CardamomOT.inference.trajectory
===============================

.. py:module:: CardamomOT.inference.trajectory

.. autoapi-nested-parse::

   Core functions for the inference of trajectories, mainly used in loop_trajectories



Functions
---------

.. autoapisummary::

   CardamomOT.inference.trajectory.find_next_prot_mixed
   CardamomOT.inference.trajectory.find_next_prot
   CardamomOT.inference.trajectory.filter_network_fancy


Module Contents
---------------

.. py:function:: find_next_prot_mixed(d1, P0, M0, M1, mode_init, mode_end, alpha, s, delta_t)

   Deterministic flow interpolating between two points


.. py:function:: find_next_prot(d1, P0, M0, M1, mode_init, mode_end, alpha, s, delta_t)

   Deterministic flow interpolating between two points


.. py:function:: filter_network_fancy(T, N_traj, prot_traj, ks, basal_ref, inter_ref, seuil_intensity=0.05, seuil_variations=0.05, seuil_synergy=1.5, samples_data=None, n_jobs=-1)

   Structured three-pass edge filtering that avoids both the greedy
   order-dependence problem and the silent-synergy problem.

   Pass 1 — Canonical individual scoring
       Each edge is evaluated independently against the full reference
       network. Produces a continuous impact score per edge.

   Pass 2 — Conditional re-evaluation of rejected edges
       Edges rejected in pass 1 are re-evaluated inside the filtered
       network (without other rejected edges). An edge that was silent
       in the full network but becomes significant once dominant edges
       are removed was being masked — it is reinstated.

   Pass 3 — Pairwise synergy check among weak edges
       Among edges with low but non-zero individual scores, pairs are
       tested jointly. If removing both edges together produces an impact
       significantly greater than the sum of their individual impacts
       (super-additivity), both edges are retained.

   :param T: Number of time steps.
   :type T: int
   :param N_traj: Number of trajectories.
   :type N_traj: int
   :param prot_traj: Protein trajectories, shape (T * N_traj, G).
   :type prot_traj: array
   :param ks: Kinetic parameters.
   :type ks: array
   :param basal_ref: Basal transcription rates, shape (G, n_networks) or
                     (n_samples, G, n_networks).
   :type basal_ref: array
   :param inter_ref: Reference interaction matrix, shape (G, G, n_networks).
   :type inter_ref: array
   :param seuil_intensity: Minimum absolute edge weight to enter evaluation at all.
   :type seuil_intensity: float
   :param seuil_variations: Minimum variation score to retain an edge (main threshold).
   :type seuil_variations: float
   :param seuil_synergy: Multiplicative factor above which joint impact is considered
                         super-additive. An edge pair (e1, e2) is flagged as synergistic
                         if impact(e1 + e2 removed) > seuil_synergy * (score(e1) + score(e2)).
                         Default 2.0 (joint impact must be at least twice the additive sum).
   :type seuil_synergy: float
   :param samples_data: Additional per-sample data passed to _kon_per_sample.
   :type samples_data: optional
   :param n_jobs: Number of parallel jobs for joblib (-1 = all cores).
   :type n_jobs: int

   :returns: * **inter** (*array, shape (G, G, n_networks)*) -- Filtered interaction matrix (binary mask applied to inter_ref).
             * **inter_t** (*array, shape (T, G, G, n_networks)*) -- Time-resolved interaction tensor.
             * **report** (*dict*) -- Diagnostic report with keys:
               - 'scores_pass1'   : raw scores from pass 1, shape (G, G, n_networks)
               - 'reinstated'     : list of edges reinstated by pass 2
               - 'synergistic'    : list of edge pairs flagged as synergistic in pass 3
               - 'retained_mask'  : final boolean retention mask