multiple parameter values with data_set()

[kyralianaka]

I've been having trouble using data_set to set the parameters of multiple synapse or cells at once as shown in this little example:

comp = jx.Compartment()
branch = jx.Branch(comp, nseg=1)
cell = jx.Cell([branch], parents=[-1])
network = jx.Network([cell, cell, cell])

jx.fully_connect(network.cell("all"), network.cell("all"), IonotropicSynapse())

n_edges = len(network.edges)
params = [{"IonotropicSynapse_gS": jnp.ones(n_edges) * 0.001}]

# n_cells = len(network.nodes)
# params = [{"radius": jnp.ones(n_cells)}]

network.cell(0).record()

def simulate(params):

    # Set the parameters with data_set
    pstates = None
    pstates = network.IonotropicSynapse.data_set("IonotropicSynapse_gS", list(params[0].values())[0], pstates)
    #pstates = network.data_set("radius", list(params[0].values())[0], pstates)

    v = jx.integrate(network, t_max=10, param_state=pstates)
    return v

s = simulate(params)

The issue ends up being that the shape of the indices array is incompatible with the shape of the values array (they are transposes). In comments in Jaxley (base.py lines 573-577), I see that inds is supposed to be shape (num_params, num_comps_per_param), but then this becomes (1, num_synapses) or (1, num_cells) in the example above, which is incompatible with the values array which gets projected to (num_params, 1), where num_params here is the number of edges or cells.

I have tried to adjust this on the user end, and maybe there's a way to do that that I haven't tried yet, but I think it would be nice to make the code above work. I don't have the time in the next two weeks, maybe later, but of course if someone else wants to have a look please do:)

[kyralianaka]

I will do some more experimenting, but I think just changing the following two lines would allow both specifying single or multiple parameter values (one for each compartment, which conveniently works out to each cell when each cell is one compartment). In _data_set() line 396: "val": jnp.atleast_1d(jnp.asarray(val)), -->"val": jnp.atleast_2d(jnp.asarray(val)),, same as is done with the indices, and then in get_all_parameters(), called by integrate(), line 577: params[key] = params[key].at[inds].set(set_param[:, None]) --> params[key] = params[key].at[inds].set(set_param). I think it is then intuitive if the values and the indices can have the same shape. I'll look into it a bit more though.

[kyralianaka]

Can now be done using the select method to define indices from #447