Table of Contents

DDM

Contents

Overview

The DDM Mechanism implements the “Drift Diffusion Model” (also know as the Diffusion Decision, Accumulation to Bound, Linear IntegratorFunction, and First Passage Time Model for a Wiener Process. This corresponds to a continuous version of the sequential probability ratio test (SPRT), that is the statistically optimal procedure for two alternative forced choice (TAFC) decision making (see drift-diffusion model in partciular).

The DDM Mechanism may be constructed with a choice of several functions that fall into to general categories: analytic solutions and path integration (see DDM Function Types below for more about these options.)

Creating a DDM Mechanism

A DDM Mechanism can be instantiated directly by calling its constructor, or by using the mechanism command and specifying DDM as its mech_spec argument. The model implementation is selected using the function argument. The function selection can be simply the name of a DDM function:

>>> import psyneulink as pnl
>>> my_DDM = pnl.DDM(function=pnl.DriftDiffusionAnalytical)

or a call to the function with arguments specifying its parameters:

>>> my_DDM = pnl.DDM(function=pnl.DriftDiffusionAnalytical(drift_rate=0.2, threshold=1.0))

Structure

The DDM Mechanism implements a general form of the decision process.

Input

The input to the function of a DDM Mechanism is always a scalar, irrespective of type of function that is used. Accordingly, the default InputPort for a DDM takes a single scalar value as its input, that represents the stimulus for the decision process. However, this can be configured using the input_format argument of the DDM’s consructor, to accomodate use of the DDM with other Mechanisms that generate a stimulus array (e.g., representing the stimuli associated with each of the two choices). By default, the input_format is SCALAR. However, if it is specified as ARRAY, the DDM’s InputPort is configured to accept a 1d 2-item vector, and to use Reduce as its Function, which subtracts the 2nd element of the vector from the 1st, and provides this as the input to the DDM’s function. If ARRAY is specified, two Standard OutputPorts are added to the DDM, that allow the result of the decision process to be represented as an array corresponding to the input array (see below).

Output

The DDM Mechanism can generate two different types of results depending on which function is selected. When a function representing an analytic solution is selected, the mechanism generates a single estimation for the process. When the path integration function is selected, the mechanism carries out step-wise integration of the process; each execution of the mechanism computes one step. (see DDM Function Types and Execution for additional details).

The value of the DDM Mechanism may have up to ten items. The first two of these are always assigned, and are represented by the DDM Mechanism’s two default OutputPorts`: DECISION_VARIABLE and RESPONSE_TIME. Other output_ports may be automatically assigned, depending on the function that has been assigned to the DDM, as shown in the table below:

The output_ports assigned to a DDM can explicilty determined by specifying ones from its list of standard_output_ports in the output_ports argument of its constructor, or the OUTPUT_PORTS entry of an OutputPort specification dictionary. This can include any of the OutputPorts listed above `DDM_Default_OutputPorts, as well as two additional ones – – DECISION_VARIABLE_ARRAY and SELECTED_INPUT_ARRAY – that are available if the ARRAY option is specified in its input_format argument (see Input). All of these Standard OutputPorts are listed in the DDM’s standard_output_ports attribute. As with any Mechanism, customized OutputPorts can also be created and assigned.

DDM Function Types

Analytic Solutions

The Drift Diffusion Model Functions that calculate analytic solutions [Bogacz et al (2006), Srivastava et al. (2016)] is DriftDiffusionAnalytical function, the mechanism generates a single estimate of the outcome for the decision process (see Execution for details). In addition to DECISION_VARIABLE and RESPONSE_TIME, the Function returns an accuracy value (represented in the PROBABILITY_UPPER_THRESHOLD OutputPort), and an error rate value (in the PROBABaILITY_LOWER_THRESHOLD OutputPort, and moments (mean, variance, and skew) for conditional (correctpositive or incorrectnegative) response time distributions. These are; the mean RT for correct responses (RT_CORRECT_MEAN, the RT variance for correct responses (RT_CORRECT_VARIANCE, the RT skew for correct responses (RT_CORRECT_SKEW, the mean RT for incorrect responses (RT_INCORRECT_MEAN, the RT variance for incorrect responses (RT_INCORRECT_VARIANCE, the RT skew for incorrect responses (RT_INCORRECT_SKEW.

An example that illustrate all of the parameters is shown below:

DriftDiffusionAnalytical Function:

>>> my_DDM_DriftDiffusionAnalytical = pnl.DDM(
...     function=pnl.DriftDiffusionAnalytical(
...         drift_rate=0.08928,
...         starting_value=0.5,
...         threshold=0.2645,
...         noise=0.5,
...         non_decision_time=0.15
...     ),
...     name='my_DDM_DriftDiffusionAnalytical'
... )

Path Integration

The Drift Diffusion Model Function that calculates a path integration is DriftDiffusionIntegrator. The DDM Mechanism uses the Euler method to carry out numerical step-wise integration of the decision process (see Execution below). In this mode, only the DECISION_VARIABLE and RESPONSE_TIME are available.

IntegratorFunction Function:

>>> my_DDM_path_integrator = pnl.DDM(
...     function=pnl.DriftDiffusionIntegrator(
...         noise=0.5,
...         initializer=1.0,
...         non_decision_time=2.0,
...         rate=3.0
...     ),
...     name='my_DDM_path_integrator'
... )

Execution

When a DDM Mechanism is executed, it computes the decision process either analytically or by numerical step-wise integration of its path. The method used is determined by its function (see DDM Function Types). The DDM’s function always returns values for the DECISION_VARIABLE and RESPONSE_TIME, and assigns these as the first two items of its value attribute, irrespective of its function.

When an analytic function is selected, the same set of values is returned for every execution. The returned values are determined entirely by the set of parameters passed to its function.

When the path integration, function is selected, a single step of integration is conducted each time the Mechanism is executed. The returned values accumulate on every execution.

The analytic functions return a final positon and time of the model, along with other statistics, where as the path integration function returns intermediate position and time values. The two types of functions can be thought of as happening on different time scales: trial (analytic) and time step (path integration).

References

Class Reference

class psyneulink.library.components.mechanisms.processing.integrator.ddm.DDM(default_variable=None, function=DriftDiffusionAnalytical)

Implements a drift diffusion process (also known as the Diffusion Decision Model, either by calculating an analytic solution or carrying out step-wise numerical integration. See Mechanism for additional arguments and attributes.

The default behaviour for the DDM is to reset its integration state on each new trial, this can be overridden by setting the reset_stateful_function_when attribute to a different condition. In addition, unlike TransferMechamism, the DDM’s execute_until_finished attribute is set to True by default. This will cause the DDM to execute multiple time steps per call to its execute method until the decision threshold is reached. This default behavior can be changed by setting execute_until_finished to False.

Parameters

  • default_variable (value, list or np.ndarray : default FUNCTION_PARAMS[STARTING_VALUE]) – the input to the Mechanism used if none is provided in a call to its execute or run methods; also serves as a template to specify the length of the variable for its function, and the primary OutputPort of the DDM (see Input <DDM_Creation>` for how an input with a length of greater than 1 is handled).

  • function (IntegratorFunction : default DriftDiffusionAnalytical) – specifies the function to use to execute the decision process; determines the mode of execution (see function and DDM Function Types for additional information).

variable

the input to Mechanism’s execute method. Serves as the “stimulus” component of the function’s drift_rate parameter.

Type

value : default FUNCTION_PARAMS[STARTING_VALUE]

function

the function used to execute the decision process; determines the mode of execution. If it is DriftDiffusionAnalytical, an analytic solution is calculated (note: the latter requires that the MatLab engine is installed); if it is an IntegratorFunction Function with an integration_type of DIFFUSION, then numerical step-wise integration is carried out. See DDM Function Types and Execution for additional information.

Type

IntegratorFunction : default DriftDiffusionAnalytical

value

result of executing DDM function; has six items, that are assigned based on the function attribute. The first two items are always assigned the values of DECISION_VARIABLE and RESPONSE_TIME (though their interpretation depends on the function and corresponding mode of of operation). See DDM Function Types, Execution, and Output for additional information about other values that can be reported and their interpretation.

Type

2d np.array[array(float64),array(float64),array(float64),array(float64)]

random_state

private pseudorandom number generator

Type

numpy.RandomState

output_ports

list of the DDM’s OutputPorts. There are always two OutputPorts, DECISION_VARIABLE and RESPONSE_TIME; additional ones may be included based on the function and/or any specifications made in the output_ports argument of the DDM’s constructor (see Output for additional details).

Type

ContentAddressableList[OutputPort]

output_values

each item is the value of the corresponding OutputPort in output_ports. The first two items are always the values of the DECISION_VARIABLE and RESPONSE_TIME OutputPorts; additional ones may be included, based on the function and any specifications made in the output_ports argument of the DDM’s constructor (see Output for additional details).

Type

List[array(float64)]

standard_output_ports

list of Standard OutputPorts that includes the following addition to the standard_output_ports of a Mechanism:

DECISION_VARIABLEfloat

  • analytic mode: the value of the threshold crossed by the decision variable on the current TRIAL (which is either the value of the DDM function’s threshold attribute or its negative);

  • integration mode: the value of the decision variable at the current TIME_STEP of execution.

Corresponds to the 1st item of the DDM’s value.

DECISION_VARIABLE_ARRAY1d nparray

Note

This is available only if input_format is specified as ARRAY in the DDM Mechanism’s constructor (see Input).

  • analytic mode: two element array, with the decision variable (1st item of the DDM’s value) as the 1st element if the decision process crossed the upper threshold, and the 2nd element if it is closer to the lower threshold; the other element is set to 0.

  • integration mode: the value of the decision variable at the current TIME_STEP of execution, assigned to the 1st element if the decision variable is closer to the upper threshold, and to the 2nd element if it is closer to the lower threshold; the other element is set to 0.

Type

list[str]

SELECTED_INPUT_ARRAY1d nparray

Note

This is available only if input_format is specified as ARRAY in the DDM Mechanism’s constructor (see Input).

  • analytic mode: two element array, with one (“selected”) element – determined by the outcome of the decision process – set to the value of the corresponding element in the stimulus array (i.e., the DDM’s input_port variable). The “selected” element is the 1st one if the decision process resulted in crossing the upper threshold, and the 2nd if it crossed the lower threshold; the other element is set to 0.

  • integration mode: the value of the element in the stimulus array based on the decision variable (1st item of the DDM’s value) at the current TIME_STEP of execution: it is assigned to the 1st element if the decision variable is closer to the upper threshold, and to the 2nd element if the decision variable is closer to the lower threshold; the other element is set to 0.

RESPONSE_TIMEfloat

  • analytic mode: mean time (in seconds) for the decision variable to reach the positive or negative value of the DDM function’s threshold attribute as estimated by the analytic solution calculated by the function);

  • integration mode: the number of TIME_STEP that have occurred since the DDM began to execute in the current TRIAL or, if it has reached the positive or negative value of the DDM function’s threshold attribute, the TIME_STEP at which that occurred.

Corresponds to the 2nd item of the DDM’s value.

PROBABILITY_UPPER_THRESHOLDfloat

  • analytic mode: the probability of the decision variable reaching the positive value of the DDM function’s threshold attribute as estimated by the analytic solution calculated by the function; often, by convention, the positive (upper) threshold is associated with the correct response, in which case PROBABILITY_UPPER_THRESHOLD corresponds to the accuracy of the decision process.

  • integration mode: None.

Corresponds to the 3rd item of the DDM’s value.

PROBABILITY_LOWER_THRESHOLDfloat

  • analytic mode: the probability of the decision variable reaching the negative value of the DDM function’s threshold attribute as estimated by the analytic solution calculate by the function); often, by convention, the negative (lower) threshold is associated with an error response, in which case PROBABILITY_LOWER_THRESHOLD corresponds to the error rate of the decision process;

  • integration mode: None.

Corresponds to the 4th item of the DDM’s value.

RT_CORRECT_MEANfloa

(only applicable if function is DriftDiffusionAnalytical)

Corresponds to the 5th item of the DDM’s value.

RT_CORRECT_VARIANCEfloat

(only applicable if function is DriftDiffusionAnalytical)

Corresponds to the 6th item of the DDM’s value.

RT_CORRECT_SKEWfloat

(only applicable if function is DriftDiffusionAnalytical)

Corresponds to the 7th item of the DDM’s value.

RT_INCORRECT_MEANfloat

(only applicable if function is DriftDiffusionAnalytical)

Corresponds to the 5th item of the DDM’s value.

RT_INCORRECT_VARIANCEfloat

(only applicable if function is DriftDiffusionAnalytical)

Corresponds to the 6th item of the DDM’s value.

RT_INCORRECT_SKEWfloat

(only applicable if function is DriftDiffusionAnalytical)

Corresponds to the 7th item of the DDM’s value.

plot(stimulus=1.0, threshold=10.0)

Generate a dynamic plot of the DDM integrating over time towards a threshold.

Note

The plot method is only available when the DriftDiffusionIntegrator function is in use. The plot method does not represent the results of this DDM mechanism in particular, and does not affect the current state of this mechanism’s DriftDiffusionIntegrator. The plot method is only meant to visualize a possible path of a DDM mechanism with these function parameters.

Parameters

  • stimulus (float: default 1.0) – specify a stimulus value for the AdaptiveIntegrator function

  • threshold (float: default 10.0) – specify the threshold at which the DDM will stop integrating

Returns

Mechanism’s function plot – Matplotlib window of the Mechanism’s function plotting dynamically over time with specified parameters towards a specified threshold

Return type

Matplotlib window

_validate_variable(variable, context=None)

Ensures that input to DDM is a single value. Remove when MULTIPROCESS DDM is implemented.

_validate_params(request_set, target_set=None, context=None)

validate TimeScale, INPUT_PORTS, FUNCTION_PARAMS, OUTPUT_PORTS and MONITOR_FOR_CONTROL

Go through target_set params (populated by Component._validate_params) and validate values for:
  • INPUT_PORTS:

    <MechanismsInputPort or Projection object or class, specification dict for one, 2-item tuple, or numeric value(s)>; if it is missing or not one of the above types, it is set to self.defaults.variable

  • FUNCTION_PARAMS: <dict>, every entry of which must be one of the following:

    ParameterPort or Projection object or class, specification dict for one, 2-item tuple, or numeric value(s); if invalid, default is assigned

  • OUTPUT_PORTS:

    <MechanismsOutputPort object or class, specification dict, or numeric value(s); if it is missing or not one of the above types, it is set to None here;

    and then to default value of value (output of execute method) in instantiate_output_port (since execute method must be instantiated before self.defaults.value is known)

    if OUTPUT_PORTS is a list or OrderedDict, it is passed along (to instantiate_output_ports) if it is a OutputPort class ref, object or specification dict, it is placed in a list

  • MONITORED_PORTS:

    ** DOCUMENT

Note: PARAMETER_PORTS are validated separately – ** DOCUMENT WHY

TBI - Generalize to go through all params, reading from each its type (from a registry),

and calling on corresponding subclass to get default values (if param not found) (as PROJECTION_TYPE and PROJECTION_SENDER are currently handled)

_execute(variable=None, context=None, runtime_params=None)

Execute DDM function (currently only trial-level, analytic solution) Execute DDM and estimate outcome or calculate trajectory of decision variable Currently implements only trial-level DDM (analytic solution) and returns:

  • stochastically estimated decion outcome (convert mean ER into value between 1 and -1)

  • mean ER

  • mean RT

  • mean, variance, and skew of RT for correct (posititive threshold) responses

  • mean, variance, and skew of RT for incorrect (negative threshold) responses

Return current decision variable (self.outputPort.value) and other output values (self.output_ports[].value Arguments: # CONFIRM: variable (float): set to self.value (= self.input_value) - params (dict): runtime_params passed from Mechanism, used as one-time value for current execution:

  • DRIFT_RATE (float)

  • THRESHOLD (float)

  • kwDDM_Bias (float)

  • NON_DECISION_TIME (float)

  • NOISE (float)

  • context (str)

Returns the following values in self.value (2D np.array) and in

the value of the corresponding outputPort in the self.output_ports dict: - decision variable (float) - mean error rate (float) - mean RT (float) - mean, variance, and skew of RT for correct (posititive threshold) responses - mean, variance, and skew of RT for incorrect (negative threshold) responses

Parameters

self –

:param variable (float) :param params: (dict) :param context: (str) :rtype self.outputPort.value: (number)

reset(*args, force=False, context=None, **kwargs)

Reset value if Mechanisms is stateful.

If the mechanism’s function is an IntegratorFunction, or if the mechanism has and integrator_function (see TransferMechanism), this method effectively begins the function’s accumulation over again at the specified value, and updates related attributes on the mechanism. It also clears the value history, thus effectively setting the previous value to None.

If the mechanism’s function is an IntegratorFunction, its reset method:

  1. Calls the function’s own reset method (see Note below for details)

  2. Sets the mechanism’s value to the output of the function’s reset method

  3. Updates its output ports based on its new value

If the mechanism has an integrator_function, its reset method:

(1) Calls the `integrator_function's <TransferMechanism.integrator_function>` own `reset
    <IntegratorFunction.reset>` method (see Note below for details)

(2) Executes its `function <Mechanism_Base.function>` using the output of the `integrator_function's
    <TransferMechanism.integrator_function>` `reset <IntegratorFunction.reset>` method as
    the function's variable

(3) Sets the mechanism's `value <Mechanism_Base.value>` to the output of its function

(4) Updates its `output ports <Mechanism_Base.output_port>` based on its new `value
    <Mechanism_Base.value>`

Note

The reset method of an IntegratorFunction Function typically resets the function’s previous_value (and any other stateful_attributes) and value to the quantity (or quantities) specified. If reset is called without arguments, the initializer value (or the values of each of the attributes in initializers) is used instead. The reset method may vary across different Integrators. See individual functions for details on their stateful_attributes, as well as other reinitialization steps that the reset method may carry out.

is_finished(context=None)

Set by a Component to signal completion of its execution in a TRIAL; used by Component-based Conditions to predicate the execution of one or more other Components on a Component.

exception psyneulink.library.components.mechanisms.processing.integrator.ddm.DDMError(error_value)