API reference
Connectivity builders
Signatures, parameters, return contracts, and source for connectivity builders.
Overview
PyNN-style builders wire one population of n neurons, optionally restricted to pre source and post target IDs, and return an ordinary validated ExplicitConnectome for create_population. Each edge takes its source neuron’s role (neuron_roles, +1 or -1; None makes every neuron excitatory) and a destination channel drawn uniformly from the checkpoint channels with that role (channel_roles), unless explicit channels are given. Efficacies and delays are constants, one value per edge, or callables fn(sources, targets, generator); delays are integer native milliseconds of at least MIN_DELAY_MS = 4, the Lite P4 recurrent minimum. One seeded generator is drawn in a fixed order (topology, channels, efficacies, delays), so a seed reproduces the exact graph, and edge IDs follow the returned edge order. Shared keyword arguments are neuron_roles, channel_roles, channels, efficacies, and delays. The builders are exported from axosim; MIN_DELAY_MS is imported from axosim.connectors.
Source revision: caef7be03abd. Public export index.
all_to_all
axosim.connectors.all_to_all(n: int, *, seed: int=0, allow_self: bool=False, pre=None, post=None, **kwargs) -> ExplicitConnectome
Connect every pre neuron to every post neuron once (source-major).
Parameters
nint- required. Positive population size; neuron IDs are 0..n-1.
seedint- keyword-only, default=0. Integer seed of the one generator drawn for topology, channels, efficacies, and delays, in that order; equal seeds give identical graphs.
allow_selfbool- keyword-only, default=False. Keep edges from a neuron to itself.
preunannotated- keyword-only, default=None. Source neuron IDs; None uses all n neurons.
postunannotated- keyword-only, default=None. Target neuron IDs; None uses all n neurons.
kwargsunannotated- variadic. Shared edge options: neuron_roles ((n,) +1/-1, default all excitatory), channel_roles (the checkpoint's (input_dim,) +1/-1 vector, required unless channels is given), channels, efficacies (default 1.0), and delays (default 4 ms). Values are a constant, one per edge, or a callable fn(sources, targets, generator).
Returns
connectomeExplicitConnectome- One edge per (pre, post) pair, source-major, without self-edges unless allow_self.
one_to_one
axosim.connectors.one_to_one(n: int, *, seed: int=0, pre=None, post=None, **kwargs) -> ExplicitConnectome
Connect pre[i] to post[i]; the default pre == post gives autapses.
Parameters
nint- required. Positive population size; neuron IDs are 0..n-1.
seedint- keyword-only, default=0. Integer seed of the one generator drawn for topology, channels, efficacies, and delays, in that order; equal seeds give identical graphs.
preunannotated- keyword-only, default=None. Source neuron IDs; None uses all n neurons.
postunannotated- keyword-only, default=None. Target neuron IDs; None uses all n neurons.
kwargsunannotated- variadic. Shared edge options: neuron_roles ((n,) +1/-1, default all excitatory), channel_roles (the checkpoint's (input_dim,) +1/-1 vector, required unless channels is given), channels, efficacies (default 1.0), and delays (default 4 ms). Values are a constant, one per edge, or a callable fn(sources, targets, generator).
Returns
connectomeExplicitConnectome- Edges pre[i] -> post[i]; pre and post must have equal length.
fixed_probability
axosim.connectors.fixed_probability(n: int, p: float | Callable[[torch.Tensor, torch.Tensor], torch.Tensor], *, seed: int=0, allow_self: bool=False, pre=None, post=None, **kwargs) -> ExplicitConnectome
Keep each pre -> post pair independently with probability p.
Parameters
nint- required. Positive population size; neuron IDs are 0..n-1.
pfloat | Callable[[torch.Tensor, torch.Tensor], torch.Tensor]- required. Connection probability in [0, 1], or a callable p(sources, targets) returning one probability per candidate pair, evaluated in chunks; for example a distance-dependent kernel over neuron positions.
seedint- keyword-only, default=0. Integer seed of the one generator drawn for topology, channels, efficacies, and delays, in that order; equal seeds give identical graphs.
allow_selfbool- keyword-only, default=False. Keep edges from a neuron to itself.
preunannotated- keyword-only, default=None. Source neuron IDs; None uses all n neurons.
postunannotated- keyword-only, default=None. Target neuron IDs; None uses all n neurons.
kwargsunannotated- variadic. Shared edge options: neuron_roles ((n,) +1/-1, default all excitatory), channel_roles (the checkpoint's (input_dim,) +1/-1 vector, required unless channels is given), channels, efficacies (default 1.0), and delays (default 4 ms). Values are a constant, one per edge, or a callable fn(sources, targets, generator).
Returns
connectomeExplicitConnectome- Independently kept pairs in source-major order, without parallel edges.
fixed_indegree
axosim.connectors.fixed_indegree(n: int, k: int, *, seed: int=0, allow_self: bool=False, with_replacement: bool=False, pre=None, post=None, **kwargs) -> ExplicitConnectome
Give every post neuron exactly k sources drawn uniformly from pre.
Parameters
nint- required. Positive population size; neuron IDs are 0..n-1.
kint- required. Nonnegative number of sources per post neuron.
seedint- keyword-only, default=0. Integer seed of the one generator drawn for topology, channels, efficacies, and delays, in that order; equal seeds give identical graphs.
allow_selfbool- keyword-only, default=False. Keep edges from a neuron to itself.
with_replacementbool- keyword-only, default=False. Draw sources with replacement; repeated draws become parallel edges. Without replacement the k sources of a target are distinct.
preunannotated- keyword-only, default=None. Source neuron IDs; None uses all n neurons.
postunannotated- keyword-only, default=None. Target neuron IDs; None uses all n neurons.
kwargsunannotated- variadic. Shared edge options: neuron_roles ((n,) +1/-1, default all excitatory), channel_roles (the checkpoint's (input_dim,) +1/-1 vector, required unless channels is given), channels, efficacies (default 1.0), and delays (default 4 ms). Values are a constant, one per edge, or a callable fn(sources, targets, generator).
Returns
connectomeExplicitConnectome- Exactly k edges into every post neuron, ordered by target, then source.
from_list
axosim.connectors.from_list(n: int, edges, *, efficacies: EdgeValues | None=None, delays: EdgeValues | None=None, seed: int=0, **kwargs) -> ExplicitConnectome
Build from rows (source, target[, efficacy[, delay]]) in the given order.
Parameters
nint- required. Positive population size; neuron IDs are 0..n-1.
edgesunannotated- required. Rows (source, target[, efficacy[, delay]]) in edge order; integer neuron IDs in [0, n).
efficaciesEdgeValues | None- keyword-only, default=None. Nonnegative edge magnitudes: a constant, one value per edge, or a callable; exclusive with an efficacy column. None means 1.0.
delaysEdgeValues | None- keyword-only, default=None. Integer native delays of at least 4 ms: a constant, one value per edge, or a callable; exclusive with a delay column. None means 4 ms.
seedint- keyword-only, default=0. Integer seed of the one generator drawn for topology, channels, efficacies, and delays, in that order; equal seeds give identical graphs.
kwargsunannotated- variadic. Shared edge options: neuron_roles ((n,) +1/-1, default all excitatory), channel_roles (the checkpoint's (input_dim,) +1/-1 vector, required unless channels is given), channels, efficacies (default 1.0), and delays (default 4 ms). Values are a constant, one per edge, or a callable fn(sources, targets, generator).
Returns
connectomeExplicitConnectome- The listed edges in the given order.
concatenate
axosim.connectors.concatenate(*connectomes: ExplicitConnectome) -> ExplicitConnectome
Join graphs edge-wise; part i’s edge IDs follow those of parts < i.
Parameters
connectomesExplicitConnectome- variadic. ExplicitConnectome parts; the outgoing edges of each neuron must share one role across parts.
Returns
connectomeExplicitConnectome- All edges joined; the edge IDs of part i follow those of the parts before it.