like in neuron

like neuron

the idea is, given some neuron or network, one should be able to tune the parameters of this model so that some target behaviour is reached.

https://codecov.io/

one for getting the full state back, and another for getting just V and Ca -- which will be much faster because it doesn't have to poll the full state every iteration

and it has to be the first compartment.

using the compiler specified by the inner folder name

xolotl will introduce type of constructing models using what I call "relational parameters". Instead of parameters being scalars, they can be functions of other parameters. This means that:

1. Changing them by hand breaks the relationship they have to other parameters
2. They cannot be manipulated, since they are functions of other parameters
3. Changing other parameters automatically updates these parameters

For now, we're going to avoid the problem of nested relational parameters (so if y = f(x), where x and y are parameters, then there can be no z: z = g(y))

Currently, AB.NaV.gbar is mapped onto gbar_NaV_AB using struct2vec

Instead, we should map it onto AB_NaV_gbar since this is much easier to reverse.

The Liu/Prinz one is a simplification of this, and does not scale

one that doesn't need mex

like a scope

file names

we should be able to know the value of every single dynamic variable at every single time step

and in addition some helpful things like E_Ca

into outputs in MATLAB, in the same format that the input was recieved

so that the MATLAB and C++ objects can be kept in sync

names of classes in C++ code -- like CaS or Glutamatergic

since hash only includes the short name of the channel, it cannot differentiate from channels from different sources.

for example, if a model uses Prinz channels, it has the same hash as a model that uses Liu channels.

then we can also include Prinz gating functions

progressively setting gbar of all conductances in a neuron causes voltage integration to explode --

because sigma_gE = 0, and sigma_g =0, and 0/0 is not defined, and yeilds a NaN

it's worse -- because the voltage of this step is NaN, the integration is ruined for all time steps in the future, and also for all compartments/neurons connected to this.

and different types of calcium equations, etc.

so that we can quickly check if binaries match the current state

then we would have to label binaries with the hash too

but where? perhaps in the full_cond_state, at the very end. this would change the dimensions of this matrix everywhere...

/ and \ need to be replaced with oss

that converts a xolotl object into a cell array of vectors that can be fed as arguments into the linked binary, and a corresponding cell array of strings with parameter/value names

which means xolotl cannot update synapses to reflect actual state.

shouldn't sigma_g include the g from synapses?

individually, and test them