Continuous parameter

As explained in Experimentation for Engineers

A system parameter whose value can vary over a continuous numerical range rather than being restricted to discrete choices.

A continuous parameter can take ordered numerical values with further values between any two choices. This structure lets an experiment search a range rather than merely compare a fixed set of variants. Response surface methodology (RSM) models the business metric between measured settings, then optimizes that model to estimate an optimum at an unmeasured value. section 4.0 section 4.1.1

Use domain constraints to define a feasible range before measuring. In the trading example, a threshold is limited to 0 < threshold < 2: negative values can make BUY and SELL conditions overlap, while values above 2 are rarely reached by a normally distributed signal. section 4.1.1

Interpolating a trading threshold

If markout profit is measured as 5 at threshold 1 and 6 at threshold 2, the midpoint threshold (1 + 2) / 2 can be assigned an initial interpolated estimate of 5.5. Two measurements alone place the interpolated maximum at an endpoint; measuring at least three points can reveal an optimum between measured settings. section 4.1.1

The same workflow extends to several tunable parameters, such as threshold and order size. Design choices, regression models, and numerical optimization become more involved as dimensions increase; RSM is best suited to three or fewer parameters. section 4.2

Figure 4.1 Overview of response surface methodology (RSM). RSM is an iterative process. First, in the design stage, we choose parameter values to measure. Next, we run the experiment to measure the business metric at those parameters. In the analyze stage, we use RSM techniques—interpolation and optimization—to estimate which parameter value is optimal. If the analysis fails, we return to the design stage. Otherwise, we validate the estimated optimal parameter value by measuring its business metric. If the measurement does not match the estimate, we return to the design stage. Otherwise, RSM has produced an optimal parameter value.

Trace the decision loop from selecting settings through production measurement, surrogate optimization, and validation-driven redesign. section 4.1

How Experimentation for Engineers covers it

How deep it goes

Defines ordered interior values, demonstrates interpolation, and develops one- and multi-parameter RSM procedures.

Examples

Uses a simulated trading strategy that tunes signal threshold and, later, order size against markout profit.

What it assumes

Builds from familiar A/B-testing measurement ideas while introducing continuous-parameter interpolation and optimization.

What you can do after

Apply it via response surface methodology to bound, measure, interpolate, optimize, and validate continuous settings.

How it connects

Connects response surface methodology with A/B testing, multi-armed bandits, surrogate functions, interpolation, linear regression, and numerical optimization.

liveBookmanning.com

Read Experimentation for Engineers on liveBook Get the book at manning.com

Related concepts

• response surface methodologyused by · A statistical optimization methodology that models a response as a function of controllable factors to identify promising settings.
• interpolationuses · Interpolation is the estimation or construction of values or representations between known points according to an assumed relationship or specified rule.
• surrogate modelused by · A surrogate model is an approximate model that estimates the behavior or outputs of a more complex, costly, or otherwise unavailable model, process, or function.

This explainer was generated by Manning's knowledge graph from the text of Experimentation for Engineers. The section links point to the parts of the book it draws on.