Procrustes alignment

.gitignore

@@ -58,6 +58,9 @@ coverage.xml
 .hypothesis/
 .pytest_cache/
 cover/
+cache/
+WDC_scripts/linking_scripts/AML/AML_v3.2/data
+WDC_scripts/linking_scripts/AML/AML_v3.2/logs
 
 # Translations
 *.mo
@@ -175,4 +178,13 @@ knowledge_graph.nt
 wikidata_classes_by_language.csv
 wikidata_classes.nt
 *.xml
-*.nt
+*.nt
+*.nq
+*.sh
+!WDC_scripts/linking_scripts/limes/extract_configs_with_props_use_limes.sh
+
+WDC_scripts/linking_scripts/limes/datasets/
+Test_dataset/
+*.json
+checkpoints/
+Experiments/

Alignment/README.md

@@ -0,0 +1,36 @@
+# Procrustes Alignment Module
+
+## Overview
+
+This module implements Procrustes alignment, a mathematical method for aligning embeddings from two knowledge graphs into a shared vector space. By aligning embeddings, entities and relations from different datasets can be directly compared, improving tasks such as link prediction and knowledge graph integration. The module supports alignment dictionaries from tools like LIMES or manually curated datasets, ensuring scalability and interpretability.
+
+
+## Key Features
+
+**Dynamic Loading of Embeddings and Mappings:** Automatically identifies and loads embeddings, entities, and relations from dataset folders.
+**Flexible Alignment Dictionaries:** Accepts both pre-aligned dictionaries or LIMES-generated sameAs links for entity alignment.
+**Bidirectional Procrustes Alignment:** Forward direction: Aligns Dataset A to Dataset B. Reverse direction: Aligns Dataset B to Dataset A.
+**Integration of Non-Aligned Entities:**  Processes non-aligned embeddings and integrates them into the aligned space.
+**Unified Embedding Representation:**  Combines aligned and non-aligned embeddings into a single representation for downstream tasks.
+**Output Models:** Saves aligned embeddings and trained models, ready for tasks like link prediction, clustering, and graph completion.
+
+
+## Applications
+
+**Cross-Knowledge Graph Tasks:** Facilitates link prediction and other tasks requiring compatibility across datasets.
+**Knowledge Graph Integration:** Aligns embeddings to merge multiple knowledge graphs seamlessly.
+**Scalable and Explainable AI:** Maintains interpretability while being computationally efficient for large-scale datasets.
+
+
+## Inputs
+
+**Embedding Folder:** 
+- Directory containing subfolders for the two datasets.
+- Each dataset subfolder must include:
+- model.pt: Trained model file with embeddings.
+- entity_to_idx.p: Entity-to-index mapping file.
+- relation_to_idx.p: Relation-to-index mapping file.
+
+**Alignment Dictionaries:** 
+- Pre-Aligned Dictionary: A file (e.g., .nt, .ttl, .txt) containing known alignments between entities in the two datasets.
+- LIMES Alignment Dictionary: A file generated by LIMES that provides sameAs links for alignment.