pyphi.landscape#
Continuous-parameter analysis of IIT quantities over substrate space.
Every quantity PyPhi computes is a function of the substrate’s parameters
(connection weights, TPM entries). This module evaluates those functions
along one parameter axis: landscape_section() computes a system
irreducibility analysis at every point of a 1-D grid and returns a tidy
table that tracks not only φ but the identity of every discrete selection
behind it (MIP partition, specified cause and effect states) and the
selection margins; perturb() estimates local derivatives at a single
point by central finite differences.
The φ landscape is piecewise-smooth: within a region where every selection
picks the same winner (a selection regime), φ is a smooth function of the
parameters, but the reported φₛ can jump where the MIP or a specified state
switches. The regime column and LandscapeSection.boundaries
locate those switches; Perturbation.same_regime flags a derivative
estimate that straddles one. Because the raw φₛ jumps at MIP switches while
the normalized value stays continuous, signed_phi and
signed_normalized_phi are the better-behaved quantities for numerical
work — the positive-part clamp makes phi exactly flat wherever the raw
integration is negative.
A parameter axis is any callable mapping a float to a
Substrate; weight_axis() builds one for the
common case of varying a single connection weight of a
build_substrate() substrate.
Functions
|
Evaluate the system irreducibility analysis along a parameter grid. |
|
Estimate the local derivative of an IIT quantity at one point. |
|
Return a parameter axis varying one weight of a generated substrate. |
Classes
|
A 1-D section of the φ landscape along one parameter axis. |
|
Local finite-difference behavior of one quantity at one point. |