Record and fadvance instead of run

When Python is the interpreter, what is a good
design for the interface to the basic NEURON
concepts.

Moderator: hines

Post Reply
fkolbl
Posts: 2
Joined: Thu Jul 16, 2020 10:02 am

Record and fadvance instead of run

Post by fkolbl »

Hi,

I try to run a simulation step by step, using 'fadvance()' with a loop instead of 'run()'. Here is a minimal code of what I am trying to do :

Code: Select all

import neuron
from neuron import h
h.load_file('stdrun.hoc')
import numpy as np

v_init = -65

axon1 = h.Section(name='U_axon')
axon1.L = 5000
axon1.diam = 1
axon1.nseg = 100
axon1.insert('pas')
axon1.insert('hh')
	
h.define_shape()

h.tstop = 2
h.celsius = 33       	
h.finitialize(v_init)  
h.v_init=v_init
h.dt = 0.001
h.fcurrent()

timeVector = h.Vector()
timeVector.record(h._ref_t)
#neuron.h.run()
while h.t < h.tstop:
	h.fadvance()
t = np.array(timeVector)
print(t)


by the end of the example there is nothing recorded in the variable 't', it is an empty numpy array...if I comment the while and the content of the loop to replace it with the 'run', I get the correct time vector... What did I miss so that the steps (and all possible other variables) are not stored by using fadvance ?

I thank you in advance for your time and consideration.
all the best
ramcdougal
Posts: 267
Joined: Fri Nov 28, 2008 3:38 pm
Location: Yale School of Public Health

Re: Record and fadvance instead of run

Post by ramcdougal »

Vectors must be initialized after calling record before advancing. Your other code works with h.run() because a run is basically an h.finitialize(h.v_init) combined with an h.continuerun(h.tstop).

The solution is to declare your timeVector before the h.finitialize call, i.e. something like:

Code: Select all

timeVector = h.Vector()
timeVector.record(h._ref_t)
h.finitialize(v_init)
# ...
while h.t < h.tstop:
    h.fadvance()
A few other suggestions:
  • You probably don't need to do each individual h.fadvance() separately. As written, your while loop is equivalent to h.continuerun(h.tstop).
  • Also, note that every vector has a .as_numpy() method which returns a numpy array mapping to the same memory (so changes to one are seen in the other). If you need a copy, do what you're doing, but you almost never need a copy (and, honestly, Vector objects mostly just work where a numpy array would work, especially in 7.7+).
  • It's better to use an odd value for nseg, if for no other reason than because it makes the center segment axon(0.5) unambiguous.
  • Consider using e.g. from neuron.units import mV, ms, um and using those values to make the meaning of the numbers clearer (e.g. h.tstop = 2 * ms, v_init = -65 * mV, axon1.L = 5000 * um)
  • Assuming you're running 7.7+, you can combine the record and declaration onto one line: timeVector = h.Vector().record(h._ref_t)
fkolbl
Posts: 2
Joined: Thu Jul 16, 2020 10:02 am

Re: Record and fadvance instead of run

Post by fkolbl »

Hello,
thank you very much for your fast and clear answer, all the comment on the code that will be really helpful. It seems to me that h.continuerun() is what I will use. I am avoiding run since I add an extracellular potential which has a relatively high frequency and when using play I have to compute a large array which fills the RAM for almost nothing... basically I will make a loop with a time of run of the stimulus sampling frequency. I was initially inspired by a code in hoc doing almost the same with fadvance on modelDB.

all the best and again thank you !
Post Reply