I interpret that Fuzzy Selector is superior to Strict Selector since it is the default targetting method, but the way they are presented makes Strict Selector appear superior.
Editing a Strict Selector via the property panel provides options to validate, indicate, repair and highlight along with a color-coded modular overview of individual attributes and one-click away to edit using UI Explorer. Fuzzy Selectors, however, just open up in a plain text field, and it is not possible to edit the settings/attributes “Text Is / Ignore Text” and “Accuracy” from the property panel.
If Fuzzy Selector is intended to be superior, I suggest making the editing experience on par with Strict Selector editing experience.
Fuzzy selectors should be avoided as much as possible. I’d rather see then phasing them out completely, since they are dangerous for your automation.
Web pages often have multiple elements that look alike, both in name and appearance. But the impact of entering data, or clicking on items wrongly can have a huge impact on a process. So letting UiPath target an element that somewhat looks like but isn’t a 100% match should never be allowed. And, to be honest, luckily I have never even needed a ‘fuzzy’, all my selector targets I was always able to resolve using strict.
So strict is indeed the ‘superior’ one.
If you want your studio to prefer or even ignore fuzzy during element selection, it can be changed in your project settings, under Modern UI settings.
First of all, thank you for raising this, it’s a very healthy discussion to have. As part of the UIAutomation team, let me address a few of the issues you raised.
Fuzzy selectors are recommended by default especially for Web (where Strict is disabled), but also for Desktop and Java. For SAP we recommend using Strict selectors alone:
Point taken. While I can’t promise you a definitive release date, I can say we are very aware of this and improving the Fuzzy Selector experience is in our (mid-term) plans.
Nice that you recommend this, but can you argue ‘why’ fuzzy is recommended for web (or any other application layer for that matter), especcially put off against the argument I raised on the possible effect of a fuzzy selector?
This is a sign we’re not doing a good job at explaining our tools, so thank you, we really appreciate this kind of (blunt) feedback, even if it’s not the easiest for us to read.
Strict selector can also be less than 100% accurate, depending on the attributes you use. Fuzzy Selectors are meant to be used with anchors that start acting as extra attributes, along with the geometrical conditions (between the target and the anchor(s)) we use. In a way, any targeting system is a game of balancing precision with recall: yes, you do want to have as less false positives as possible - preferrably zero - but you also need resilience in time when facing UI changes. We never aimed at generating a different balance point with Fuzzy vs. Strict: ideally, we want them to be equally balanced; it’s just that for Fuzzy you’ll need to leverage anchors a lot more.
Looking at the telemetry numbers, Strict is still a bit better in terms of precision, but the differences are very small vs. Fuzzy on all use-cases. So I’d say Fuzzy is not that “dangerous” but we’ll need to document and promote this a lot better than we do now.
That’s great, this means we were able to cover your target scenarios with a single tool - we’re always glad when users are happy
We think no targeting method is truly “superior” unless we’re dealing with specific use-cases; this is why we tried to recommend the default values above per use-case (Web/Desktop/Java/SAP). In what we call pure CV scenarios (Citrix automations with no access to selectors), Computer Vision targeting is king (it’s better than Image). So it all depends on your target scenarios.
Trying to find the best targeting method is like trying to find the best car: an F1 car is great on a track, but not so great in the Dakar Rally
Thank you, good to know it is among your plans! And thank you for informing me about the different recommendations depending on web/desktop/java/SAP. That graphic has eluded me. Which guide/page is it from?
Agreed. I found that setting Accuracy to 1 (100%) neutralize this. I have found use cases when Fuzzy Selector is able to accurately identify and reliably act on specific text elements which for some reason has no attribute with Strict Selector, but are identifiable with Fuzzy Selector.
It’s just a collection of screenshots taken directly from Project Settings. But yes, would be nice to have them centralised in our docs as well, I’ll create a task for this, thank you for the idea
- but you also need resilience in time when facing UI changes. We never aimed at generating a different balance point with Fuzzy vs. Strict: ideally, we want them to be equally balanced; it’s just that for Fuzzy you’ll need to leverage anchors a lot more.
No offense is meant.
we meant it when we said we preffer it this way; let’s just say we’re more efficient