Struct BatchCollector 

pub struct BatchCollector { /* private fields */ }

Collects neural predicate calls for batched evaluation.

During proof search, calls are accumulated. Before neural evaluation, they are grouped by network name for efficient batched forward passes.

Implementations

impl BatchCollector

pub fn new() -> Self

Create a new empty collector.

pub fn with_capacity(capacity: usize) -> Self

Create a collector with pre-allocated capacity.

pub fn add(&mut self, call: NeuralCall)

Add a neural call to the collector.

pub fn collect(&self) -> HashMap<String, Vec<&NeuralCall>>

Group calls by network name for batched evaluation.

Returns a map from network name to list of calls for that network.

pub fn indices_for_network(&self, network: &str) -> Vec<usize>

Get all input indices for a specific network.

These indices can be used to gather inputs from the tensor source into a batched tensor for the forward pass.

pub fn network_names(&self) -> Vec<String>

Get the names of all networks that have been called.

pub fn call_count_for_network(&self, network: &str) -> usize

Get the number of calls for a specific network.

pub fn total_input_count(&self) -> usize

Get the total number of input indices across all calls.

pub fn len(&self) -> usize

Total number of calls.

pub fn is_empty(&self) -> bool

Check if the collector is empty.

pub fn clear(&mut self)

Clear all collected calls.

pub fn iter(&self) -> impl Iterator<Item = &NeuralCall>

Iterate over all calls.

pub fn take(&mut self) -> Vec<NeuralCall>

Take ownership of all calls.

Trait Implementations

impl Default for BatchCollector

fn default() -> Self

Returns the “default value” for a type.