What’s New: MindsDB January 2021

Here’s What’s new to MindsDB in the first month of 2021.

Improvements

  • Improvements to logs handler to avoid unuseful warnings messages.
  • Added support for returning class probabilities in all mixers.
  • Improved autodetection of GPU.
  • New LightGBM mixer that speeds up training and reduces memory usage.
  • Improved quick_learn and quick_predict functionality.
  • We’ve reverted back to reporting a within-bounds accuracy for numerical tasks. This metric indicates the proportion of true values that fall inside the bounds of our model’s prediction range.
  • New option to filter outliers in the target column when training a predictor.
  • Improved model importing functionality.
  • Added new integration tests for most of the database integrations.
  • Improved the design of query modal in MindsDB Studio.
  • Visual improvements on data preview dashboard in MindsDB Studio.

Bug Fixes

  • Fixed KeyError bug for categorical time series targets.
  • Fixed bug when historical_columns feature is used with time series data.
  • Minor bug fixes in NnMixer and SkLearnMixer.
  • Fixed issues with MSSQL integration which is now in a stable version.
  • Fixed issues when the dataset contains NaN values.
  • Fixed data analysis issues when a model is trained with ignored_columns.
  • Fix the wrong default directory for mindsdb configuration.
  • Fix MindsDB Studio check for connection functionality.
  • Fixed the issue for selecting the GPU/CPU training option in MindsDB Studio.
  • New predictors that are in the training phase will be visible in the MindsDB Studio predictors list.