search

SpineML / SpineCreator

A GUI for neural network models. A graphical editor for SpineML models.
http://spineml.github.io/spinecreator/
GNU General Public License v3.0
17 stars 7 forks source link

Graphs interface has bug(s)? #82

Closed sebjameswml closed 9 years ago

sebjameswml commented 9 years ago

I just had a situation where the Graphs interface was showing crazy results for a simple network. After a while spent investigating what might be wrong with the network output, I found that the output is fine, it was the SpineCreator graphing that had gone wrong.

sebjameswml commented 9 years ago

I have a demo of the problem.

The data is here: https://sebjames.zapto.org/owncloud/index.php/s/PxKYU0sKdNMhemv

On a freshly started SpineCreator (from the master branch), go to the Graphs interface and use the "folder" icon to load the log "SNr_v_log.bin". From this, make a graph of index 8. You should see 3 spikes, then a buildup towards a 4th spike, then a drop to the 'C' (as in the Izhikevich ABCD model) reset value for the voltage, then a jump to voltage > Vpeak.

It looks like this:

sc_graphs_iface_wrong

Opening the same log using the load_sc_data function from SpineCreator/analysis_utils/matlab/ shows sensible data:

a = load_sc_data ('STN_v_log.bin',60) % The log contains data for 60 neurons plot (a(9,:)) % Index 9 in matlab/octave is index 8 in the SC Graphs interface.

Gives:

octave_load_sc_data_right

ajc158 commented 9 years ago

Looks like one of those graphs is the STN, and one the snr... On 28 Sep 2015 1:59 pm, "Seb James" notifications@github.com wrote:

I have a demo of the problem.

The data is here: https://sebjames.zapto.org/owncloud/index.php/s/PxKYU0sKdNMhemv

On a freshly started SpineCreator (from the master branch), go to the Graphs interface and use the "folder" icon to load the log "SNr_v_log.bin". From this, make a graph of index 8. You should see 3 spikes, then a buildup towards a 4th spike, then a drop to the 'C' (as in the Izhikevich ABCD model) reset value for the voltage, then a jump to voltage > Vpeak.

It looks like this:

[image: sc_graphs_iface_wrong] https://cloud.githubusercontent.com/assets/130308/10135823/cd108d82-65e8-11e5-8964-e5d7f7e76641.png

Opening the same log using the load_sc_data function from SpineCreator/analysis_utils/matlab/ shows sensible data:

a = load_sc_data ('STN_v_log.bin',60) % The log contains data for 60 neurons plot (a(9,:)) % Index 9 in matlab/octave is index 8 in the SC Graphs interface.

Gives:

[image: octave_load_sc_data_right] https://cloud.githubusercontent.com/assets/130308/10135834/e280c5ec-65e8-11e5-9be7-3c9670769682.png

— Reply to this email directly or view it on GitHub https://github.com/SpineML/SpineCreator/issues/82#issuecomment-143737366 .

sebjameswml commented 9 years ago

Oh, bother so they are. Hold on. Yes, it's the component output that's at fault, rather than the interface.

sebjameswml commented 9 years ago

Back to debugging the component. Not good.