Module mcmc_indexclass
source code
This module provides several classes to manage the parameters of an
algorithm, particularly so that you can run the mcmc.py optimizer on it.
Essentially, the main problem that it solves is how to keep track of
human-readable names assigned to components of a vector of
parameters.
You have an index_base object i, and out can get a
parameter from it like this: i.p("velocity",
"car", "x"). That would get out a number
that might correspond to the x-velocity of a car. That's the
human-readable side, but you can also get a vector of all the parameters
via i.get_prms() and get a mapping between names and indices
into that vector via i.map.
You can also have parameters that do not appear in the vector of
adjustable parameters. These are known as "fixed" parameters;
their intent is to let you fix some of the parameters of an optimization,
and let the optimizer play with the remainder.
Use guess first to establish the mapping, write it to a file
and guess reasonable starting values for an optimization. Then, use index
to get parameter values corresponding to names.
|
|
|
|
|
|
|
indexer
|
reindex(q,
ref)
This remaps the parameters of q so that they are in the
same order as in ref. |
source code
|
|
|
|
Debug = False
|
|
|
__package__ = 'gmisclib' |
Imports:
re,
os,
numpy,
die,
dictops,
g_encode
|
This remaps the parameters of q so that they are in the
same order as in ref.
- Returns: indexer
- an indexer that contains the same data as
q, but
shares its parameter mapping with ref.
Note:
The parameters in q must be a superset of the
adjustable parameters in ref. Extra parameters in
q will be carried over as fixed parameters.
|