2 h0f457ee_0 conda-forge
referencing 0.33.0 <pip>
requests 2.31.0 <pip>
rfc3339-validator 0.1.4 <pip>
rfc3986-validator 0.1.1 <pip>
rpds-py 0.18.0 <pip>
scikit-learn 1.4.0 <pip>
scipy 1.10.1 <pip>
seaborn 0.13.2 <pip>
Send2Trash 1.8.2 <pip>
setuptools 65.5.1 pyhd8ed1ab_0 conda-forge
six 1.16.0 <pip>
sniffio 1.3.0 <pip>
soupsieve 2.5 <pip>
ssm-simulators 0.6.1 <pip>
stack-data 0.6.3 <pip>
sympy 1.12 <pip>
terminado 0.18.0 <pip>
threadpoolctl 3.3.0 <pip>
tinycss2 1.2.1 <pip>
tk 8.6.12 h27826a3_0 conda-forge
toolz 0.12.1 <pip>
tornado 6.4 <pip>
tqdm 4.66.2 <pip>
traitlets 5.14.1 <pip>
types-python-dateutil 2.8.19.20240106 <pip>
typing_extensions 4.9.0 <pip>
tzdata 2024.1 <pip>
tzdata 2022f h191b570_0 conda-forge
uri-template 1.3.0 <pip>
urllib3 2.2.0 <pip>
wcwidth 0.2.13 <pip>
webcolors 1.13 <pip>
webencodings 0.5.1 <pip>
websocket-client 1.7.0 <pip>
wheel 0.38.4 pyhd8ed1ab_0 conda-forge
xarray 2024.1.1 <pip>
xarray-einstats 0.7.0 <pip>
xz 5.2.6 h166bdaf_0 conda-forgeHi @igrahek,
No I can't seem to reproduce this error. This seems to be an error that is produced at runtime. We do use the CheckParameter function from PyMC to ensure that a >= 0 during the computation of the analytical log-likelihood function, but with log-logit, priors are chosen to prevent sampling negative as.
What is the platform where your code is run? I could check on that platform as well. One solution to this is that we can remove this check in the next version since there are already mechanisms preventing out of bound samples
igrahek
commented
4 months ago Thanks for checking @digicosmos86!
I'm running this on Oscar, in a conda environment that I sent above. I assume you're using Oscar too? Do you see any obvious differences between your environment and mine?
digicosmos86
commented
4 months ago @igrahek I finally got a chance to test this code on Oscar. It seems that the code is running fine at the moment. Here's what I did to try to reproduce the error that you got:
First, I created a new conda environment. I used the miniforge module since it has mamba by default:
module load miniforge
mamba create -n test-hssm python=3.11
mamba activate test-hssm
mamba install pymc=5.9.2
pip install hssmThen I ran your code
import hssm
hssm.set_floatX("float32")
cav_data = hssm.load_data("cavanagh_theta")
model_safe = hssm.HSSM(
data=cav_data,
hierarchical=True,
model="ddm",
loglik_kind="analytical"
)
model_safe.sample(sampler="mcmc")Can you try this and see if this works for you?
digicosmos86
commented
4 months ago Hi @igrahek,
Is this issue still persisting for you? It seems that some of the other issues are the same symptoms of the same underlying problem, so I just want to find a fix for all of them.
Thanks! Paul
Hi guys, I'm just trying out the new version, and I'm still running into this error which likely has to do with where samples are initialized. I thought that the log_logit would fix that, so that I don't have to manually specify init values.
Here's the code:
Here's the error:
Traceback (most recent call last): File "/users/igrahek/.conda/envs/pyHSSM/lib/python3.11/site-packages/pytensor/compile/function/types.py", line 970, in call self.vm() pymc.logprob.utils.ParameterValueError: a >= 0