Struct NetworkConfig 

#[non_exhaustive]
pub struct NetworkConfig {
    pub name: String,
    pub batching: bool,
    pub k: Option<usize>,
    pub det: bool,
    pub cache_enabled: bool,
    pub cache_size: usize,
    pub arity: Option<usize>,
    pub arg_sorts: Option<Vec<i64>>,
    pub artifact_hash: Option<String>,
}

Configuration for registering a neural network.

This mirrors the DeepProbLog register_network options.

Fields (Non-exhaustive)

Non-exhaustive structs could have additional fields added in future. Therefore, non-exhaustive structs cannot be constructed in external crates using the traditional Struct { .. } syntax; cannot be matched against without a wildcard ..; and struct update syntax will not work.

name: String

Unique name identifying this network (must match nn() declarations)

batching: bool

Whether to batch inputs for efficient GPU processing. When true, multiple queries are grouped into a single forward pass.

k: Option<usize>

Top-k sampling: if Some(k), only consider the top k outputs. Useful for large output spaces where most classes have near-zero probability.

det: bool

Deterministic mode: use argmax instead of probabilistic sampling. Useful for debugging and when you want reproducible results.

cache_enabled: bool

Whether to cache network outputs. Caching avoids redundant forward passes for repeated inputs.

cache_size: usize

Maximum number of entries in the output cache.

arity: Option<usize>

Number of arguments the network’s declared predicate(s) take, as stated by the caller at registration time. None means unstated. When present, this is validated against every nn/4 declaration bound to this network name (see register_network), so a mismatch between what the caller claims and what the program actually declared is rejected rather than trusted.

arg_sorts: Option<Vec<i64>>

Per-argument catalog sort ids, in declared-argument order. Carried opaquely: the registry does not interpret or validate these against the program, it only stores them so callers (e.g. slot-matching enumeration) can sort-match candidate slots by id. Validated at the pyo3 boundary (register_network) to be Python ints, with bool explicitly excluded even though isinstance(True, int) holds.

artifact_hash: Option<String>

Content hash of the registered artifact (module weights). Carried opaquely, like arg_sorts: a retrained network is expected to mint a new hash, so this field is the identity of this registration, not a claim the registry checks.

Implementations

impl NetworkConfig

pub fn default(name: &str) -> Self

Create a default configuration for a network with the given name.

Default settings:

• batching: true
• k: None (consider all outputs)
• det: false (probabilistic mode)
• cache_enabled: true
• cache_size: 10000

pub fn deterministic(name: &str) -> Self

Create a configuration for a deterministic network.

pub fn with_top_k(name: &str, k: usize) -> Self

Create a configuration with top-k sampling.

pub fn batching(self, enabled: bool) -> Self

Builder method to set batching.

pub fn k(self, k: Option<usize>) -> Self

Builder method to set top-k.

pub fn det(self, det: bool) -> Self

Builder method to set deterministic mode.

pub fn cache(self, enabled: bool, size: usize) -> Self

Builder method to set cache.

Trait Implementations

impl Clone for NetworkConfig

fn clone(&self) -> NetworkConfig

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for NetworkConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.