Error using MCMC
Opened this issue · 2 comments
doctorwes commented
I am using the same model as in #195 (closed; thank you!), but attempting to use MCMC rather than VI.
(Note that I can successfully run the sample notebook MCMC-logregression.ipynb)
MC = inf.inference.MCMC(num_results=1000)
m.fit({"x": data}, MC)
I get the following error:
---------------------------------------------------------------------------
FailedPreconditionError Traceback (most recent call last)
/home/wkp/.conda/envs/WKPenv/lib/python3.6/site-packages/tensorflow/python/client/session.py in _do_call(self, fn, *args)
1355 try:
-> 1356 return fn(*args)
1357 except errors.OpError as e:
/home/wkp/.conda/envs/WKPenv/lib/python3.6/site-packages/tensorflow/python/client/session.py in _run_fn(feed_dict, fetch_list, target_list, options, run_metadata)
1340 return self._call_tf_sessionrun(
-> 1341 options, feed_dict, fetch_list, target_list, run_metadata)
1342
/home/wkp/.conda/envs/WKPenv/lib/python3.6/site-packages/tensorflow/python/client/session.py in _call_tf_sessionrun(self, options, feed_dict, fetch_list, target_list, run_metadata)
1428 self._session, options, feed_dict, fetch_list, target_list,
-> 1429 run_metadata)
1430
FailedPreconditionError: Error while reading resource variable mcmc_sample_chain_1/mh_bootstrap_results/hmc_kernel_bootstrap_results/maybe_call_fn_and_grads/value_and_gradients/value_and_gradient/dense_40/bias from Container: localhost. This could mean that the variable was uninitialized. Not found: Resource localhost/mcmc_sample_chain_1/mh_bootstrap_results/hmc_kernel_bootstrap_results/maybe_call_fn_and_grads/value_and_gradients/value_and_gradient/dense_40/bias/N10tensorflow3VarE does not exist.
[[{{node mcmc_sample_chain_1/mh_bootstrap_results/hmc_kernel_bootstrap_results/maybe_call_fn_and_grads/value_and_gradients/value_and_gradient/dense_40/BiasAdd/ReadVariableOp}}]]
rcabanasdepaz commented
For probabilistic models with (trainable) NNs, it is not possible to use MCMC. Instead, you should use variational inference.
doctorwes commented
Thank you for your advice - I was not aware of that.
Wesley Phoa
Sent
… On Feb 24, 2020, at 2:58 AM, Rafael Cabañas de Paz ***@***.***> wrote:
For probabilistic models with (trainable) NNs, it is not possible to use MCMC. Instead, you should use variational inference.
—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub, or unsubscribe.