How to Optimize Your System with Exploratory Data Analysis

We all feel more comfortable when we know how things work. That is so partly because that same knowledge makes us feel more confident as to the response those things will provide and, thus, allows us to anticipate.

Can AutoDiscovery help us adjusting our system in order to optimize the response we get?

The answer is YES and now I will proceed to explain you how to do it.

A best practice

In order to illustrate the application field of AutoDiscovery in an optimization process, we will use a simulation based on real data acquired during a neurogenesis study.

Basically, the object of this experiment consists in understanding how learning and memory are impacted by neural structure and anxiety in rats, our experimental subjects.

In order to reach our goal, we count on four groups of information clearly differentiated:

• The characteristics of each experimental subject.

• The number of each subject’s different neuron types.

• The results of each subject’s anxiety tests.

• The results of each subject’s learning/memory test.

These four groups are precisely the data sources of our project.

Our aim now will be to identify in which circumstances the best learning and memory results are achieved.

The recipe to optimization

As previously commented, the optimization process consists in adjusting part of the system in order to make other parts respond the way we require. Consequently, before we get started with an optimization process we need to clearly establish:

1. What part of the system we are to optimize?

That is, which variables (out of our dataset) are the ones we need to get closer to, so we may work over them.

In our case, these variables are precisely the results coming from the learning and memory tests, with special attention to the “balance” variables between problem solving speed (learning) and memories duration (memory) of the experimental subjects.

Thus, we will tag them as “response variables” from this moment on.

2. What part of the system may be adjusted (or are they already adjusted?)

That is, which variables may be stablished manually and/or which come preset from the beginning.

In our example, only “subject group” is under our control. It indirectly determines the conditions we have made the experimental subject to work under.

3. What kind of response we wish to obtain?

Basically, whether we want to obtain the maximum or the minimum value out of the “response variables”.

As previously stated, in our case we want to look for the best results or, in other words, the maximum “response variables” values.

Setting up AutoDiscovery: the response variables

Once all the information has been aggregated to AutoDiscovery, we need to indicate which variables are the ones to be optimized, that is, which are our response variables.

The “Variables to Explain” option in the “Configure Variables” window of AutoDiscovery provides us with precisely that possibility.

At this moment, the discovery process will be limited to obtain a description of the interactions amongst all system variables and our selection of response variables.

Asking AutoDiscovery to describe the mechanisms of the response

The discovery process we launched will now allow us to describe the mechanisms of our response variables. That is, understand and explain which variables intervene in the behavior of the system, what kind of relationships there are among them, and in which particular circumstances those relationships take place.

In that sense, we will use the Discovery Map, which shows us that, indeed, there exist relationships among the response variables (the results we got from both the learning and memory tests) and the adjusted variables (subject properties).

In order to focus on the concrete relationships between the subject and the learning/memory balance, we will use the Hypo Booster tool:

Identifying the optimal circumstances

With our focus set on these relationships, we only need to look for those circumstances or conditions in which the response variables show their maximum value.

In our example, we can identify that the different experimental subject groups show a significantly different learning/memory balance specifically in day 5 of the experiment:

The previous chart suggests that the maximum balance takes place in the subjects of group “P1”, which means that, from the exploratory point of view, everything seems to aim at these particular circumstances as the origin to that optimal response.

Conclusions

The capacity of AutoDiscovery to describe the behavior of a complex system from an exploratory point of view is useful when we are to identify the potential circumstances in which the best response is achieved.

As always, a confirmatory study will help us validate that promising result.

So ... do you dare trying with your own data? You cannot imagine what you could discover today ...