Thanks for lending us a hand!
1)
The Script attached seemingly has messed up the environment. Running an arbitrary python script results in the following error.
What happened here?
cd mtcode
module load spikey
srun -p spikey --gres station666 python synfire_STDP.py
cp: cannot stat `STDPtestxResults/0/potential.png': No such file or directory
cp: cannot stat `STDPtestxResults/1/potential.png': No such file or directory
cp: cannot stat `STDPtestxResults/2/potential.png': No such file or directory
cp: cannot stat `STDPtestxResults/3/potential.png': No such file or directory
cp: cannot stat `STDPtestxResults/4/potential.png': No such file or directory
cp: cannot stat `STDPtestxResults/5/potential.png': No such file or directory
cp: cannot stat `STDPtestxResults/6/potential.png': No such file or directory
cp: cannot stat `STDPtestxResults/7/potential.png': No such file or directory
Traceback (most recent call last):
File "synfire_STDP.py", line 12, in <module>
import matplotlib as mpl
File "/usr/lib/pymodules/python2.7/matplotlib/__init__.py", line 134, in <module>
from matplotlib.rcsetup import (defaultParams,
File "/usr/lib/pymodules/python2.7/matplotlib/rcsetup.py", line 18, in <module>
from matplotlib.fontconfig_pattern import parse_fontconfig_pattern
File "/usr/lib/pymodules/python2.7/matplotlib/fontconfig_pattern.py", line 23, in <module>
from matplotlib.pyparsing import Literal, ZeroOrMore, \
File "/usr/lib/pymodules/python2.7/matplotlib/pyparsing.py", line 3207, in <module>
_escapedHexChar = Combine( Suppress(_bslash + "0x") + Word(hexnums) ).setParseAction(lambda s,l,t:unichr(int(t[0],16)))
File "/usr/lib/pymodules/python2.7/matplotlib/pyparsing.py", line 2923, in __init__
self.leaveWhitespace()
File "/usr/lib/pymodules/python2.7/matplotlib/pyparsing.py", line 2565, in leaveWhitespace
self.expr = self.expr.copy()
File "/usr/lib/pymodules/python2.7/matplotlib/pyparsing.py", line 681, in copy
cpy = copy.copy( self )
AttributeError: 'module' object has no attribute 'copy'
srun: error: HBPHost6: task 0: Exited with exit code 1
2)
Attached, find a simple test that we ran. Run runBatch and afterwards copy and you will get an output similar to the pictures attached.
Potential_10_X starts a single pulse at 10ms through a casual static synapse and plots the membrane potential of the post neuron.
Also, there is another STDP synapse, with a single pulse running at 20ms, on two different neurons.
Potential_15_X starts the static pulse at 15, same network else.
When starting at 10ms, it clearly suffices to get the post neuron spiking, when starting at 15ms, it sometimes spikes, overall the energy seems leaky, but why would that occur here?
3)
When removing the STDP synapse from the sample script, it would not matter whether we induce the spike at 10 or 15ms. All would cause the post neuron to spike afterwards.
4)
I wanted to play a little with hwa.setIcb and hwa.autoSTDPFrequency as I don't know how exactly they influence the network, but can't right now..
Can you explain setIcb? Will dig again into the paper to understand the STDP Controller..
Greets
Bjoern & Frank
-----Original Message-----
From: Spikey neuromorphic system - User [mailto:[log in to unmask]] On Behalf Of Thomas Pfeil
Sent: Montag, 26. Oktober 2015 21:35
To: [log in to unmask]
Subject: Re: Regularity in Spikey(-STDP) noise
Hello Bjoern & Frank,
first of all, sorry for my late reply.
Depending on your description I can not really reconstruct your
experiment setting.
Ideally, please provide a minimal example.
Interesting additional information would be:
- In the case with STDP: how large are EPSPs in experiments without
spiking? Larger than in the case without STDP?
- Provide details regarding your stimulation protocol.
Reminders that may help:
- If you change the order of connecting external spike sources to
hardware neurons, different synapses are assigned on hardware. Due to
fixed-pattern noise across synapses (static parameter variations due
to production imperfections) EPSPs on the membrane potential are
likely to differ in size.
- Depending on the number of synapses (and their synaptic weight),
leak currents pull the membrane potential to the respective reversal
potential.
Solution: Try to use as identical network configurations in both
experiments as possible, e.g., do not provide input to the plastic
synapse in your static case. Alternatively, compensate for the
different excitabilities of the neuron (caused by leak currents) by
adjusting synaptic weights.
Regarding the "sampling frequency" of the STDP controller:
This frequency defines how fast the global weight update controller
steps through the rows of synapses.
The default value is maximum frequency to achieve the highest temporal
resolution of possible weight changes.
For most use cases I recommend this setting.
For details, see: http://arxiv.org/abs/1201.6255
Kind regards,
Thomas
########################################################################
To unsubscribe from the KIP-SPIKEY-USERS list, click the following link:
https://listserv.uni-heidelberg.de/cgi-bin/wa?SUBED1=KIP-SPIKEY-USERS
|
