Monitoring weight dynamics of Projection between PopulationViews

[dkoller]

Hi all,

I have created a model of two separate neuron populations A and B. Next, I create a PopulationView of the first 500 neurons of each population. I connect both PopulationViews via a Projection with STDP and monitor the weight dynamics. This works fine and results in increasing weights, however, if I create a PopulationView using neurons 500 to 999, i.e. PVB = B[500:1000], instead of B[:500], the resulting monitored weights are a mix of list([]), NaNs and extremely small values 4.9347e-320. My expectation would be that the outcomes using the first 500 neurons or the second 500 neurons should be the same. I couldn't figure out why this is happening. Does someone have an idea why this happens?

Best,

Dominik

[julien...@gmail.com]

Hi Dominik,

this sounds like a mean bug in the Monitors, we have a look ASAP.

Best

Julien

[julien...@gmail.com]

We have published a potential fix in the master branch. Can you test it? If it works, we will do the same for CUDA and make a new release 4.6.10.1.

[dkoller]

Hi Julien,

thanks for the quick fix! It works well now. 

Best,

Dominik

[julien...@gmail.com]

Great! I wonder since how long that bug was there... Anyway, recording the evolution of connectivity matrices is not a good idea as it can very quickly fill up the RAM for big networks and long simulations. Make sure to dump the recordings into a file quite often.

[Helge Dinkelbach]

Hi,

maybe as addition, there is also the possibility to use the period argument to record only after X steps. This could reduce the pressure on the memory.

Best,

Helge

--
You received this message because you are subscribed to the Google Groups "ANNarchy" group.
To unsubscribe from this group and stop receiving emails from it, send an email to annarchy+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/annarchy/887d073d-f436-4685-a30a-1fba1ffba04en%40googlegroups.com.

[dkoller]

Hi all,

thanks for your suggestions! I'm actually using the period argument and so far I had no issues with RAM. 

Best,

Dominik