Brainspace

Brainspace 6.4 introduced a range of new and updated features. This document provides details on each of those features to help understand the differences and requirements to use the features.

New visualization to help evaluate how complete the classifiers training is.

The following features are available on the Documents Scored Across Training Rounds tab

Callout 1: Score Band buttons when clicked take the current round score band to the Analytics view for the user to take further action. The bands are set to 5 default ranges that are currently fixed which are 0.8-1, 0.6-0.8, 0.4-0.6, 0.2-0.4 and 0-0.2. If you would like to specify the exact range of records you wish to analyze you should use the Predictive Rank graph Trim feature.

Callout 2: Download Score Distributions Report button downloads a report to your local downloads folder. This report can be viewed and edited for distribution to stakeholders that do not have access to Brainspace.

Callout 3: Round and Score dropdowns allow any of the rounds or score bands Documents In Score Range, Total Coded and Positive Coded counts to be displayed.

Feature availability

The round scores visuals will be the new default visual when in a CMML Classifier. The graph will automatically be created for every training round.

For classifiers created prior to 6.4, the historical round information will be created when the next training round is closed.

Credit Card Entity detection has been improved to be more accurate and reduce false positives by detecting specific credit card types.

These improvements also allow for the searching of specific credit card types as entities. For example, if you wish to find all VISA numbers only, this is now possible.

A new search field <brs_has_entity_type> is now available to search for any specific entity type including the newly introduced credit card types:

In addition, a new field <brs_has_entity> has also been added to return any document that includes (or not) any entity.

Feature availability

This feature is only available for NEW datasets due to a schema update.

The implementation of Emoji support provides:

Searching for Emojis:

Using the OS provided tools:

Other methods:

A new search filed <brs_has_emoji> has also been added.

Feature availability

Changes have been implemented to core supervised learning. The main benefits include round to round stability, improved effectiveness and improved portable model interpretability.

The main area of change that will be observed is within portable models.

The feature weighting changes from a scale of +100 to -100 to a scale of +10 to -10 with a single decimal place.

Less extreme weighting will also be observed.

What should I do if I have existing models that I have crafted with +100 to -100 weights?

When using models with integers higher than 10, every integer higher will be treated as a value of 10 when the model is used with a new classifier.

As such, moving the decimal place 1 point before using the portable model is recommended. For example:

Feature availability

When the next training round is closed, the system will switch from the old method to the new method automatically.

Document scores, ordering and any Control Set statistics will typically change.