Introduction to Simulation Configurations

The following has been copy-pasted from the documentation available in GitHub. Some of the relative links broke in the process, but we thought it made sense to have this is a thread here to make it more easily accessible and for feedback.


Simulation Configuration

Introduction

Given a Simulation Configuration, cadCAD produces datasets that represent the evolution of the state of a system
over discrete time. The state of the
system is described by a set of State Variables. The dynamic of the system is described by
Policy Functions and State Update Functions, which are evaluated by
cadCAD according to the definitions set by the user in Partial State Update Blocks.

A Simulation Configuration is comprised of a System Model and a set of
Simulation Properties

append_configs, stores a Simulation Configuration to be Executed by cadCAD

from cadCAD.configuration import append_configs

append_configs(
    initial_state = ..., # System Model
    partial_state_update_blocks = .., # System Model
    policy_ops = ..., # System Model
    sim_configs = ... # Simulation Properties
)

Parameters:

Simulation Properties

Simulation properties are passed to append_configs in the sim_configs parameter. To construct this parameter, we
use the config_sim function in cadCAD.configuration.utils

from cadCAD.configuration.utils import config_sim

c = config_sim({
    "N": ...,
    "T": range(...),
    "M": ...
})

append_configs(
    ...
    sim_configs = c # Simulation Properties
)

T - Simulation Length

Computer simulations run in discrete time:

Discrete time views values of variables as occurring at distinct, separate “points in time”, or equivalently as being
unchanged throughout each non-zero region of time (“time period”)—that is, time is viewed as a discrete variable. (…)
This view of time corresponds to a digital clock that gives a fixed reading of 10:37 for a while, and then jumps to a
new fixed reading of 10:38, etc.
(source: Wikipedia)

As is common in many simulation tools, in cadCAD too we refer to each discrete unit of time as a timestep. cadCAD
increments a “time counter”, and at each step it updates the state variables according to the equations that describe
the system.

The main simulation property that the user must set when creating a Simulation Configuration is the number of timesteps
in the simulation. In other words, for how long do they want to simulate the system that has been modeled.

N - Number of Runs

cadCAD facilitates running multiple simulations of the same system sequentially, reporting the results of all those
runs in a single dataset. This is especially helpful for running
Monte Carlo Simulations.

M - Parameters of the System

Parameters of the system, passed to the state update functions and the policy functions in the params parameter are
defined here. See System Model Parameter Sweep for more information.

System Model

The System Model describes the system that will be simulated in cadCAD. It is comprised of a set of
State Variables and the State Update Functions that determine the
evolution of the state of the system over time. Policy Functions (representations of user policies
or internal system control policies) may also be part of a System Model.

State Variables

A state variable is one of the set of variables that are used to describe the mathematical “state” of a dynamical
system. Intuitively, the state of a system describes enough about the system to determine its future behaviour in the
absence of any external forces affecting the system. (source: Wikipedia)

cadCAD can handle state variables of any Python data type, including custom classes. It is up to the user of cadCAD to
determine the state variables needed to sufficiently and accurately describe the system they are interested in.

State Variables are passed to append_configs along with its initial values, as a Python dict where the dict_keys
are the names of the variables and the dict_values are their initial values.

from cadCAD.configuration import append_configs

genesis_states = {
    'state_variable_1': 0,
    'state_variable_2': 0,
    'state_variable_3': 1.5,
    'timestamp': '2019-01-01 00:00:00'
}

append_configs(
    initial_state = genesis_states,
    ...
)

State Update Functions

State Update Functions represent equations according to which the state variables change over time. Each state update
function must return a tuple containing a string with the name of the state variable being updated and its new value.
Each state update function can only modify a single state variable. The general structure of a state update function is:

def state_update_function_A(_params, substep, sH, s, _input):
    ...
    return 'state_variable_name', new_value

Parameters:

  • _params : dict - System parameters
  • substep : int - Current substep
  • sH : list[list[dict]] - Historical values of all state variables for the simulation. See
    Historical State Access for details
  • s : dict - Current state of the system, where the dict_keys are the names of the state variables and the
    dict_values are their current values.
  • _input : dict - Aggregation of the signals of all policy functions in the current
    Partial State Update Block

Return:

  • tuple containing a string with the name of the state variable being updated and its new value.

State update functions should not modify any of the parameters passed to it, as those are mutable Python objects that
cadCAD relies on in order to run the simulation according to the specifications.

Policy Functions

A Policy Function computes one or more signals to be passed to State Update Functions
(via the _input parameter). Read
this article
for details on why and when to use policy functions.

The general structure of a policy function is:

def policy_function_1(_params, substep, sH, s):
    ...
    return {'signal_1': value_1, ..., 'signal_N': value_N}

Parameters:

  • _params : dict - System parameters
  • substep : int - Current substep
  • sH : list[list[dict]] - Historical values of all state variables for the simulation. See
    Historical State Access for details
  • s : dict - Current state of the system, where the dict_keys are the names of the state variables and the
    dict_values are their current values.

Return:

Policy functions should not modify any of the parameters passed to it, as those are mutable Python objects that cadCAD
relies on in order to run the simulation according to the specifications.

At each Partial State Update Block (PSUB), the dicts returned by all policy functions
within that PSUB dictionaries are aggregated into a single dict using an initial reduction function
(a key-wise operation, default: dic1['keyA'] + dic2['keyA']) and optional subsequent map functions. The resulting
aggregated dict is then passed as the _input parameter to the state update functions in that PSUB. For more
information on how to modify the aggregation method, see Policy Aggregation.

Partial State Update Blocks

A Partial State Update Block (PSUB) is a set of State Update Functions and Policy Functions such that State Update
Functions in the set are independent from each other and Policies in the set are independent from each other and from
the State Update Functions in the set. In other words, if a state variable is updated in a PSUB, its new value cannot
impact the State Update Functions and Policy Functions in that PSUB - only those in the next PSUB.

Partial State Update Blocks are passed to append_configs as a List of Python dicts where the dict_keys are named
"policies" and "variables" and the values are also Python dicts where the keys are the names of the policy and
state update functions and the values are the functions.

PSUBs = [
    {
        "policies": {
            "b_1": policy_function_1,
            ...
            "b_J": policy_function_J
        },
        "variables": {
            "s_1": state_update_function_1,
            ...
            "s_K": state_update_function_K
        }
    }, #PSUB_1,
    {...}, #PSUB_2,
    ...
    {...} #PSUB_M
]

append_configs(
    ...
    partial_state_update_blocks = PSUBs,
    ...
)

Substep

At each timestep, cadCAD iterates over the partial_state_update_blocks list. For each Partial State Update Block,
cadCAD returns a record containing the state of the system at the end of that PSUB. We refer to that subdivision of a
timestep as a substep.

Result Dataset

cadCAD returns a dataset containing the evolution of the state variables defined by the user over time, with three int
indexes:

  • run - id of the run
  • timestep - discrete unit of time (the total number of timesteps is defined by the user in the
    T Simulation Parameter)
  • substep - subdivision of timestep (the number of substeps is the same as the number of Partial State
    Update Blocks)

Therefore, the total number of records in the resulting dataset is N x T x len(partial_state_update_blocks)

System Simulation Execution

4 Likes

The link to system model parameter sweeping is broken both here and in gitHub. I’m having a hard time getting the M input to work the way I would expect. Is there documentation or an example somewhere?

Thanks for bringing this up. We fixed the links in the documentation. Parameter sweeping is described here. You can also find an example in this Gist (see how params['r'] is used inside state update function func_b)

1 Like