Package com.helix.core.reorder

Class NeuralAstReorderingPolicy

java.lang.Object

com.helix.core.reorder.NeuralAstReorderingPolicy

All Implemented Interfaces:

ReorderingPolicy

public class NeuralAstReorderingPolicy
extends Object
implements ReorderingPolicy

Neural AST Reordering Policy utilizing the embedded ast_reorder_policy.onnx model.

Encodes candidate AST node statistics into an 82-dimensional float observation vector, evaluates candidate action logits via OnnxSessionPool, and applies action masking to decode an optimal permutation sequence.

Seamlessly falls back to RatioSortPolicy if the neural policy model or runtime session pool is unavailable.

Field Summary

Fields

Modifier and Type	Field	Description
static final String	INPUT_TENSOR_NAME
static final int	MAX_CANDIDATE_NODES
static final int	OBSERVATION_DIM
static final String	OUTPUT_TENSOR_NAME
static final String	POLICY_MODEL_NAME

Constructor Summary

Constructors

Constructor	Description
NeuralAstReorderingPolicy()
NeuralAstReorderingPolicy(OnnxSessionPool sessionPool)

Method Summary

Modifier and Type	Method	Description
List<Integer>	determineOrder(List<NodeStats> nodes)	Determines the optimal evaluation order of candidate AST clauses.
void	encodeObservation(float[] obs, List<NodeStats> nodes, int n, int step)	Encodes candidate node statistics into an 82-dimensional float observation vector.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Details

POLICY_MODEL_NAME

public static final String POLICY_MODEL_NAME

See Also:

• Constant Field Values

INPUT_TENSOR_NAME

public static final String INPUT_TENSOR_NAME

See Also:

• Constant Field Values

OUTPUT_TENSOR_NAME

public static final String OUTPUT_TENSOR_NAME

See Also:

• Constant Field Values

MAX_CANDIDATE_NODES

public static final int MAX_CANDIDATE_NODES

See Also:

• Constant Field Values

OBSERVATION_DIM

public static final int OBSERVATION_DIM

See Also:

• Constant Field Values

Constructor Details

NeuralAstReorderingPolicy

public NeuralAstReorderingPolicy()

NeuralAstReorderingPolicy

public NeuralAstReorderingPolicy(OnnxSessionPool sessionPool)

Method Details

determineOrder

public List<Integer> determineOrder(List<NodeStats> nodes)

Description copied from interface: ReorderingPolicy

Determines the optimal evaluation order of candidate AST clauses.

Specified by:

determineOrder in interface ReorderingPolicy

Parameters:

nodes - list of node statistics corresponding to candidate clauses

Returns:

list of 0-based indices representing the new evaluation sequence

encodeObservation

public void encodeObservation(float[] obs,
 List<NodeStats> nodes,
 int n,
 int step)

Encodes candidate node statistics into an 82-dimensional float observation vector.

Parameters:

obs - 82-dimensional target float array

nodes - candidate nodes

n - number of active candidate nodes

step - current decoding step (0 to n - 1)