risk-network 0.0.9b26__py3-none-any.whl → 0.0.9b27__py3-none-any.whl

risk/__init__.py

@@ -7,4 +7,4 @@ RISK: Regional Inference of Significant Kinships
 
 from risk.risk import RISK
 
-__version__ = "0.0.9-beta.26"
+__version__ = "0.0.9-beta.27"

risk/annotations/annotations.py

@@ -16,7 +16,7 @@ from nltk.tokenize import word_tokenize
 from nltk.corpus import stopwords
 
 from risk.log import logger
-from scipy.sparse import csr_matrix
+from scipy.sparse import coo_matrix
 
 
 def _setup_nltk():
@@ -41,14 +41,13 @@ stop_words = set(stopwords.words("english"))
 def load_annotations(
     network: nx.Graph, annotations_input: Dict[str, Any], min_nodes_per_term: int = 2
 ) -> Dict[str, Any]:
-    """Convert annotations input to a DataFrame and reindex based on the network's node labels.
+    """Convert annotations input to a sparse matrix and reindex based on the network's node labels.
 
     Args:
         network (nx.Graph): The network graph.
         annotations_input (Dict[str, Any]): A dictionary with annotations.
         min_nodes_per_term (int, optional): The minimum number of network nodes required for each annotation
             term to be included. Defaults to 2.
-        use_sparse (bool, optional): Whether to return the annotations matrix as a sparse matrix. Defaults to True.
 
     Returns:
         Dict[str, Any]: A dictionary containing ordered nodes, ordered annotations, and the sparse binary annotations
@@ -58,51 +57,53 @@ def load_annotations(
         ValueError: If no annotations are found for the nodes in the network.
         ValueError: If no annotations have at least min_nodes_per_term nodes in the network.
     """
-    # Flatten the dictionary to a list of tuples for easier DataFrame creation
-    flattened_annotations = [
-        (node, annotation) for annotation, nodes in annotations_input.items() for node in nodes
-    ]
-    # Create a DataFrame from the flattened list
-    annotations = pd.DataFrame(flattened_annotations, columns=["node", "annotations"])
-    annotations["is_member"] = 1
-    # Pivot to create a binary matrix with nodes as rows and annotations as columns
-    annotations_pivot = annotations.pivot_table(
-        index="node", columns="annotations", values="is_member", fill_value=0, dropna=False
-    )
-    # Reindex the annotations matrix based on the node labels from the network
-    node_label_order = (attr["label"] for _, attr in network.nodes(data=True) if "label" in attr)
-    annotations_pivot = annotations_pivot.reindex(index=node_label_order)
-    # Raise an error if no valid annotations are found for the nodes in the network
-    if annotations_pivot.notnull().sum().sum() == 0:
+    # Step 1: Map nodes and annotations to indices
+    node_label_order = [attr["label"] for _, attr in network.nodes(data=True) if "label" in attr]
+    node_to_idx = {node: i for i, node in enumerate(node_label_order)}
+    annotation_to_idx = {annotation: i for i, annotation in enumerate(annotations_input)}
+    # Step 2: Construct a sparse binary matrix directly
+    row = []
+    col = []
+    data = []
+    for annotation, nodes in annotations_input.items():
+        for node in nodes:
+            if node in node_to_idx and annotation in annotation_to_idx:
+                row.append(node_to_idx[node])
+                col.append(annotation_to_idx[annotation])
+                data.append(1)
+
+    # Create a sparse binary matrix
+    num_nodes = len(node_to_idx)
+    num_annotations = len(annotation_to_idx)
+    annotations_pivot = coo_matrix((data, (row, col)), shape=(num_nodes, num_annotations)).tocsr()
+    # Step 3: Filter out annotations with fewer than min_nodes_per_term occurrences
+    valid_annotations = annotations_pivot.sum(axis=0).A1 >= min_nodes_per_term
+    annotations_pivot = annotations_pivot[:, valid_annotations]
+    # Step 4: Raise errors for empty matrices
+    if annotations_pivot.nnz == 0:
         raise ValueError("No terms found in the annotation file for the nodes in the network.")
 
-    # Filter out annotations with fewer than min_nodes_per_term occurrences
-    num_terms_before_filtering = annotations_pivot.shape[1]
-    annotations_pivot = annotations_pivot.loc[
-        :, (annotations_pivot.sum(axis=0) >= min_nodes_per_term)
-    ]
-    num_terms_after_filtering = annotations_pivot.shape[1]
-    # Log the number of annotations before and after filtering
-    logger.info(f"Minimum number of nodes per annotation term: {min_nodes_per_term}")
-    logger.info(f"Number of input annotation terms: {num_terms_before_filtering}")
-    logger.info(f"Number of remaining annotation terms: {num_terms_after_filtering}")
-    if num_terms_after_filtering == 0:
+    num_remaining_annotations = annotations_pivot.shape[1]
+    if num_remaining_annotations == 0:
         raise ValueError(
             f"No annotation terms found with at least {min_nodes_per_term} nodes in the network."
         )
 
-    # Extract ordered nodes and annotations
-    ordered_nodes = tuple(annotations_pivot.index)
-    ordered_annotations = tuple(annotations_pivot.columns)
-    # Convert the annotations_pivot matrix to a numpy array or sparse matrix
-    annotations_pivot_binary = (annotations_pivot.fillna(0).to_numpy() > 0).astype(int)
-    # Convert the binary annotations matrix to a sparse matrix
-    annotations_pivot_binary = csr_matrix(annotations_pivot_binary)
+    # Step 5: Extract ordered nodes and annotations
+    ordered_nodes = tuple(node_label_order)
+    ordered_annotations = tuple(
+        annotation for annotation, is_valid in zip(annotation_to_idx, valid_annotations) if is_valid
+    )
+
+    # Log the filtering details
+    logger.info(f"Minimum number of nodes per annotation term: {min_nodes_per_term}")
+    logger.info(f"Number of input annotation terms: {num_annotations}")
+    logger.info(f"Number of remaining annotation terms: {num_remaining_annotations}")
 
     return {
         "ordered_nodes": ordered_nodes,
         "ordered_annotations": ordered_annotations,
-        "matrix": annotations_pivot_binary,
+        "matrix": annotations_pivot,
     }
 
 

risk/neighborhoods/api.py

@@ -368,7 +368,7 @@ class NeighborhoodsAPI:
         )
 
         # Make a copy of the network to avoid modifying the original
-        network = copy.deepcopy(network)
+        network = copy.copy(network)
         # Load neighborhoods based on the network and distance metric
         neighborhoods = self._load_neighborhoods(
             network,
@@ -438,9 +438,5 @@ class NeighborhoodsAPI:
             random_seed=random_seed,
         )
 
-        # Ensure the neighborhood matrix is in sparse format
-        if not isinstance(neighborhoods, csr_matrix):
-            neighborhoods = csr_matrix(neighborhoods)
-
         # Return the sparse neighborhood matrix
         return neighborhoods

risk/neighborhoods/community.py

@@ -10,22 +10,23 @@ import networkx as nx
 import numpy as np
 from leidenalg import find_partition, RBConfigurationVertexPartition
 from networkx.algorithms.community import greedy_modularity_communities
+from scipy.sparse import csr_matrix
 
 from risk.log import logger
 
 
 def calculate_greedy_modularity_neighborhoods(
     network: nx.Graph, fraction_shortest_edges: float = 1.0
-) -> np.ndarray:
-    """Calculate neighborhoods using the Greedy Modularity method.
+) -> csr_matrix:
+    """Calculate neighborhoods using the Greedy Modularity method with CSR matrix output.
 
     Args:
         network (nx.Graph): The network graph.
         fraction_shortest_edges (float, optional): Shortest edge rank fraction threshold for creating
-            subgraphs before clustering.
+            subgraphs before clustering. Defaults to 1.0.
 
     Returns:
-        np.ndarray: A binary neighborhood matrix where nodes in the same community have 1, and others have 0.
+        csr_matrix: A binary neighborhood matrix (CSR) where nodes in the same community have 1, and others have 0.
     """
     # Create a subgraph with the shortest edges based on the rank fraction
     subnetwork = _create_percentile_limited_subgraph(
@@ -36,35 +37,36 @@ def calculate_greedy_modularity_neighborhoods(
     # Get the list of nodes in the original NetworkX graph
     nodes = list(network.nodes())
     node_index_map = {node: idx for idx, node in enumerate(nodes)}
-    # Create a binary neighborhood matrix
-    num_nodes = len(nodes)
-    # Initialize neighborhoods with zeros and set self-self entries to 1
-    neighborhoods = np.eye(num_nodes, dtype=int)
-    # Fill in the neighborhood matrix for nodes in the same community
+    # Prepare data for CSR matrix
+    row_indices = []
+    col_indices = []
     for community in communities:
-        # Iterate through all pairs of nodes in the same community
-        for node_i in community:
-            for node_j in community:
-                idx_i = node_index_map[node_i]
-                idx_j = node_index_map[node_j]
-                # Set them as neighbors (1) in the binary matrix
-                neighborhoods[idx_i, idx_j] = 1
+        mapped_indices = [node_index_map[node] for node in community]
+        for i in mapped_indices:
+            for j in mapped_indices:
+                row_indices.append(i)
+                col_indices.append(j)
+
+    # Create a CSR matrix
+    num_nodes = len(nodes)
+    data = np.ones(len(row_indices), dtype=int)
+    neighborhoods = csr_matrix((data, (row_indices, col_indices)), shape=(num_nodes, num_nodes))
 
     return neighborhoods
 
 
 def calculate_label_propagation_neighborhoods(
     network: nx.Graph, fraction_shortest_edges: float = 1.0
-) -> np.ndarray:
+) -> csr_matrix:
     """Apply Label Propagation to the network to detect communities.
 
     Args:
         network (nx.Graph): The network graph.
         fraction_shortest_edges (float, optional): Shortest edge rank fraction threshold for creating
-            subgraphs before clustering.
+            subgraphs before clustering. Defaults to 1.0.
 
     Returns:
-        np.ndarray: A binary neighborhood matrix on Label Propagation.
+        csr_matrix: A binary neighborhood matrix (CSR) on Label Propagation.
     """
     # Create a subgraph with the shortest edges based on the rank fraction
     subnetwork = _create_percentile_limited_subgraph(
@@ -77,17 +79,21 @@ def calculate_label_propagation_neighborhoods(
     # Get the list of nodes in the network
     nodes = list(network.nodes())
     node_index_map = {node: idx for idx, node in enumerate(nodes)}
-    # Create a binary neighborhood matrix
-    num_nodes = len(nodes)
-    # Initialize neighborhoods with zeros and set self-self entries to 1
-    neighborhoods = np.eye(num_nodes, dtype=int)
+    # Prepare data for CSR matrix
+    row_indices = []
+    col_indices = []
     # Assign neighborhoods based on community labels using the mapped indices
     for community in communities:
-        for node_i in community:
-            for node_j in community:
-                idx_i = node_index_map[node_i]
-                idx_j = node_index_map[node_j]
-                neighborhoods[idx_i, idx_j] = 1
+        mapped_indices = [node_index_map[node] for node in community]
+        for i in mapped_indices:
+            for j in mapped_indices:
+                row_indices.append(i)
+                col_indices.append(j)
+
+    # Create a CSR matrix
+    num_nodes = len(nodes)
+    data = np.ones(len(row_indices), dtype=int)
+    neighborhoods = csr_matrix((data, (row_indices, col_indices)), shape=(num_nodes, num_nodes))
 
     return neighborhoods
 
@@ -97,18 +103,18 @@ def calculate_leiden_neighborhoods(
     resolution: float = 1.0,
     fraction_shortest_edges: float = 1.0,
     random_seed: int = 888,
-) -> np.ndarray:
-    """Calculate neighborhoods using the Leiden method.
+) -> csr_matrix:
+    """Calculate neighborhoods using the Leiden method with CSR matrix output.
 
     Args:
         network (nx.Graph): The network graph.
         resolution (float, optional): Resolution parameter for the Leiden method. Defaults to 1.0.
         fraction_shortest_edges (float, optional): Shortest edge rank fraction threshold for creating
-            subgraphs before clustering.
+            subgraphs before clustering. Defaults to 1.0.
         random_seed (int, optional): Random seed for reproducibility. Defaults to 888.
 
     Returns:
-        np.ndarray: A binary neighborhood matrix where nodes in the same community have 1, and others have 0.
+        csr_matrix: A binary neighborhood matrix (CSR) where nodes in the same community have 1, and others have 0.
     """
     # Create a subgraph with the shortest edges based on the rank fraction
     subnetwork = _create_percentile_limited_subgraph(
@@ -126,17 +132,20 @@ def calculate_leiden_neighborhoods(
     # Get the list of nodes in the original NetworkX graph
     nodes = list(network.nodes())
     node_index_map = {node: idx for idx, node in enumerate(nodes)}
-    # Create a binary neighborhood matrix
-    num_nodes = len(nodes)
-    # Initialize neighborhoods with zeros and set self-self entries to 1
-    neighborhoods = np.eye(num_nodes, dtype=int)
-    # Assign neighborhoods based on community partitions using the mapped indices
+    # Prepare data for CSR matrix
+    row_indices = []
+    col_indices = []
     for community in partition:
-        for node_i in community:
-            for node_j in community:
-                idx_i = node_index_map[igraph_network.vs[node_i]["_nx_name"]]
-                idx_j = node_index_map[igraph_network.vs[node_j]["_nx_name"]]
-                neighborhoods[idx_i, idx_j] = 1
+        mapped_indices = [node_index_map[igraph_network.vs[node]["_nx_name"]] for node in community]
+        for i in mapped_indices:
+            for j in mapped_indices:
+                row_indices.append(i)
+                col_indices.append(j)
+
+    # Create a CSR matrix
+    num_nodes = len(nodes)
+    data = np.ones(len(row_indices), dtype=int)
+    neighborhoods = csr_matrix((data, (row_indices, col_indices)), shape=(num_nodes, num_nodes))
 
     return neighborhoods
 
@@ -146,18 +155,18 @@ def calculate_louvain_neighborhoods(
     resolution: float = 0.1,
     fraction_shortest_edges: float = 1.0,
     random_seed: int = 888,
-) -> np.ndarray:
+) -> csr_matrix:
     """Calculate neighborhoods using the Louvain method.
 
     Args:
         network (nx.Graph): The network graph.
         resolution (float, optional): Resolution parameter for the Louvain method. Defaults to 0.1.
         fraction_shortest_edges (float, optional): Shortest edge rank fraction threshold for creating
-            subgraphs before clustering.
+            subgraphs before clustering. Defaults to 1.0.
         random_seed (int, optional): Random seed for reproducibility. Defaults to 888.
 
     Returns:
-        np.ndarray: A binary neighborhood matrix on the Louvain method.
+        csr_matrix: A binary neighborhood matrix in CSR format.
     """
     # Create a subgraph with the shortest edges based on the rank fraction
     subnetwork = _create_percentile_limited_subgraph(
@@ -170,83 +179,110 @@ def calculate_louvain_neighborhoods(
     # Get the list of nodes in the network and create a mapping to indices
     nodes = list(network.nodes())
     node_index_map = {node: idx for idx, node in enumerate(nodes)}
-    # Create a binary neighborhood matrix
-    num_nodes = len(nodes)
-    # Initialize neighborhoods with zeros and set self-self entries to 1
-    neighborhoods = np.eye(num_nodes, dtype=int)
     # Group nodes by community
     community_groups = {}
     for node, community in partition.items():
         community_groups.setdefault(community, []).append(node)
 
-    # Assign neighborhoods based on community partitions using the mapped indices
-    for community, nodes in community_groups.items():
-        for node_i in nodes:
-            for node_j in nodes:
-                idx_i = node_index_map[node_i]
-                idx_j = node_index_map[node_j]
-                neighborhoods[idx_i, idx_j] = 1
+    # Prepare data for CSR matrix
+    row_indices = []
+    col_indices = []
+    for community_nodes in community_groups.values():
+        mapped_indices = [node_index_map[node] for node in community_nodes]
+        for i in mapped_indices:
+            for j in mapped_indices:
+                row_indices.append(i)
+                col_indices.append(j)
+
+    # Create a CSR matrix
+    num_nodes = len(nodes)
+    data = np.ones(len(row_indices), dtype=int)
+    neighborhoods = csr_matrix((data, (row_indices, col_indices)), shape=(num_nodes, num_nodes))
 
     return neighborhoods
 
 
 def calculate_markov_clustering_neighborhoods(
     network: nx.Graph, fraction_shortest_edges: float = 1.0
-) -> np.ndarray:
-    """Apply Markov Clustering (MCL) to the network and return a binary neighborhood matrix.
+) -> csr_matrix:
+    """Apply Markov Clustering (MCL) to the network and return a binary neighborhood matrix (CSR).
 
     Args:
         network (nx.Graph): The network graph.
         fraction_shortest_edges (float, optional): Shortest edge rank fraction threshold for creating
-            subgraphs before clustering.
+            subgraphs before clustering. Defaults to 1.0.
 
     Returns:
-        np.ndarray: A binary neighborhood matrix on Markov Clustering.
+        csr_matrix: A binary neighborhood matrix (CSR) on Markov Clustering.
+
+    Warning:
+        This function temporarily converts the adjacency matrix to a dense format, which may lead to
+        high memory consumption for large graphs.
     """
     # Create a subgraph with the shortest edges based on the rank fraction
     subnetwork = _create_percentile_limited_subgraph(
         network, fraction_shortest_edges=fraction_shortest_edges
     )
-    # Step 1: Convert the subnetwork to an adjacency matrix
+    # Check if the subgraph has edges
+    if subnetwork.number_of_edges() == 0:
+        raise ValueError("The subgraph has no edges. Adjust the fraction_shortest_edges parameter.")
+
+    # Step 1: Convert the subnetwork to an adjacency matrix (CSR)
     subnetwork_nodes = list(subnetwork.nodes())
-    adjacency_matrix = nx.to_numpy_array(subnetwork, nodelist=subnetwork_nodes)
-    # Step 2: Run Markov Clustering (MCL) on the subnetwork's adjacency matrix
-    result = mc.run_mcl(adjacency_matrix)
+    adjacency_matrix = nx.to_scipy_sparse_array(subnetwork, nodelist=subnetwork_nodes)
+    # Ensure the adjacency matrix is valid
+    if adjacency_matrix.shape[0] == 0 or adjacency_matrix.shape[1] == 0:
+        raise ValueError(
+            "The adjacency matrix is empty. Check the input graph or filtering criteria."
+        )
+
+    # Convert the sparse matrix to dense format for MCL
+    dense_matrix = adjacency_matrix.toarray()
+    # Step 2: Run Markov Clustering (MCL) on the dense adjacency matrix
+    try:
+        result = mc.run_mcl(dense_matrix)
+    except Exception as e:
+        raise RuntimeError(f"Markov Clustering failed: {e}")
+
     clusters = mc.get_clusters(result)
     # Step 3: Prepare the original network nodes and indices
     nodes = list(network.nodes())
     node_index_map = {node: idx for idx, node in enumerate(nodes)}
     num_nodes = len(nodes)
-    # Step 4: Initialize the neighborhood matrix for the original network
-    neighborhoods = np.eye(num_nodes, dtype=int)
-    # Step 5: Fill the neighborhoods matrix using the clusters from the subnetwork
+    # Step 4: Prepare data for CSR matrix
+    row_indices = []
+    col_indices = []
     for cluster in clusters:
         for node_i in cluster:
             for node_j in cluster:
                 # Map the indices back to the original network's node indices
                 original_node_i = subnetwork_nodes[node_i]
                 original_node_j = subnetwork_nodes[node_j]
-
                 if original_node_i in node_index_map and original_node_j in node_index_map:
                     idx_i = node_index_map[original_node_i]
                     idx_j = node_index_map[original_node_j]
-                    neighborhoods[idx_i, idx_j] = 1
+                    row_indices.append(idx_i)
+                    col_indices.append(idx_j)
+
+    # Step 5: Create a CSR matrix
+    data = np.ones(len(row_indices), dtype=int)
+    neighborhoods = csr_matrix((data, (row_indices, col_indices)), shape=(num_nodes, num_nodes))
 
     return neighborhoods
 
 
 def calculate_spinglass_neighborhoods(
     network: nx.Graph, fraction_shortest_edges: float = 1.0
-) -> np.ndarray:
+) -> csr_matrix:
     """Apply Spinglass Community Detection to the network, handling disconnected components.
 
     Args:
         network (nx.Graph): The network graph.
         fraction_shortest_edges (float, optional): Shortest edge rank fraction threshold for creating
-            subgraphs before clustering.
+            subgraphs before clustering. Defaults to 1.0.
 
     Returns:
-        np.ndarray: A binary neighborhood matrix based on Spinglass communities.
+        csr_matrix: A binary neighborhood matrix (CSR) based on Spinglass communities.
     """
     # Create a subgraph with the shortest edges based on the rank fraction
     subnetwork = _create_percentile_limited_subgraph(
@@ -254,12 +290,11 @@ def calculate_spinglass_neighborhoods(
     )
     # Step 1: Find connected components in the graph
     components = list(nx.connected_components(subnetwork))
-    # Prepare to store community results
+    # Prepare data for CSR matrix
     nodes = list(network.nodes())
     node_index_map = {node: idx for idx, node in enumerate(nodes)}
-    num_nodes = len(nodes)
-    # Initialize neighborhoods with zeros and set self-self entries to 1
-    neighborhoods = np.eye(num_nodes, dtype=int)
+    row_indices = []
+    col_indices = []
     # Step 2: Run Spinglass on each connected component
     for component in components:
         # Extract the subgraph corresponding to the current component
@@ -280,27 +315,34 @@ def calculate_spinglass_neighborhoods(
 
         # Step 3: Assign neighborhoods based on community labels
         for community in communities:
-            for node_i in community:
-                for node_j in community:
-                    idx_i = node_index_map[igraph_subgraph.vs[node_i]["_nx_name"]]
-                    idx_j = node_index_map[igraph_subgraph.vs[node_j]["_nx_name"]]
-                    neighborhoods[idx_i, idx_j] = 1
+            mapped_indices = [
+                node_index_map[igraph_subgraph.vs[node]["_nx_name"]] for node in community
+            ]
+            for i in mapped_indices:
+                for j in mapped_indices:
+                    row_indices.append(i)
+                    col_indices.append(j)
+
+    # Step 4: Create a CSR matrix
+    num_nodes = len(nodes)
+    data = np.ones(len(row_indices), dtype=int)
+    neighborhoods = csr_matrix((data, (row_indices, col_indices)), shape=(num_nodes, num_nodes))
 
     return neighborhoods
 
 
 def calculate_walktrap_neighborhoods(
     network: nx.Graph, fraction_shortest_edges: float = 1.0
-) -> np.ndarray:
-    """Apply Walktrap Community Detection to the network.
+) -> csr_matrix:
+    """Apply Walktrap Community Detection to the network with CSR matrix output.
 
     Args:
         network (nx.Graph): The network graph.
         fraction_shortest_edges (float, optional): Shortest edge rank fraction threshold for creating
-            subgraphs before clustering.
+            subgraphs before clustering. Defaults to 1.0.
 
     Returns:
-        np.ndarray: A binary neighborhood matrix on Walktrap communities.
+        csr_matrix: A binary neighborhood matrix (CSR) on Walktrap communities.
     """
     # Create a subgraph with the shortest edges based on the rank fraction
     subnetwork = _create_percentile_limited_subgraph(
@@ -313,17 +355,20 @@ def calculate_walktrap_neighborhoods(
     # Get the list of nodes in the original NetworkX graph
     nodes = list(network.nodes())
     node_index_map = {node: idx for idx, node in enumerate(nodes)}
-    # Create a binary neighborhood matrix
-    num_nodes = len(nodes)
-    # Initialize neighborhoods with zeros and set self-self entries to 1
-    neighborhoods = np.eye(num_nodes, dtype=int)
-    # Assign neighborhoods based on community labels
+    # Prepare data for CSR matrix
+    row_indices = []
+    col_indices = []
     for community in communities:
-        for node_i in community:
-            for node_j in community:
-                idx_i = node_index_map[igraph_network.vs[node_i]["_nx_name"]]
-                idx_j = node_index_map[igraph_network.vs[node_j]["_nx_name"]]
-                neighborhoods[idx_i, idx_j] = 1
+        mapped_indices = [node_index_map[igraph_network.vs[node]["_nx_name"]] for node in community]
+        for i in mapped_indices:
+            for j in mapped_indices:
+                row_indices.append(i)
+                col_indices.append(j)
+
+    # Create a CSR matrix
+    num_nodes = len(nodes)
+    data = np.ones(len(row_indices), dtype=int)
+    neighborhoods = csr_matrix((data, (row_indices, col_indices)), shape=(num_nodes, num_nodes))
 
     return neighborhoods