neurophysik/jitcxde_common

Common code for JiTC*DE

jitcxde_common's Issues

Bug in the `conditional' routine

Hello, it's me again.

I found a bug in the conditional routine (from jitcxde_common). The description says the function returns value_if if observable is higher than threshold. However, the returned value is
return value_if+(1+tanh((observable-threshold)/width))/2*(value_else-value_if)

As you see, if observable is higher than threshold, then tanh > 0 and the returned value is actually value_else. The solution would be either to change the description, or change the last part of the returned value to (value_if-value_else).

Thanks!

Table look-up

Hey again :)
Not sure whether this should come here or to specific jitc*de I am using, but I feel it is a general question / problem.

So, I would need to make a table look-up as part of my dynamics. The thing is I am using a model (mean-field approximation of spiking neuron model) which uses pre-computed quantities in order to avoid solving Fokker-Planck equation each time step. So this table "converts" mean and standard deviation of membrane potential into firing rate. You can imagine it as

x_idx = int(<some function of mean potential>)
y_idx = int(<some function of std potential>)
firing_rate = table[x_idx, y_idx]

The problem is, of course, that the indices x_idx and y_idx are symbolic. Just to make a point x_idx is

floor(14.0*(-0.5 + sqrt(2.25 + 472.392*current_y(9)/(2.0 + 20.0*(1.0 + del_fr_exc_exc)) + 2178.0*current_y(10)/(5.0 + 20.0*(1.0 + del_fr_exc_inh)))))

So what I tried:

symengine.Lambdify

the code is roughly as this:

def compute_from_cascade(..., x_idx, y_idx, dx_idx, dy_idx)
        table_se = sympy.MatrixSymbol("table_se", *table.shape)
        expr = (
            table_se[y_idx, x_idx] * (1 - dx_idx) * (1 - dy_idx)
            + table_se[y_idx, x_idx + 1] * dx_idx * (1 - dy_idx)
            + table_se[y_idx + 1, x_idx] * (1 - dx_idx) * dy_idx
            + table_se[y_idx + 1, x_idx + 1] * dx_idx * dy_idx
        )
        func = se.Lambdify(table_se, expr)
        return func(table)

so I need to define Matrix as a symbol and lambdify it in order to get correct indices for a table lookup. However, this is not working as

symengine.lib.symengine_wrapper.SympifyError: sympy2symengine: Cannot convert 'table_se[0, 0]' (of type <class 'sympy.matrices.expressions.matexpr.MatrixElement'>) to a symengine type.

Then I found out that MatrxSymbol is not implemented in symengine and nobody knows if and when they'd implement it (I've spent an afternoon searching their github issues)

Use sympy's fallback symbols y and t

so the other option was to use "slow" sympy (still faster I guess than pure python) and instead of

from jitcdde import t as time_vector
from jitcdde import y as state_vector_dde

I did

from jitcdde.sympy_symbols import t as time_vector
from jitcdde.sympy_symbols import y as state_vector_dde

unfortunately, this is not working as this time I am getting

TypeError: cannot determine truth value of Relational

unrelated to the table lookup and this is an earlier error, which means that sympy is not really fully compatible with symenigne since symengine has no problems in determining the truth value of Relational...

However, the problem is that in order to work this with sympy I would need to rewrite almost all integration logic which then would become unusable with symengine... Since this is the only model I am having problem implementing in jitc*de I really do not want to rewrite the core logic to be compatible with sympy, I am more than happy with symengine.

I guess you're not an expert on symengine but the main question is whether you have some idea how you would do symbolic table lookup, given that symengine does not implement 2D symbolic arrays? Should I somehow approximate the table with a function?

Thanks a lot!
N.

May I contribute your package to conda-forge?

Dear neurophysik,

Hello neurophysik, my name is JunSeokCha and I am a student. I found that your package is on pypi but not on conda-forge. May I submit your package to conda-forge? Would you allow me to add jitcxde, jitcode, jitcdde, jitcsde to conda-forge?

If you would like me to do so, then who would you like to be the maintainer of the conda-forge feedstock for your package?

The job of the maintainer of the feedstock is to:
Keep the feedstock updated by merging eventual maintenance PRs from conda-forge’s bots.
Keep the feedstock on par with new releases of the source package by:
Bumping the version number and checksum.
Making sure that the feedstock’s requirements stay accurate.
Make sure the test requirements match those of the updated package.
Answer eventual questions about the package on the feedstock issue tracker.

If you would like to be the maintainer of the conda-forge feedstock, then I will submit your name as the maintainer of feedstock in the conda-forge recipe. If you would not like to be the maintainer of the conda-forge feedstock, then I will submit my name on the conda-forge recipe as maintainer of feedstock.

Sincerely,
JunSeokCha