What’s New: MindsDB November 2020

As we are approaching the end of 2020 our team is busy with finishing the latest integrations of AI Tables in new Database systems that will be announced soon. Apart from that we still have managed to release new features and bug fixes.

We improved MindsDB installation by providing new easy-to-use installers for Linux and macOS and updated our Docker image that runs MindsDB server and Scout. The latest MindsDB 2.13.8 also contains:

Improvements

  • Improvements to the DateTime encoder to determine the monthly number of days given the year and month.
  • Added Snowflake integration through our graphical user interface(Scout).
  • Improved generalization in neural network mixer.
  • Increased code test coverage and major improvements around tests, testing datasets and user-behavior tests.
  • Added validation to not allow training models with a dataset that contains less than 10 rows.
  • Improvements to fix the confidence interval out of range bug.
  • Major changes around Datasources functionality.
  • Redesign the confusion matrix for model quality in Scout.
  • New functionalities for datasource analysis feature to speed up and improve data analysis.
  • Major improvements around AI Tables configuration and added backward compatibility with older configuration versions.
  • Async download of Scout files (that should significant reduce time to open HTTP interface)
  • New design of the Query view dashboard in Scout.
  • Improved Advanced mode in Scout to allow removing multiple rows for model training.
  • Added Onboarding tour in Scout that would make the whole navigation easier for new users.
  • New Data dashboard in Scout that simplifies the data upload and connection to database steps.
  • Redesign of the Explainability section.
  • Added caching to the datasource_analyze to improve loading time.

Bug Fixes

  • Fix naming schemas in data sources to avoid collisions with data column names.
  • Fixed issue with our dependencies (tokenizers) on macOS.
  • Fix for the negative value prediction in positive numerical tasks.
  • Fixed the issue with stop_training_in_x_seconds parameter.
  • Fixed word distribution in data analysis.
  • Fix for multiple target prediction. Now, users can train the model to predict more than one target variable.
  • Fixed ‘import mindsdb’ in Jupyter notebook.
  • Fix for ignoring columns functionality. Now, users can specify columns to ignore when training the model through SQL.
  • Fixed invalid date status when training model from Scout.
  • Fixed issue with uploading data that contain Null values through SQL.
  • Fixed storage_dir issue when a relative path is added to the configuration.