src/backend/api/services/constants.py

METHODS = [
    {
        "id": "coins",
        "name": "COINs - Knowledge Graph Reasoning",
        "thesis_section": "3.1",
        "description": (
            "Community-Informed Graph Embeddings (COINs) for scalable knowledge graph link prediction "
            "and complex query answering. Uses community detection to localize embedding computation, "
            "achieving significant speedups over full-graph methods."
        ),
    },
    {
        "id": "multiproxan",
        "name": "MultiProxAn - Graph Generation",
        "thesis_section": "4.3",
        "description": (
            "Discrete denoising diffusion model for graph generation with MultiProx sampling. "
            "Generates molecular graphs (QM9) and synthetic community graphs using iterative "
            "multi-measurement Gibbs sampling for improved sample quality."
        ),
    },
    {
        "id": "kg_anomaly",
        "name": "KG Anomaly Correction",
        "thesis_section": "4.4",
        "description": (
            "Diffusion-based knowledge graph subgraph correction. Applies the DiGress denoising "
            "diffusion model to knowledge graph subgraphs to detect and correct anomalous edges."
        ),
    },
]

COINS_DATASET_META = {
    "freebase": {
        "name": "FB15k-237",
        "description": "Subset of Freebase knowledge base with 237 relation types",
        "data_dir": "FB15k-237",
    },
    "wordnet": {
        "name": "WN18RR",
        "description": "Subset of WordNet lexical database with 11 relation types",
        "data_dir": "WN18RR",
    },
    "nell": {
        "name": "NELL-995",
        "description": "Never-Ending Language Learner knowledge base with 200 relation types",
        "data_dir": "NELL-995",
    },
}

COINS_MODELS = [
    {
        "algorithm": "transe",
        "name": "TransE",
        "description": "Translation-based embedding model",
        "supported_query_structures": ["1p"],
    },
    {
        "algorithm": "distmult",
        "name": "DistMult",
        "description": "Bilinear diagonal embedding model",
        "supported_query_structures": ["1p"],
    },
    {
        "algorithm": "complex",
        "name": "ComplEx",
        "description": "Complex-valued embedding model",
        "supported_query_structures": ["1p"],
    },
    {
        "algorithm": "rotate",
        "name": "RotatE",
        "description": "Rotation-based embedding model in complex space",
        "supported_query_structures": ["1p"],
    },
    {
        "algorithm": "q2b",
        "name": "Query2Box",
        "description": "Box embedding model for complex logical queries",
        "supported_query_structures": ["1p", "2p", "3p", "2i", "3i", "ip", "pi"],
    },
    {
        "algorithm": "kbgat",
        "name": "KBGAT",
        "description": "Knowledge base graph attention network",
        "supported_query_structures": ["1p"],
    },
]

QUERY_STRUCTURES = [
    {
        "id": "1p",
        "name": "Single Hop",
        "description": "Direct link prediction: who/what is connected to the anchor via this relation?",
        "nodes": [
            {"id": "a", "type": "anchor", "label": "Anchor"},
            {"id": "t", "type": "target", "label": "?"},
        ],
        "edges": [
            {"id": "r1", "source": "a", "target": "t", "label": "Relation"},
        ],
    },
    {
        "id": "2p",
        "name": "Two Hop",
        "description": "Two-step chain: anchor -> variable -> target",
        "nodes": [
            {"id": "a", "type": "anchor", "label": "Anchor"},
            {"id": "v1", "type": "variable", "label": "Variable"},
            {"id": "t", "type": "target", "label": "?"},
        ],
        "edges": [
            {"id": "r1", "source": "a", "target": "v1", "label": "Relation 1"},
            {"id": "r2", "source": "v1", "target": "t", "label": "Relation 2"},
        ],
    },
    {
        "id": "3p",
        "name": "Three Hop",
        "description": "Three-step chain: anchor -> v1 -> v2 -> target",
        "nodes": [
            {"id": "a", "type": "anchor", "label": "Anchor"},
            {"id": "v1", "type": "variable", "label": "Variable 1"},
            {"id": "v2", "type": "variable", "label": "Variable 2"},
            {"id": "t", "type": "target", "label": "?"},
        ],
        "edges": [
            {"id": "r1", "source": "a", "target": "v1", "label": "Relation 1"},
            {"id": "r2", "source": "v1", "target": "v2", "label": "Relation 2"},
            {"id": "r3", "source": "v2", "target": "t", "label": "Relation 3"},
        ],
    },
    {
        "id": "2i",
        "name": "Two Intersection",
        "description": "Intersection of two single-hop queries sharing the same target",
        "nodes": [
            {"id": "a1", "type": "anchor", "label": "Anchor 1"},
            {"id": "a2", "type": "anchor", "label": "Anchor 2"},
            {"id": "t", "type": "target", "label": "?"},
        ],
        "edges": [
            {"id": "r1", "source": "a1", "target": "t", "label": "Relation 1"},
            {"id": "r2", "source": "a2", "target": "t", "label": "Relation 2"},
        ],
    },
    {
        "id": "3i",
        "name": "Three Intersection",
        "description": "Intersection of three single-hop queries sharing the same target",
        "nodes": [
            {"id": "a1", "type": "anchor", "label": "Anchor 1"},
            {"id": "a2", "type": "anchor", "label": "Anchor 2"},
            {"id": "a3", "type": "anchor", "label": "Anchor 3"},
            {"id": "t", "type": "target", "label": "?"},
        ],
        "edges": [
            {"id": "r1", "source": "a1", "target": "t", "label": "Relation 1"},
            {"id": "r2", "source": "a2", "target": "t", "label": "Relation 2"},
            {"id": "r3", "source": "a3", "target": "t", "label": "Relation 3"},
        ],
    },
    {
        "id": "ip",
        "name": "Intersection then Projection",
        "description": "Two anchors intersect, then the result projects via a third relation to the target",
        "nodes": [
            {"id": "a1", "type": "anchor", "label": "Anchor 1"},
            {"id": "a2", "type": "anchor", "label": "Anchor 2"},
            {"id": "v1", "type": "variable", "label": "Variable"},
            {"id": "t", "type": "target", "label": "?"},
        ],
        "edges": [
            {"id": "r1", "source": "a1", "target": "v1", "label": "Relation 1"},
            {"id": "r2", "source": "a2", "target": "v1", "label": "Relation 2"},
            {"id": "r3", "source": "v1", "target": "t", "label": "Relation 3"},
        ],
    },
    {
        "id": "pi",
        "name": "Projection then Intersection",
        "description": "One anchor projects then intersects with a direct connection from a second anchor",
        "nodes": [
            {"id": "a1", "type": "anchor", "label": "Anchor 1"},
            {"id": "v1", "type": "variable", "label": "Variable"},
            {"id": "a2", "type": "anchor", "label": "Anchor 2"},
            {"id": "t", "type": "target", "label": "?"},
        ],
        "edges": [
            {"id": "r1", "source": "a1", "target": "v1", "label": "Relation 1"},
            {"id": "r2", "source": "v1", "target": "t", "label": "Relation 2"},
            {"id": "r3", "source": "a2", "target": "t", "label": "Relation 3"},
        ],
    },
]

GRAPHGEN_DATASETS = {
    "qm9": {
        "name": "QM9",
        "type": "molecular",
        "description": "Small organic molecules with up to 9 heavy atoms (C, N, O, F)",
        "node_types": ["C", "N", "O", "F"],
        "edge_types": ["none", "single", "double", "triple", "aromatic"],
        "max_nodes": 9,
    },
    "comm20": {
        "name": "Community20",
        "type": "synthetic",
        "description": "Synthetic community-structured graphs with 12-20 nodes",
        "node_types": ["node"],
        "edge_types": ["none", "edge"],
        "max_nodes": 20,
    },
}

GRAPHGEN_SAMPLING_MODES = [
    {
        "id": "standard",
        "name": "Standard Denoising",
        "description": "Iterative denoising from T to 0. Full quality, slower.",
        "parameters": [
            {
                "name": "diffusion_steps",
                "type": "integer",
                "description": "Number of diffusion steps T",
                "default": 500,
                "min": 50,
                "max": 1000,
            },
            {
                "name": "chain_frames",
                "type": "integer",
                "description": "Number of denoising snapshots in the GIF",
                "default": 20,
                "min": 10,
                "max": 30,
            },
        ],
    },
    {
        "id": "multiprox",
        "name": "MultiProx Sampling",
        "description": (
            "Multi-measurement Gibbs sampling with proximal steps. "
            "Step-by-step generation with controllable noise levels."
        ),
        "parameters": [
            {
                "name": "diffusion_steps",
                "type": "integer",
                "description": "Number of diffusion steps T",
                "default": 500,
                "min": 50,
                "max": 1000,
            },
            {
                "name": "m",
                "type": "integer",
                "description": "Number of parallel samples per multi-measurement step",
                "default": 10,
                "min": 2,
                "max": 100,
            },
            {
                "name": "t",
                "type": "float",
                "description": "First noise level (normalized, 0-1)",
                "default": 0.5,
                "min": 0.0,
                "max": 1.0,
            },
            {
                "name": "t_prime",
                "type": "float",
                "description": "Second noise level (normalized, 0-1). Must satisfy t_prime <= t.",
                "default": 0.1,
                "min": 0.0,
                "max": 1.0,
            },
        ],
    },
]

# --- COINs predict helpers ---

QUERY_STRUCTURE_INTERNAL = {
    "1p": "1p", "2p": "2p", "3p": "3p",
    "2i": "2i", "3i": "3i",
    "ip": "2i1p", "pi": "1p2i",
}

# Maps API node/edge IDs to query tree node/edge indices.
# Anchor/variable node IDs → tree vertex index; edge IDs → tree edge index.
# Key insight: edge_index == target_node_index in all COINs query trees.
QUERY_TREE_MAPPINGS = {
    "1p": {"nodes": {"a": 0}, "edges": {"r1": 0}},
    "2p": {"nodes": {"a": 0, "v1": 1}, "edges": {"r1": 0, "r2": 1}},
    "3p": {"nodes": {"a": 0, "v1": 1, "v2": 2}, "edges": {"r1": 0, "r2": 1, "r3": 2}},
    "2i": {"nodes": {"a1": 0, "a2": 2}, "edges": {"r1": 0, "r2": 2}},
    "3i": {"nodes": {"a1": 0, "a2": 2, "a3": 4}, "edges": {"r1": 0, "r2": 2, "r3": 4}},
    "ip": {"nodes": {"a1": 0, "a2": 2, "v1": 4}, "edges": {"r1": 0, "r2": 2, "r3": 4}},
    "pi": {"nodes": {"a1": 0, "v1": 1, "a2": 3}, "edges": {"r1": 0, "r2": 1, "r3": 3}},
}

COINS_CONFIG_SUFFIX = {
    "transe": "", "distmult": "_distmult", "complex": "_complex",
    "rotate": "_rotate", "q2b": "_q2b", "kbgat": "_gnn",
}


KG_ANOMALY_DATASET_META = {
    "freebase": {
        "name": "FB15k-237",
        "description": "Diffusion model trained on Freebase subgraphs",
    },
    "wordnet": {
        "name": "WN18RR",
        "description": "Diffusion model trained on WordNet subgraphs",
    },
    "nell": {
        "name": "NELL-995",
        "description": "Diffusion model trained on NELL subgraphs",
    },
}