search

ioam / topographica

A general-purpose neural simulator focusing on topographic maps.
topographica.org
BSD 3-Clause "New" or "Revised" License
53 stars 32 forks source link

optimized.so: undefined symbol: PyFPE_jbuf #697

Open dancehours opened 3 years ago

dancehours commented 3 years ago

Dear Bednar,

I have a problem for using load_snapshot, for example I try load_snapshot('/GCAL_2000000.typ') and get the following errors: Loading error: Traceback (most recent call last): File "/home/wenqi/miniconda2/envs/venv/topographica/topo/command/init.py", line 341, in load_snapshot pickle.load(snapshot) ImportError: /home/wenqi/miniconda2/envs/venv/topographica/topo/optimized/optimized.so: undefined symbol: PyFPE_jbuf

I would like to ask: how to solve this problem ?

jbednar commented 3 years ago

An error like that typically indicates that you have non-matching versions somewhere, either between some python code and the corresponding binary libraries, or maybe between the version used to create the .typ file and the current version. To decide between those cases, see if you can create a network, save a snapshot, and load it back in. If that works fine, then presumably the .typ file was saved with some older version, and you'd need to recreate the older environment if you want to use that .typ file. If you get the problem even with a brand-new .typ file, then it's likely an issue with your current environment, in which case you should try building it fresh.

dancehours commented 3 years ago

Could it be possible to remove the optimized part connecting to load_snapshot function? I don't see what is the effect for load_snapshot to use optimized part. Recently I built new snapshots of GCAL model, but get the same error.

jbednar commented 3 years ago

I don't think that we optimized snapshot loading, but the actual operation of the network. So you see an error to do with the optimized object just because you are trying to restore the state of an optimized object. You can turn off optimization altogether, but things will get 100X slower, and so that's not normally practical. In any case if you get the same error for newly saved snapshots I'd try to create your environment fresh, using older versions of the dependent code that were previously tested with Topographica.