tnfr 4.5.2__py3-none-any.whl → 8.5.0__py3-none-any.whl

tnfr/dynamics/__init__.pyi

@@ -0,0 +1,83 @@
+from __future__ import annotations
+
+from typing import Any, Literal, Sequence
+
+from tnfr.types import GlyphCode, TNFRGraph
+
+__all__: tuple[str, ...]
+
+dnfr: Any
+integrators: Any
+metabolism: Any
+
+ALIAS_D2EPI: Sequence[str]
+ALIAS_DNFR: Sequence[str]
+ALIAS_DSI: Sequence[str]
+ALIAS_EPI: Sequence[str]
+ALIAS_SI: Sequence[str]
+ALIAS_VF: Sequence[str]
+
+AbstractSelector: Any
+DefaultGlyphSelector: Any
+ParametricGlyphSelector: Any
+StructuralFeedbackLoop: Any
+AdaptiveSequenceSelector: Any
+StructuralHomeostasis: Any
+AdaptiveLearningSystem: Any
+_SelectorPreselection: Any
+_apply_glyphs: Any
+_apply_selector: Any
+_choose_glyph: Any
+_configure_selector_weights: Any
+ProcessPoolExecutor: Any
+_maybe_remesh: Any
+_normalize_job_overrides: Any
+_prepare_dnfr: Any
+_prepare_dnfr_data: Any
+_prepare_selector_preselection: Any
+_resolve_jobs_override: Any
+_resolve_preselected_glyph: Any
+_run_after_callbacks: Any
+_run_before_callbacks: Any
+_run_validators: Any
+_selector_parallel_jobs: Any
+_update_epi_hist: Any
+_update_node_sample: Any
+_update_nodes: Any
+_compute_dnfr: Any
+_compute_neighbor_means: Any
+_init_dnfr_cache: Any
+_refresh_dnfr_vectors: Any
+adapt_vf_by_coherence: Any
+coordinate_global_local_phase: Any
+default_compute_delta_nfr: Any
+default_glyph_selector: Any
+dnfr_epi_vf_mixed: Any
+dnfr_laplacian: Any
+dnfr_phase_only: Any
+get_numpy: Any
+apply_glyph: Any
+parametric_glyph_selector: Any
+
+AbstractIntegrator: Any
+DefaultIntegrator: Any
+
+def prepare_integration_params(
+    G: TNFRGraph,
+    dt: float | None = ...,
+    t: float | None = ...,
+    method: Literal["euler", "rk4"] | None = ...,
+) -> tuple[float, int, float, Literal["euler", "rk4"]]: ...
+
+run: Any
+set_delta_nfr_hook: Any
+step: Any
+
+def update_epi_via_nodal_equation(
+    G: TNFRGraph,
+    *,
+    dt: float | None = ...,
+    t: float | None = ...,
+    method: Literal["euler", "rk4"] | None = ...,
+    n_jobs: int | None = ...,
+) -> None: ...

tnfr/dynamics/adaptation.py

@@ -0,0 +1,267 @@
+"""νf adaptation routines for TNFR dynamics."""
+
+from __future__ import annotations
+
+import math
+from concurrent.futures import ProcessPoolExecutor
+from typing import Any, cast
+
+from ..alias import collect_attr, set_vf
+from ..constants import get_graph_param
+from ..utils import clamp, resolve_chunk_size
+from ..metrics.common import ensure_neighbors_map
+from ..types import CoherenceMetric, DeltaNFR, TNFRGraph
+from ..utils import get_numpy
+from .aliases import ALIAS_DNFR, ALIAS_SI, ALIAS_VF
+
+__all__ = ("adapt_vf_by_coherence",)
+
+
+def _vf_adapt_chunk(
+    args: tuple[list[tuple[Any, int, tuple[int, ...]]], tuple[float, ...], float],
+) -> list[tuple[Any, float]]:
+    """Return proposed νf updates for ``chunk`` of stable nodes."""
+
+    chunk, vf_values, mu = args
+    updates: list[tuple[Any, float]] = []
+    for node, idx, neighbor_idx in chunk:
+        vf = vf_values[idx]
+        if neighbor_idx:
+            mean = math.fsum(vf_values[j] for j in neighbor_idx) / len(neighbor_idx)
+        else:
+            mean = vf
+        updates.append((node, vf + mu * (mean - vf)))
+    return updates
+
+
+def adapt_vf_by_coherence(G: TNFRGraph, n_jobs: int | None = None) -> None:
+    """Synchronise νf to the neighbour mean once ΔNFR and Si stay coherent.
+
+    Parameters
+    ----------
+    G : TNFRGraph
+        Graph that stores the TNFR nodes and configuration required for
+        adaptation. The routine reads ``VF_ADAPT_TAU`` (τ) to decide how many
+        consecutive stable steps a node must accumulate in ``stable_count``
+        before updating. The adaptation weight ``VF_ADAPT_MU`` (μ) controls how
+        quickly νf moves toward the neighbour mean. Stability is detected when
+        the absolute ΔNFR stays below ``EPS_DNFR_STABLE`` and the sense index Si
+        exceeds the selector threshold ``SELECTOR_THRESHOLDS['si_hi']`` (falling
+        back to ``GLYPH_THRESHOLDS['hi']``). Only nodes that satisfy both
+        thresholds for τ successive evaluations are eligible for μ-weighted
+        averaging.
+    n_jobs : int or None, optional
+        Number of worker processes used for eligible nodes. ``None`` (the
+        default) keeps the adaptation serial, ``1`` disables parallelism, and
+        any value greater than one dispatches chunks of nodes to a
+        :class:`~concurrent.futures.ProcessPoolExecutor` so large graphs can
+        adjust νf without blocking the main dynamic loop.
+
+    Returns
+    -------
+    None
+        The graph is updated in place; no value is returned.
+
+    Raises
+    ------
+    KeyError
+        Raised when ``G.graph`` lacks the canonical adaptation parameters and
+        defaults have not been injected.
+
+    Examples
+    --------
+    >>> from tnfr.constants import inject_defaults
+    >>> from tnfr.dynamics import adapt_vf_by_coherence
+    >>> from tnfr.structural import create_nfr
+    >>> G, seed = create_nfr("seed", vf=0.2)
+    >>> _, anchor = create_nfr("anchor", graph=G, vf=1.0)
+    >>> G.add_edge(seed, anchor)
+    >>> inject_defaults(G)
+    >>> G.graph["VF_ADAPT_TAU"] = 2      # τ: consecutive stable steps
+    >>> G.graph["VF_ADAPT_MU"] = 0.5      # μ: neighbour coupling strength
+    >>> G.graph["SELECTOR_THRESHOLDS"] = {"si_hi": 0.8}
+    >>> for node in G.nodes:
+    ...     G.nodes[node]["Si"] = 0.9      # above ΔSi threshold
+    ...     G.nodes[node]["ΔNFR"] = 0.0    # within |ΔNFR| ≤ eps guard
+    ...     G.nodes[node]["stable_count"] = 1
+    >>> adapt_vf_by_coherence(G)
+    >>> round(G.nodes[seed]["νf"], 2), round(G.nodes[anchor]["νf"], 2)
+    (0.6, 0.6)
+    >>> G.nodes[seed]["stable_count"], G.nodes[anchor]["stable_count"] >= 2
+    (2, True)
+    """
+
+    required_keys = ("VF_ADAPT_TAU", "VF_ADAPT_MU")
+    missing_keys = [key for key in required_keys if key not in G.graph]
+    if missing_keys:
+        missing_list = ", ".join(sorted(missing_keys))
+        raise KeyError(
+            "adapt_vf_by_coherence requires graph parameters "
+            f"{missing_list}; call tnfr.constants.inject_defaults(G) "
+            "before adaptation."
+        )
+
+    tau = get_graph_param(G, "VF_ADAPT_TAU", int)
+    mu = float(get_graph_param(G, "VF_ADAPT_MU"))
+    eps_dnfr = cast(DeltaNFR, get_graph_param(G, "EPS_DNFR_STABLE"))
+    thr_sel = get_graph_param(G, "SELECTOR_THRESHOLDS", dict)
+    thr_def = get_graph_param(G, "GLYPH_THRESHOLDS", dict)
+    si_hi = cast(
+        CoherenceMetric,
+        float(thr_sel.get("si_hi", thr_def.get("hi", 0.66))),
+    )
+    vf_min = float(get_graph_param(G, "VF_MIN"))
+    vf_max = float(get_graph_param(G, "VF_MAX"))
+
+    nodes = list(G.nodes)
+    if not nodes:
+        return
+
+    neighbors_map = ensure_neighbors_map(G)
+    node_count = len(nodes)
+    node_index = {node: idx for idx, node in enumerate(nodes)}
+
+    jobs: int | None
+    if n_jobs is None:
+        jobs = None
+    else:
+        try:
+            jobs = int(n_jobs)
+        except (TypeError, ValueError):
+            jobs = None
+        else:
+            if jobs <= 1:
+                jobs = None
+
+    np_mod = get_numpy()
+    use_np = np_mod is not None
+
+    si_values = collect_attr(G, nodes, ALIAS_SI, 0.0, np=np_mod if use_np else None)
+    dnfr_values = collect_attr(G, nodes, ALIAS_DNFR, 0.0, np=np_mod if use_np else None)
+    vf_values = collect_attr(G, nodes, ALIAS_VF, 0.0, np=np_mod if use_np else None)
+
+    if use_np:
+        np = np_mod  # type: ignore[assignment]
+        assert np is not None
+        si_arr = si_values.astype(float, copy=False)
+        dnfr_arr = np.abs(dnfr_values.astype(float, copy=False))
+        vf_arr = vf_values.astype(float, copy=False)
+
+        prev_counts = np.fromiter(
+            (int(G.nodes[node].get("stable_count", 0)) for node in nodes),
+            dtype=int,
+            count=node_count,
+        )
+        stable_mask = (si_arr >= si_hi) & (dnfr_arr <= eps_dnfr)
+        new_counts = np.where(stable_mask, prev_counts + 1, 0)
+
+        for node, count in zip(nodes, new_counts.tolist()):
+            G.nodes[node]["stable_count"] = int(count)
+
+        eligible_mask = new_counts >= tau
+        if not bool(eligible_mask.any()):
+            return
+
+        max_degree = 0
+        if node_count:
+            degree_counts = np.fromiter(
+                (len(neighbors_map.get(node, ())) for node in nodes),
+                dtype=int,
+                count=node_count,
+            )
+            if degree_counts.size:
+                max_degree = int(degree_counts.max())
+
+        if max_degree > 0:
+            neighbor_indices = np.zeros((node_count, max_degree), dtype=int)
+            mask = np.zeros((node_count, max_degree), dtype=bool)
+            for idx, node in enumerate(nodes):
+                neigh = neighbors_map.get(node, ())
+                if not neigh:
+                    continue
+                idxs = [node_index[nbr] for nbr in neigh if nbr in node_index]
+                if not idxs:
+                    continue
+                length = len(idxs)
+                neighbor_indices[idx, :length] = idxs
+                mask[idx, :length] = True
+            neighbor_values = vf_arr[neighbor_indices]
+            sums = (neighbor_values * mask).sum(axis=1)
+            counts = mask.sum(axis=1)
+            neighbor_means = np.where(counts > 0, sums / counts, vf_arr)
+        else:
+            neighbor_means = vf_arr
+
+        vf_updates = vf_arr + mu * (neighbor_means - vf_arr)
+        for idx in np.nonzero(eligible_mask)[0]:
+            node = nodes[int(idx)]
+            vf_new = clamp(float(vf_updates[int(idx)]), vf_min, vf_max)
+            set_vf(G, node, vf_new)
+        return
+
+    si_list = [float(val) for val in si_values]
+    dnfr_list = [abs(float(val)) for val in dnfr_values]
+    vf_list = [float(val) for val in vf_values]
+
+    prev_counts = [int(G.nodes[node].get("stable_count", 0)) for node in nodes]
+    stable_flags = [
+        si >= si_hi and dnfr <= eps_dnfr for si, dnfr in zip(si_list, dnfr_list)
+    ]
+    new_counts = [
+        prev + 1 if flag else 0 for prev, flag in zip(prev_counts, stable_flags)
+    ]
+
+    for node, count in zip(nodes, new_counts):
+        G.nodes[node]["stable_count"] = int(count)
+
+    eligible_nodes = [node for node, count in zip(nodes, new_counts) if count >= tau]
+    if not eligible_nodes:
+        return
+
+    if jobs is None:
+        for node in eligible_nodes:
+            idx = node_index[node]
+            neigh_indices = [
+                node_index[nbr]
+                for nbr in neighbors_map.get(node, ())
+                if nbr in node_index
+            ]
+            if neigh_indices:
+                total = math.fsum(vf_list[i] for i in neigh_indices)
+                mean = total / len(neigh_indices)
+            else:
+                mean = vf_list[idx]
+            vf_new = vf_list[idx] + mu * (mean - vf_list[idx])
+            set_vf(G, node, clamp(float(vf_new), vf_min, vf_max))
+        return
+
+    work_items: list[tuple[Any, int, tuple[int, ...]]] = []
+    for node in eligible_nodes:
+        idx = node_index[node]
+        neigh_indices = tuple(
+            node_index[nbr] for nbr in neighbors_map.get(node, ()) if nbr in node_index
+        )
+        work_items.append((node, idx, neigh_indices))
+
+    approx_chunk = math.ceil(len(work_items) / jobs) if jobs else None
+    chunk_size = resolve_chunk_size(
+        approx_chunk,
+        len(work_items),
+        minimum=1,
+    )
+    chunks = [
+        work_items[i : i + chunk_size] for i in range(0, len(work_items), chunk_size)
+    ]
+    vf_tuple = tuple(vf_list)
+    updates: dict[Any, float] = {}
+    with ProcessPoolExecutor(max_workers=jobs) as executor:
+        args = ((chunk, vf_tuple, mu) for chunk in chunks)
+        for chunk_updates in executor.map(_vf_adapt_chunk, args):
+            for node, value in chunk_updates:
+                updates[node] = float(value)
+
+    for node in eligible_nodes:
+        vf_new = updates.get(node)
+        if vf_new is None:
+            continue
+        set_vf(G, node, clamp(float(vf_new), vf_min, vf_max))

tnfr/dynamics/adaptation.pyi

@@ -0,0 +1,7 @@
+from __future__ import annotations
+
+from ..types import TNFRGraph
+
+__all__ = ["adapt_vf_by_coherence"]
+
+def adapt_vf_by_coherence(G: TNFRGraph, n_jobs: int | None = None) -> None: ...

tnfr/dynamics/adaptive_sequences.py

@@ -0,0 +1,189 @@
+"""Adaptive sequence selection for TNFR operator trajectories.
+
+This module implements learning-based selection of operator sequences.
+Rather than executing fixed sequences, the system learns which sequences
+work best for given contexts and adapts its selection over time.
+
+The approach combines predefined canonical sequences with epsilon-greedy
+exploration to balance exploitation (use known good sequences) with
+exploration (try new patterns).
+"""
+
+from __future__ import annotations
+
+import random
+from typing import TYPE_CHECKING, Any, Dict, List
+
+if TYPE_CHECKING:
+    from ..types import TNFRGraph, NodeId
+
+try:
+    import numpy as np
+except ImportError:
+    np = None  # type: ignore[assignment]
+
+from ..config.operator_names import (
+    COHERENCE,
+    DISSONANCE,
+    EMISSION,
+    MUTATION,
+    RECEPTION,
+    RECURSIVITY,
+    RESONANCE,
+    SELF_ORGANIZATION,
+    SILENCE,
+    TRANSITION,
+)
+
+__all__ = ["AdaptiveSequenceSelector"]
+
+
+class AdaptiveSequenceSelector:
+    """Learns and selects optimal operator sequences based on context.
+
+    This class maintains a pool of canonical operator sequences and tracks
+    their performance over time. It uses epsilon-greedy selection to balance
+    exploitation of known-good sequences with exploration of alternatives.
+
+    **Selection Strategy:**
+
+    - **Exploitation (80%)**: Choose sequence with best historical performance
+    - **Exploration (20%)**: Random selection to discover new patterns
+
+    Parameters
+    ----------
+    graph : TNFRGraph
+        Graph containing the node
+    node : NodeId
+        Identifier of the node
+
+    Attributes
+    ----------
+    G : TNFRGraph
+        Graph reference
+    node : NodeId
+        Node identifier
+    sequences : dict[str, list[str]]
+        Pool of canonical operator sequences
+    performance : dict[str, list[float]]
+        Historical performance for each sequence
+
+    Examples
+    --------
+    >>> from tnfr.structural import create_nfr
+    >>> from tnfr.dynamics.adaptive_sequences import AdaptiveSequenceSelector
+    >>> G, node = create_nfr("test_node")
+    >>> selector = AdaptiveSequenceSelector(G, node)
+    >>> context = {"goal": "stability", "urgency": 0.5}
+    >>> sequence = selector.select_sequence(context)
+    >>> selector.record_performance("basic_activation", 0.85)
+    """
+
+    def __init__(self, graph: TNFRGraph, node: NodeId) -> None:
+        self.G = graph
+        self.node = node
+
+        # Canonical operator sequences
+        # Note: Sequences are designed to comply with TNFR grammar rules
+        self.sequences: Dict[str, List[str]] = {
+            "basic_activation": [EMISSION, COHERENCE],
+            "deep_learning": [EMISSION, RECEPTION, COHERENCE],
+            "exploration": [EMISSION, DISSONANCE, COHERENCE],
+            "consolidation": [COHERENCE, SILENCE, RECURSIVITY],
+            "mutation": [COHERENCE, MUTATION, TRANSITION, COHERENCE],
+        }
+
+        # Performance history: sequence_name -> [coherence_gains]
+        self.performance: Dict[str, List[float]] = {
+            k: [] for k in self.sequences.keys()
+        }
+
+    def select_sequence(self, context: Dict[str, Any]) -> List[str]:
+        """Select optimal sequence based on context and historical performance.
+
+        Uses goal-based filtering and epsilon-greedy selection:
+
+        1. Filter sequences appropriate for goal
+        2. With probability 0.8: select best-performing sequence
+        3. With probability 0.2: select random sequence (exploration)
+
+        Parameters
+        ----------
+        context : dict
+            Context with keys:
+
+            - **goal** (str): "stability", "growth", or "adaptation"
+            - **urgency** (float): Urgency level (0-1), currently unused
+
+        Returns
+        -------
+        list[str]
+            Sequence of operator names to execute
+
+        Notes
+        -----
+        Goal-to-sequence mapping follows TNFR principles:
+
+        - **stability**: Sequences emphasizing IL (Coherence) and SHA (Silence)
+        - **growth**: Sequences with AL (Emission) and THOL (Self-organization)
+        - **adaptation**: Sequences with ZHIR (Mutation) and learning cycles
+        """
+        goal = context.get("goal", "stability")
+
+        # Map goals to appropriate sequence candidates
+        if goal == "stability":
+            candidates = ["basic_activation", "consolidation"]
+        elif goal == "growth":
+            candidates = ["deep_learning", "exploration"]
+        elif goal == "adaptation":
+            candidates = ["mutation", "deep_learning"]
+        else:
+            candidates = list(self.sequences.keys())
+
+        # Epsilon-greedy selection (20% exploration, 80% exploitation)
+        epsilon = 0.2
+
+        if np is not None:
+            random_val = np.random.random()
+        else:
+            random_val = random.random()
+
+        if random_val < epsilon:
+            # Exploration: random selection
+            if np is not None:
+                selected = str(np.random.choice(candidates))
+            else:
+                selected = random.choice(candidates)
+        else:
+            # Exploitation: select best-performing sequence
+            avg_perf = {
+                k: (
+                    sum(self.performance[k]) / len(self.performance[k])
+                    if self.performance[k]
+                    else 0.0
+                )
+                for k in candidates
+            }
+            selected = max(avg_perf, key=avg_perf.get)  # type: ignore[arg-type]
+
+        return self.sequences[selected]
+
+    def record_performance(self, sequence_name: str, coherence_gain: float) -> None:
+        """Record performance metric for a sequence to enable learning.
+
+        Parameters
+        ----------
+        sequence_name : str
+            Name of the sequence that was executed
+        coherence_gain : float
+            Achieved coherence improvement or other performance metric
+
+        Notes
+        -----
+        Maintains a sliding window of the last 20 executions to adapt to
+        changing dynamics. Older performance data is discarded.
+        """
+        if sequence_name in self.performance:
+            self.performance[sequence_name].append(float(coherence_gain))
+            # Keep only last 20 executions (sliding window)
+            self.performance[sequence_name] = self.performance[sequence_name][-20:]

tnfr/dynamics/adaptive_sequences.pyi

@@ -0,0 +1,14 @@
+"""Type stubs for tnfr.dynamics.adaptive_sequences module."""
+
+from typing import Any, Dict, List
+from ..types import TNFRGraph, NodeId
+
+class AdaptiveSequenceSelector:
+    G: TNFRGraph
+    node: NodeId
+    sequences: Dict[str, List[str]]
+    performance: Dict[str, List[float]]
+
+    def __init__(self, graph: TNFRGraph, node: NodeId) -> None: ...
+    def select_sequence(self, context: Dict[str, Any]) -> List[str]: ...
+    def record_performance(self, sequence_name: str, coherence_gain: float) -> None: ...

tnfr/dynamics/aliases.py

@@ -0,0 +1,23 @@
+"""Shared alias tokens used across TNFR dynamics submodules."""
+
+from __future__ import annotations
+
+from ..constants.aliases import (
+    ALIAS_D2EPI,
+    ALIAS_DNFR,
+    ALIAS_DSI,
+    ALIAS_EPI,
+    ALIAS_SI,
+    ALIAS_THETA,
+    ALIAS_VF,
+)
+
+__all__ = (
+    "ALIAS_VF",
+    "ALIAS_DNFR",
+    "ALIAS_EPI",
+    "ALIAS_SI",
+    "ALIAS_THETA",
+    "ALIAS_D2EPI",
+    "ALIAS_DSI",
+)

tnfr/dynamics/aliases.pyi

@@ -0,0 +1,19 @@
+from ..constants.aliases import (
+    ALIAS_D2EPI as ALIAS_D2EPI,
+    ALIAS_DNFR as ALIAS_DNFR,
+    ALIAS_DSI as ALIAS_DSI,
+    ALIAS_EPI as ALIAS_EPI,
+    ALIAS_SI as ALIAS_SI,
+    ALIAS_THETA as ALIAS_THETA,
+    ALIAS_VF as ALIAS_VF,
+)
+
+__all__ = [
+    "ALIAS_VF",
+    "ALIAS_DNFR",
+    "ALIAS_EPI",
+    "ALIAS_SI",
+    "ALIAS_THETA",
+    "ALIAS_D2EPI",
+    "ALIAS_DSI",
+]