kg-modality-mining

name	kg-modality-mining
skill_type	skill
description	Run data-mining over the epistemic-graph engine — association-rule mining (Apriori/FP-Growth/Eclat), clustering (DBSCAN/hierarchical/GMM/k-medoids), and anomaly detection (z-score/isolation-forest/LOF/one-class-SVM), compute-near-data, plus the deep-learning family (LSTM forecast, MLP classify, autoencoder anomaly, gradient-boosting, neural embedding) delegated to data-science-mcp. Use for "find frequent itemsets", "mine association rules", "cluster these nodes/ vectors", "detect anomalies/outliers in this series or graph", "forecast this series with a neural model", "deep-classify these rows".
license	MIT
tags	["graph-os","engine","modality","mining","clustering","anomaly","data-mining","deep-learning"]
tier	modality
wraps	["engine_mining","graph_mine","graph_mine_deep"]
metadata	{"author":"Genius","version":"0.2.0"}

KG Modality — Mining

Condensed intent-surface note (Seam 8). Under the default intent surface (MCP_TOOL_MODE=intent), engine_mining, graph_mine, graph_mine_deep are held back from the default tool list (nothing removed — REST + _execute_tool still reach them exactly as documented below). Two ways to use this skill unchanged: (1) load_tools(tools=["engine_mining"]) once per session (as below), then proceed exactly as documented; or (2) call the ask intent verb with the same natural-language request — the resolver routes to engine_mining for you and returns the result plus a routing justification. Set MCP_TOOL_MODE=condensed/verbose/both to expose the granular tools eagerly instead.

Fronts the epistemic-graph engine's mining domain (CONCEPT:EG-KG.mining.frequent-itemset-mining): association-rule mining, clustering, and anomaly detection running inside the engine, compute-near-data, with an optional writeback of results as first-class graph nodes.

Two surfaces over the same engine MiningClient:

• engine_mining — the raw modality-tier action router (1:1 over the client's methods; empty action lists what the live engine supports).
• graph_mine — the friendlier, action-routed wrapper with three fixed actions and richer input shapes (baskets, feature matrices, or a graph/vector source selector):
  • associate — frequent-itemset + rules (Apriori/FP-Growth/Eclat: support, confidence, lift). Provide transactions (baskets of item labels) OR a graph-derived source (node_label, direction, item_field, relation, limit). writeback: true ⇒ :AssociationRule nodes.
  • cluster — DBSCAN (default) / hierarchical / GMM / k-medoids over a features row matrix OR a vector source (the stored embeddings of those nodes). writeback: true ⇒ :Cluster nodes linked to members.
  • anomaly — z-score (default) / isolation-forest / LOF / one-class-SVM over features, a 1-D values series (timeseries RCA), or a vector source. writeback: true ⇒ :Anomaly nodes linked to their source.

Degrades cleanly on a no-mining engine build (returns a structured error, never a crash).

• graph_mine_deep (CONCEPT:AU-KG.mining.dsm-forecast-delegation) — the heavy-Python / deep-learning family the pure-Rust engine deliberately does NOT implement (no torch/GPU in-engine): dispatches to agents/data-science-mcp over MCP and folds the result back into the KG as typed nodes. Actions: deep_forecast (LSTM sequence forecaster), deep_classify (MLP classifier), autoencoder_anomaly (reconstruction-error outlier detection), xgboost (gradient-boosting classifier), embed (neural embedding of numeric feature rows). Every action accepts raw rows OR a graph-derived source selector; writeback: true materializes typed result nodes (:Forecast/:Classification/:Anomaly/:Embedding) linked back to their source. Degrades cleanly ({available:false, error:...}) when data-science-mcp is unreachable or its [training] extra isn't installed.

How to reach it

Via the multiplexer: load_tools(tools=["graph_mine"]) (or engine_mining for the raw action router, or graph_mine_deep for the deep-learning delegation), then call; unload_tools(...) when done.

REST twins: POST /engine/mining (raw), POST /api/mining/{associate, cluster,anomaly} (natural per-action body, same _execute_tool core as graph_mine), and POST /api/mining/deep/{deep_forecast,deep_classify, autoencoder_anomaly,xgboost,embed} (same core as graph_mine_deep).

Example

// association rules from baskets
graph_mine(action="associate",
  params_json="{\"transactions\":[[\"bread\",\"milk\"],[\"bread\",\"butter\"]],\"min_support\":0.5}")

// cluster the stored embeddings of a node set, writing back :Cluster nodes
graph_mine(action="cluster",
  params_json="{\"source\":{\"node_label\":\"Doc\"},\"algorithm\":\"kmedoids\",\"k\":3,\"writeback\":true}")

// anomaly detection over a metric series
graph_mine(action="anomaly",
  params_json="{\"values\":[1,1,1,100],\"algorithm\":\"zscore\"}")