Add predefined timestep() function
#1100
Conversation
clinssen
changed the title
timestep() function.timestep() function
clinssen
force-pushed
the
resolution_to_timestep
branch
from
139428f
to
ed55a2c
Compare
|
|
||
| onCondition(V_m >= V_peak): | ||
| onCondition(V_m >= V_peak and refr_t <= 0 ms): |
There was a problem hiding this comment.
Here and in all other neuron models with a similar construct, I'd swap the order of conditions:
onCondition(refr_t <= 0 ms and V_m >= V_peak)
| neuron like dynamics interacting by point events is described in | ||
| [1]_. A flow chart can be found in [2]_. | ||
|
|
||
| This model differs from ``iaf_psc_delta_fixed_timestep`` in that it |
There was a problem hiding this comment.
This is certainly a typo and should read:
This model differs from ``iaf_psc_delta`` ...
| spike | ||
|
|
||
| update: | ||
| if is_refractory: |
There was a problem hiding this comment.
This could be simplified by removing the boolean state variable is_refractory, and replacing
onCondition(not is_refractory)
by
onCondition(refr_t <= 0 ms)
(and similarly for all other refractoriness conditions). With this change, this version of the model becomes more similar to the version using an ODE for the refractoriness counter. In fact, only the way refr_t is updated needs to be changed. To expose the differences between the two versions, I'd keep everything else identical.
|
|
||
| emit_spike() | ||
|
|
||
| onCondition(is_refractory and refr_t <= resolution() / 2): |
There was a problem hiding this comment.
Is this block really necessary?
| @@ -260,13 +260,13 @@ In order to model an absolute refractory state, in which the neuron cannot fire | |||
| update: | |||
| if is_refractory: | |||
| # neuron is absolute refractory, do not evolve V_m | |||
| refr_t -= resolution() | |||
There was a problem hiding this comment.
Shouldn't the entire section "Implementing refractoriness" be updated to explain the new way of handling refractoriness?
Add a
timestep()function so models can be formulated in more generic terms than using theresolution()function. The latter assumes a fixed simulation timestep; the former is compatible with timesteps of arbitrary size (with subsequent steps potentially being of different size).Existing time-based functions (
resolution()andsteps()) will remain, in order to be able to formulate models optimised for fixed-timestep simulators. This is also necessary to get a numerically identical result with NEST for at least one of the tests (stdp triplet synapse test). A warning will be emitted during model validation with a recommendation to usetimestep()where possible.New refractoriness mechanisms are implemented based on a countdown ODE (thanks to Tom Tetzlaff and Markus Diesmann!) This means that neither
resolution()nortimestep()needs to be used any longer in the neuron models for the refractoriness mechanism. (See discussion in #988).TODO