Method cluster_ideas

internagent/mas/memory/long_memory.py:238–337

Perform clustering on ideas in the graph and assign cluster IDs. Args: method (str): Clustering method to use. Options: - "louvain": Community detection using Louvain algorithm (default) - "spectral": Spectral clustering -

(self, method: str = "louvain")

            return []

    def cluster_ideas(self, method: str = "louvain") -> None:
        """
        Perform clustering on ideas in the graph and assign cluster IDs.

        Args:
            method (str): Clustering method to use. Options:
                - "louvain": Community detection using Louvain algorithm (default)
                - "spectral": Spectral clustering
                - "embedding": Clustering based on embeddings (requires FINCH)
        """
        nodes = list(self.graph.nodes)

        if len(nodes) == 0:
            logger.warning("No nodes in graph, skipping clustering")
            return

        logger.info(f"Clustering {len(nodes)} ideas using method: {method}")

        if method == "louvain":
            # Use Louvain community detection
            try:
                communities = nx.community.louvain_communities(self.graph, seed=42)
                for cluster_id, community in enumerate(communities):
                    for node_id in community:
                        self.graph.nodes[node_id]['cluster_id'] = cluster_id
                logger.info(f"Louvain clustering created {len(communities)} clusters")
            except Exception as e:
                logger.error(f"Louvain clustering failed: {e}")
                # Fallback: assign all to cluster 0
                for node_id in nodes:
                    self.graph.nodes[node_id]['cluster_id'] = 0

        elif method == "spectral":
            # Use spectral clustering (requires connected components)
            try:
                # Get largest connected component
                if nx.is_connected(self.graph):
                    adj_matrix = nx.to_numpy_array(self.graph)
                    from sklearn.cluster import SpectralClustering
                    n_clusters = min(5, len(nodes))
                    clustering = SpectralClustering(n_clusters=n_clusters, affinity='precomputed')
                    labels = clustering.fit_predict(adj_matrix)
                    for node_id, label in zip(nodes, labels):
                        self.graph.nodes[node_id]['cluster_id'] = int(label)
                    logger.info(f"Spectral clustering created {n_clusters} clusters")
                else:
                    logger.warning("Graph not connected, using component-based clustering")
                    for cluster_id, component in enumerate(nx.connected_components(self.graph)):
                        for node_id in component:
                            self.graph.nodes[node_id]['cluster_id'] = cluster_id
            except Exception as e:
                logger.error(f"Spectral clustering failed: {e}")
                for node_id in nodes:
                    self.graph.nodes[node_id]['cluster_id'] = 0

        elif method == "embedding":
            # Use embedding-based clustering with FINCH or K-means
            if not CHROMA_AVAILABLE or self.collection is None: