Pipeline improvements: finalizing asset similarity

backend/api/python_endpoints/asset_endpoints.py

@@ -41,3 +41,7 @@ def add_keyword(asset_iri: str, keyword: str):
         keyword (str): _description_
     """
     ASSETS_DAO.add_keyword(asset_iri=asset_iri, keyword=keyword)
+
+
+def get_keywords_set_for_asset(asset_iri: str):
+    return ASSETS_DAO.get_keywords_set_for_asset(asset_iri=asset_iri)

backend/api/python_endpoints/file_endpoints.py

@@ -162,10 +162,6 @@ def save_extracted_text(file_iri: str, text: str):
     SUPPL_FILE_DAO.save_extracted_text(file_iri=file_iri, text=text)
 
 
-def get_keywords_set_for_asset(asset_iri: str):
-    return SUPPL_FILE_DAO.get_keywords_set_for_asset(asset_iri=asset_iri)
-
-
 def reset_dimension_clusters():
     SUPPL_FILE_DAO.reset_dimension_clusters()
 
@@ -182,3 +178,7 @@ def add_file_to_dimension_cluster(file_iri: str, cluster_iri: str):
     SUPPL_FILE_DAO.add_file_to_dimension_cluster(
         file_iri=file_iri, cluster_iri=cluster_iri
     )
+
+
+def get_dimensions_cluster_for_asset(asset_iri: str):
+    return SUPPL_FILE_DAO.get_dimensions_cluster_for_asset(asset_iri=asset_iri)

backend/knowledge_graph/dao/AssetNodesDao.py

@@ -155,3 +155,38 @@ def add_keyword(self, asset_iri: str, keyword: str):
         )
 
         self.ps.graph_create(relationship)
+
+    def get_keywords_set_for_asset(self, asset_iri: str):
+        # File keywords
+        file_keywords_table = self.ps.graph_run(
+            "MATCH p=(a:"
+            + NodeTypes.ASSET.value
+            + ' {iri: "'
+            + asset_iri
+            + '"})-[r1:'
+            + RelationshipTypes.HAS_SUPPLEMENTARY_FILE.value
+            + "]->(t)-[r2:"
+            + RelationshipTypes.KEYWORD_EXTRACTION.value
+            + "]->(c) RETURN c.keyword"
+        ).to_table()
+
+        file_keyword_list = [keyword[0] for keyword in file_keywords_table]
+
+        # Asset keywords
+        asset_keywords_table = self.ps.graph_run(
+            "MATCH p=(a:"
+            + NodeTypes.ASSET.value
+            + ' {iri: "'
+            + asset_iri
+            + '"})-[r1:'
+            + RelationshipTypes.KEYWORD_EXTRACTION.value
+            + "]->(c) RETURN c.keyword"
+        ).to_table()
+
+        asset_keyword_list = [keyword[0] for keyword in asset_keywords_table]
+
+        # Combine
+        keyword_list = file_keyword_list
+        keyword_list.extend(asset_keyword_list)
+
+        return set(keyword_list)

backend/knowledge_graph/dao/SupplementaryFileNodesDao.py

@@ -206,23 +206,6 @@ def save_extracted_text(self, file_iri: str, text: str):
         node.update(extracted_text=text)
         self.ps.graph_push(node)
 
-    def get_keywords_set_for_asset(self, asset_iri: str):
-        keywords_table = self.ps.graph_run(
-            "MATCH p=(a:"
-            + NodeTypes.ASSET.value
-            + ' {iri: "'
-            + asset_iri
-            + '"})-[r1:'
-            + RelationshipTypes.HAS_SUPPLEMENTARY_FILE.value
-            + "]->(t)-[r2:"
-            + RelationshipTypes.KEYWORD_EXTRACTION.value
-            + "]->(c) RETURN c.keyword"
-        ).to_table()
-
-        keyword_list = [keyword[0] for keyword in keywords_table]
-
-        return set(keyword_list)
-
     def reset_dimension_clusters(self):
         self.ps.graph_run(
             f"MATCH (n:{NodeTypes.DIMENSION_CLUSTER.value}) DETACH DELETE n"
@@ -252,3 +235,19 @@ def add_file_to_dimension_cluster(self, file_iri: str, cluster_iri: str):
         )
 
         self.ps.graph_create(relationship)
+
+    def get_dimensions_cluster_for_asset(self, asset_iri: str):
+
+        cluster_matches = self.ps.repo_match(model=DimensionClusterNode).where(
+            "(:"
+            + NodeTypes.ASSET.value
+            + ' {iri: "'
+            + asset_iri
+            + '"})-[:'
+            + RelationshipTypes.HAS_SUPPLEMENTARY_FILE.value
+            + "]->()-[:"
+            + RelationshipTypes.PART_OF_DIMENSION_CLUSTER.value
+            + "]->(_)"
+        )
+
+        return cluster_matches.first()

similarity_pipeline/similarity_pipeline_5_cad_analysis.py

@@ -123,9 +123,9 @@ def similarity_pipeline_5_cad_analysis():
                 max_z = max(point[stl.Dimension.Z], max_z)
                 min_z = min(point[stl.Dimension.Z], min_z)
 
-        length = max_x - min_x
-        width = max_y - min_y
-        height = max_z - min_z
+        width = max_x - min_x
+        height = max_y - min_y
+        length = max_z - min_z
 
         properties_dict = {
             "length": float(length),

similarity_pipeline/similarity_pipeline_6_image_analysis.py

@@ -72,8 +72,8 @@ def similarity_pipeline_6_image_analysis():
         tmp_file_path = "./temporary_picture.jpg"
 
         with io.FileIO(tmp_file_path, "w") as tmp_file:
-            for pdf_line in file_stream:
-                tmp_file.write(pdf_line)
+            for file_line in file_stream:
+                tmp_file.write(file_line)
 
         logger.info("Detecting objects...")
 
@@ -108,7 +108,7 @@ def similarity_pipeline_6_image_analysis():
         logger.info("Extracting text from PDF...")
         text_encoded = textract.process(
             tmp_file_path,
-            method="tesseract",
+            method="tesseract-ocr",
             # language="eng",
         )
         text = str(text_encoded, "UTF-8")

similarity_pipeline/similarity_pipeline_7_asset_similarity.py

@@ -6,8 +6,6 @@
 from numpy import nan
 from numpy import linalg as linalg
 import numpy as np
-from sklearn.cluster import DBSCAN
-from sklearn.decomposition import PCA
 
 from backend.api.python_endpoints import asset_endpoints
 from backend.api.python_endpoints import timeseries_endpoints
@@ -16,7 +14,9 @@
     TimeseriesNodeFlat,
     TimeseriesValueTypes,
 )
-from similarity_pipeline.similarity_pipeline_status_manager import SimilarityPipelineStatusManager
+from similarity_pipeline.similarity_pipeline_status_manager import (
+    SimilarityPipelineStatusManager,
+)
 from util.log import logger
 
 # #############################################################################
@@ -33,7 +33,9 @@ def similarity_pipeline_7_asset_similarity():
     logger.info("Loading assets and related nodes...")
 
     # get assets flat (just for iris)
-    asset_nodes_flat: List[TimeseriesNodeFlat] = asset_endpoints.get_asset_nodes(deep=False)
+    asset_nodes_flat: List[TimeseriesNodeFlat] = asset_endpoints.get_asset_nodes(
+        deep=False
+    )
 
     # per asset: get connected ts-clusters, keywords... (seperated, so they can be easily weighted differently)
     timeseries_clusters = []
@@ -44,9 +46,17 @@ def similarity_pipeline_7_asset_similarity():
 
     keyword_sets = []
     for asset_node in asset_nodes_flat:
-        keyword_sets.append(
-            file_endpoints.get_keywords_set_for_asset(asset_iri=asset_node.iri)
+        keyword_set = asset_endpoints.get_keywords_set_for_asset(
+            asset_iri=asset_node.iri
+        )
+        # Handle dimension clusters like key-phrases, as they can only occur once per asset
+        cluster = file_endpoints.get_dimensions_cluster_for_asset(
+            asset_iri=asset_node.iri
         )
+        if cluster is not None:
+            keyword_set.add(cluster.iri)
+
+        keyword_sets.append(keyword_set)
 
     ################################################
     logger.info("Calculating similarity scores...")
@@ -70,7 +80,8 @@ def similarity_pipeline_7_asset_similarity():
 
             cosine_similarity = round(
                 number=float(
-                    np.dot(vector1, vector2) / (linalg.norm(vector1) * linalg.norm(vector2))
+                    np.dot(vector1, vector2)
+                    / (linalg.norm(vector1) * linalg.norm(vector2))
                 ),
                 ndigits=4,
             )
@@ -98,8 +109,9 @@ def similarity_pipeline_7_asset_similarity():
                 similarity_score=combined_similarity,
             )
 
-
     ################################################
     # logger.info("Building clusters...") TODO
 
-    SimilarityPipelineStatusManager.instance().set_active(active=False, stage="asset_similarity")
+    SimilarityPipelineStatusManager.instance().set_active(
+        active=False, stage="asset_similarity"
+    )