Objective Search is one of the most powerful tools in the platform to search through all experiments for the ones that come closest to meeting a specification in your project. To access the Objective Search page click on the Calculate tab and select Objective Search.

If you have already set up a project default spec, that spec will show up on this page automatically. If not, you can either click the “Load Spec” to load spec from the current project or the “Import Spec” to load spec from other projects. A step-by-step guide on how to properly set up a project spec is listed in this article.

If you are inside a project, the objective search will look through all formulations within your project. This is the most common use case.

If you are under the “All Projects” view, the objective search will look through all formulations under the current material family. This is used less often.

Loading specs and applying filters

Once a spec is loaded, you can change priorities, settings (like upper limit, lower limit, range, goal … ) for the goals you are optimizing for. You can add hard filters to the search by clicking the purple “Add Filter” button. This filter function is consistent with the ones across our platform on several different pages. You can find a detailed article on filter function here. For instance, you might want the search to return only experiments that include a certain polymer and the cost of that formulation to be within a range (shown in the screenshot below).

You want to be aware that filters here are hard/strict filters. Thus, any formulations that do not meet those criteria will be filtered out from the objective search. If there are a bunch of filters applied on this page, you might be searching in a space which is much smaller than what you might be anticipating.

Interpreting search results

Objective search is most useful for answering questions like:

• What are the best formulations so far?
• Are there formulations that meet the specifications already? If not, which are the ones that are closest to meeting all specifications? And how close are those?

Once your preferences are defined click the Get Results button to evaluate every measurement recorded.

From there, Uncountable platform will evaluate every measurement recorded against the goal in the spec, assigning it a score that denotes how close it was to meeting the specifications. Formulations that are closer to the specifications will be ranked higher and have a higher score while the ones that are further away from meeting the specification will be ranked lower and have a lower score.

Results that are far away from the goal are penalized, and ones that meet or exceed the goal are rewarded. Every result is evaluated on all experiments, scores are summed over the full list for each experiment, and then experiments are ranked from closest to furthest away from the specification. For missing measurement values, an average score is computed across all other experiments and that average score is used.

Not only upper and lower limits are taken into account, but the priority of each goal is also taken into consideration. Setting all goals as “High” is equal to setting all goals as “Medium” or “Low”. In the above 2 formulations (YG Screening – 5 and 6), density and elongation at break are both set as Medium. At this stage, their loss is about the same for both formulations. If I change density to be high priority and elongation at break as low priority, formulations with higher density (YG Screening – 6) will get ranked higher compared to YG Screening – 5 because of its density is lower. Thus, it is always best practice to set some goals as High, some as Medium and some as Low. Creating some gradient on different goals here is helpful for the Uncountable platform to differentiate between different formulations.

If you mouse over to one of the measurement cells, you will get a breakdown of its loss (shown in screenshot below).

Measurement Value – its value under that measurement. It will be empty if nothing is recorded.

Passes Goal – Yes or No depends on its measurement value and how the goal is set up for that measurement.

Loss Contribution – This indicates how much that measurement value contributes to the loss function on a per formulation basis. It should fall roughly within 0-100%.

Inferred –  Yes or No depends on if that measurement value is missing. If the value is missing, it would be Yes (an inferred value will be used to calculate the loss for this formulation). Otherwise it would be No.

Per measurement, you can set upper and lower limits at different values and see how that impacts the loss contribution. For the ones that you are optimizing for/struggling the most with/hoping to improve the most in the next round, you want to set their upper/lower limits at a place where the loss contribution (for the top 5 or 10 formulations) comes from those measurements the most. Generally, you would want to fine tune priority and upper/lower limit of measurements so that the top 5 to 10 formulations generated by the Uncountable platform agrees roughly with your thought.

By clicking on the Compare button, you will be taken to a compare view like below, where you can view every detail (all inputs and measurements will be displayed) of your selected experiments (by default, the top 10 experiments are selected). You can think these 10 are the best selected by Uncountable platform based on the spec you set up above. The spec you selected will be displayed as a separate column right next to outputs. The color of each measurement cell will indicate whether the result has passed or failed the desired spec; red indicating a failure and green indicating a pass.

Show recipes outside project

Sometimes, it might be necessary to see the performance of recipes that are NOT in the current project.

If you are NOT under the “All Projects” view, you can check the Show Recipes Outside Project box and then Refresh Results. This option allows you to do objective search under the current material family (filters you added above would still be respected). This way, the search does not only apply to all experiments within the project, but to all other experiments within the material family. “Show Recipes Outside Project” is a quick way of seeing what others have done on the outputs in your specification.

2 pages that are closely related to objective search

At the bottom of the objective search page, you will see 2 buttons: View Tradeoffs and Suggest Experiments. The two buttons will take you to view tradeoffs and suggest experiments with AI pages on the Uncountable platform.

View Tradeoffs – It is fairly common to see tradeoffs between various measurements in the spec. Section 3.3 in this article gives more explanation on the purpose and how to use the View Tradeoff page.

Suggest Experiments – After reviewing the objective search results, you might want to design a new round of experiments. A detailed description on how to use that page can be found in this article.