ENH: Add somato example #35
Conversation
Codecov Report
@@ Coverage Diff @@
## master #35 +/- ##
=======================================
Coverage 76.17% 76.17%
=======================================
Files 11 11
Lines 1356 1356
=======================================
Hits 1033 1033
Misses 323 323Continue to review full report at Codecov.
|
jasmainak
force-pushed
the
somato_example
branch
from
1d60523
to
3ffedbd
Compare
|
@rythorpe @blakecaldwell do you want to try this example out and review the code? I updated it with the parameters that I got from Ryan. I have two questions:
Let me know. |
examples/plot_simulate_somato.py
Outdated
| params_fname = op.join(mne_neuron_root, 'param', 'default.json') | ||
| params = Params(params_fname) | ||
|
|
||
| params.update({ |
There was a problem hiding this comment.
@rythorpe I would suggest sharing the params as a "diff". That is, what do we need to change in the defaults.param or defaults.json to get the waveform. That way, it's a bit more clear to the users (and to me!) what are the canonical parameters that can be somewhat fixed and which ones are actually critical to producing the waveform.
There was a problem hiding this comment.
The waveform above is a bit different from the one produced by HNN. Here's what I get:
We could do that, though we would need to come to some sort of consensus regarding what we consider to be the "default" parameters. Were you thinking of just using the default.param file referenced in the HNN tutorial?
There was a problem hiding this comment.
Hmm ... the difference is probably because you have multiple trials. I made n_trials = 1 because we're still trying to figure out what's the best way to do the parallel processing with @blakecaldwell. There is an open pull request for that. There seems to be a fair amount of noise at the level of single trials. Did you try and see what it looks like for n_trials = 1? It should be the same because the random seeds etc have not changed.
Yes, currently I am just using the default.param from the HNN tutorials. But we could change this of course. Just want to change one thing at a time so we know where the differences are coming from (between softwares, models, examples etc.), if any.
There was a problem hiding this comment.
@jasmainak What are the units on the y-axis?
There was a problem hiding this comment.
should be the same as yours, that is nAm with a scaling factor of 30. Where do you get the 6000 cells from? It's not in your param file ...
There was a problem hiding this comment.
The 6000 (pyramidal) cells comes from the number of pyramidal cells per layer (Local Network Parameters->Cells->[Num Pyr Cells] for the x and y-dimensions) times the number of layers (2, corresponding to L2/3 and L5 Pyr) times the scaling factor. For a scaling factor of 30, this is 10x10x2x30=6,000. For the default.param simulation, it is 10x10x2x3000=600,000 and reflects a greater number of aligned pyramidal cells contributing to the net current dipole.
There was a problem hiding this comment.
oh I see, thanks for explaining. The plots look very similar to me -- the only difference being that you have smoothing over many trials. I think the next logical step would be to get back to #40 and see what can be done to make it work in parallel for multiple trials.
There was a problem hiding this comment.
There is still something slightly different between our respective single-trial simulations. In particular, the initial upward deflection isn't present in your version. Here's what I get (note that the solid black line is the first single trial and the dashed line is the 100-trial average):
|
@rythorpe I am merging this for now to move forward. I will raise an issue to dig into this. It will require more work. I need to think of a good way to debug. Thanks for helping! |
cc @rythorpe
this gives the following figure:
we can probably do a better job, but it's already a start.