|
|
--- |
|
|
tags: |
|
|
- pytorch |
|
|
- code-search |
|
|
- ablation-study |
|
|
--- |
|
|
|
|
|
# reranker - Procedural Code Search |
|
|
|
|
|
This is the `reranker` variant from the ablation study for the Structural Encoder V2. |
|
|
|
|
|
## Model Details |
|
|
- **Variant**: reranker |
|
|
- **Dataset**: POJ-104 |
|
|
- **Architecture**: |
|
|
- `only_graph`: Pure Structural Encoder (R-GNN only) |
|
|
- `concat`: GraphCodeBERT + R-GNN (Concatenation Fusion) |
|
|
- `only_graph`: Pure Structural Encoder (R-GNN only) |
|
|
- `concat`: GraphCodeBERT + R-GNN (Concatenation Fusion) |
|
|
- `v2`: GraphCodeBERT + R-GNN (Gated Fusion) |
|
|
- `reranker`: Stage B Procedural Similarity Reranker (Logistic Regression) |
|
|
|
|
|
## Usage |
|
|
This model requires the custom code `structural_encoder_ablation.py` and `structural_encoder_v2.py`. |
|
|
|
