Daylight | On-Site v.2.2 | 2020-02-29

On-site packaging and management

New Features

Science Updates

Enhancements

Fixes


Learn More

On-site menu changes

With this release, we made several changes to the on-site menu to match what you see in the new Settings page: 

  • Renamed “account” to “workspace” in all instances. This reflects our system-wide Settings update that restructured anything that was formerly an account into a workspace. 

  • Updated the Change User Access section to match the roles you can assign in the UI.

Azure VM now available 

All Luminoso products are now accessible through an Azure Virtual Machine. If you are an existing customer who is interested in using this option, contact your Luminoso Customer Success Manager for details. 

New Settings page and redesigned account management

In 2019, Luminoso redesigned account handling in Daylight and combined the former User and Account Settings pages into a unified Settings page. These changes: 

  • Clarify roles and permissions

  • Make account management more intuitive at every level

  • Introduce better password rules

  • Manage API tokens more securely

This update replaces "accounts" with the clearer designation of “workspaces,” and introduces "organizations." With this infrastructure change, you can set up separate workspaces with different permission levels so that users only see as much data as they need. For instance, one user in an organization may need to create projects in one workspace, but only has to view data in another. 

With this release, you no longer view or manage short-lived API tokens in the Daylight application, since this type of token is only for active browser sessions. Now, you can create as many long-lived API tokens for authenticating integrations as you need and tokens are only visible immediately after their creation.

Read more in our article on the new roles and actions in Daylight and a guide to the Settings page.

Delete multiple saved concepts

Starting with this release, you can delete multiple saved concepts with one action. Now, access the new Delete saved concepts button on the What concepts have I saved? card in the Highlights feature. 

  1. In Highlights, navigate to the What concepts have I saved? card. 

  2. Click Delete saved concepts. A Delete saved concepts window opens. 

  3. Check the box next to each concept to select it, or use the box at the top to select or deselect every concept.

  4. Click Delete concepts. A confirmation message appears. 

  5. Click Confirm delete. The concepts you selected are deleted. 

  6. Click X at the top right corner or anywhere outside of the Delete saved concepts window to close it. 

QuickLearn 2.0 

With this release, Luminoso introduces QuickLearn 2.0 to all Daylight users. This major improvement to Luminoso’s science expands word embeddings from 150 to 300 dimensions. Now, your projects display major improvements in conceptual matching and handling of possible social biases. 

All new Daylight projects are now built using QuickLearn 2.0. For existing projects, add new documents to take advantage of QuickLearn 2.0 science or rebuild the project through the API. Daylight on-site projects do not update to QuickLearn automatically.

As part of QuickLearn 2.0, you’ll see: 

  • Updated conceptual matches — As a result of strengthened conceptual matching, you will see more intuitive and a higher number of conceptual matches for concepts in Daylight. 

  • Improved language support — QuickLearn 2.0 expands the starting vocabulary across the 15 languages Luminoso supports. This change improves accuracy in especially non-English languages and in our classification offering, Compass.

  • More precise conceptual match scores — QuickLearn 2.0 contains a larger semantic space, so concepts may be farther apart, resulting in lower concept association scores. This change is especially visible in the Volume feature. 

  • Updated look to the Galaxy — The Galaxy feature is a 3D representation of Daylight’s semantic space. Since the space is now larger and concepts are more related, the Galaxy now appears more round. 

  • Improved score drivers — Changes to conceptual matches affect the rank of suggested drivers and a concept’s average score calculations. Overall, analysis will be more accurate.

  • Reduced bias — Enhanced understanding significantly decreases the risk of bias in natural language understanding (NLU) results, especially for categories protected from discrimination. Read more about decreasing bias in AI in this conversation with Luminoso’s Chief Science Officer, Robyn Speer


Read more about QuickLearn 2.0 in our Guide to QuickLearn 2.0

Enhanced handling of emoji variations

With this release, we introduce updates to our natural language processing that help Daylight handle emoji variations better.

For example, when you search for the (“person shrugging emoji”), you now find documents that contain any variation of the person shrugging emoji, like (“man shrugging: medium-dark skin tone”).  The Volume, Drivers and Galaxy visualizations now provide the total volume of any emoji that displays a person shrugging, but still preserve the details of the specific emoji variations that are used at the document excerpt level. 

Flag emoji that are composed of multiple characters are now preserved as a single concept. For example, the two separate concepts of "flag piece: U" and "flag piece: S" will now be preserved as a single concept

Improved handling of collocations

In this release, we improve Daylight’s handling of collocations, or groups of words. When Luminoso processes your dataset, it creates collocations out of groups of words that are significant when used together, such as “dining room.” Previously, Luminoso would not collocate a group of words with a negative term in the middle. Now, a phrase like “no issues were resolved,” in which the negation begins the collocation, is conceptually identical to “the issue was not resolved,” which contains the negation in the middle of the collocation. 

This change affects all 15 languages that Luminoso supports. You’ll see this change when you build a new project, update or rebuild an old project, or do not open a project for six months (which triggers  an automatic rebuild). To see this change in Compass projects, you must rebuild each classifier.

This change will affect each project differently. If your dataset contains a high number of negative phrases, you may see a change in your top 500 concepts or a slight difference in conceptual match scores. 

API: Science version information included in a project record’s last_build_info

Now, when you call a project record, a science version identifier is included. This string displays the version of Luminoso science at the last time the project was built. 

You can compare this number against the latest available science version that displays in the status endpoint. If your project is out-of-date, you can rebuild it using the build endpoint. For more information, read our API Documentation

API Reminder: Change to the default of legacy_term_format parameter coming in May 2020 on-site package

When we released Daylight on-site version 2.1. we added the optional parameter, legacy_term_format, to the Vectorize text API endpoint. At the time of this release, we continue to accept “true” or “false” in this parameter.

If you need time to update code on your end, specify “true” to this parameter and continue to receive the array format for terms and fragments. To update your code to accept the new format, specify “false” and receive modern object-style terms and fragments. If this change does not affect you, you do not need to specify anything for this parameter

Beginning with the Daylight on-site v2.3 release on May 30, 2020 the default is “false,” and we consider the parameter deprecated.

In May 2021, the parameter is completely removed, and modern object-style terms and fragments are the only available output.

For more information, refer to Luminoso’s API documentation on vectorizing text.

API Reminder: Deprecation of exact_only parameter coming in February 2021 on-site package

When we released Daylight on-site v2.1, we added the optional parameter match_type to the Get documents API endpoint. If you don’t specify match_type, the default response is both exact and conceptual matches.

You can specify if you’re looking for only exact or conceptual matches. 

We have deprecated the prior optional exact_only parameter in the API endpoint for Get documents, and it will be removed in December 2020. If you formerly supplied “true” for exact_only, the equivalent is “exact” for match_type. We recommend updating your scripts accordingly. 

For more information, refer to Luminoso’s API documentation on get documents.

Highlights: View saved concepts button opens Volume feature

Now, when you click the View saved concepts button on the What concepts have I saved? card in the Highlights feature, it opens the Volume feature. Use Volume to view, sort, and manage a comprehensive list of your saved concepts. 

Settings: New usage meter for uploads

With this release, we add a new usage meter to the upload page. This meter provides greater visibility into your organization’s data usage and how your uploads will affect your overall use.

Now, when you upload new documents in Daylight on the Create a project page or the Upload data page for additions to existing projects, you see a usage meter that reflects the total usage for all workspaces in your organization. 

In the meter, dark blue represents already uploaded documents, light blue represents the documents you are uploading now, and white represents your unused documents. If your upload would exceed your document limit, the meter displays a dotted line that indicates the limit. 

Concept details pane: Moved match type and sentiment icons

With this release, we moved the icons for exact and conceptual matches and for positive and negative sentiment directly to the left of the document preview. This change displays more matching document previews for a selected concept. It also makes document previews easier to scan by type. 

View this change in the Matching documents section of the Concept details pane in any analysis feature. 

Concept details pane: Edit saved concepts

With this release, you can edit your saved concepts in all analysis features. Now, the process for editing concepts in the Galaxy feature is the same as in Volume, Sentiment, and Drivers.

Select a saved concept and click the edit icon in the Concept details pane to access the Edit saved concept flyout. In this flyout, you can:

  • Edit a concept’s name and color

  • Create and edit compound concepts

To take action in the Edit saved concept flyout: 

  • Click the X or anywhere outside the flyout to exit without saving changes. 

  • Pick an option from the Color menu to change a concept’s color. 

  • Type into the text field and click Save to edit a concept’s name. 

  • Search a concept’s name in the Add concepts field and click Add to add concepts to a new or existing compound concept. The concept appears below the Add concepts text box.

  • Click the X next to a concept’s name in the editor to remove it from a compound concept. The concept disappears.

To delete a saved concept, click the delete icon next to the saved concept’s name.

The Galaxy does not display concepts with all zero vectors

Because the Galaxy represents associations between concepts, it no longer displays concepts that do not relate to other concepts. It is unusual for a concept to have no associations, and this scenario could exclusively happen if a concept only appeared by itself in every document. It could also happen if the concept was part of a collocation unique to its dataset and appeared alone in every document. This also means that the concept did not exist in the background space. 

Now, the Galaxy does not display these zero vector concepts. Concepts with zero vectors continue to display in the Volume, Sentiment and Drivers visualizations, if appropriate.

Drivers no longer crashes if no drivers are suggested

Before, if you selected a score field for a dataset where most documents didn't have that score value, Daylight couldn’t suggest any drivers. The feature would crash and return a generic error message. 

Now, if you open the Drivers feature and choose a score field that only applies to a few documents, Daylight displays a more specific error message that says "Sorry, there are no drivers returned for your current selections. Please adjust your selections.” 

The rest of the UI remains accessible, so you can easily choose another score field.