What’s New: MindsDB December 2020

2020 has been a year of a lot of new features and integrations in MindsDB thanks to our community and their feedback. Here are new product updates in December to close up 2020.

Improvements

  • Added Automatic Mixed Precision for NnMixer.
  • A lot of improvements to the MindsDB installers that should make the whole installation process easier.
  • Datetime encoder improvements to determine monthly number of days given the year and month.
  • Improvements to Numeric decoding to avoid overflow issues.
  • New Time series transformer encoder.
  • Improve GPU usage availability.
  • New accuracy metrics for numerical predictions (r2 score and confidence accuracy).
  • Added SQLite as a datasource.
  • Improved runtime performance in GPUs with Tensor Cores.
  • Improvements to Quick learn feature when used with time series data.
  • New option to disable column_importance feature.
  • Added support for connection string when MongoDB data source is used.
  • Refactor and Improvements to costly operations around datasources metadata.
  • New username/password authentication for MongoDB API.

Bug Fixes

  • Fixed issue when stop_traininx_in_x_seconds parameter is used.
  • Fixed ValueError when target variable is of type Date.
  • Fixed distilBERT out of range bug.
  • Fixed issue when models are copied to another location.
  • Fix for time series issue when nan values exist in the data.
  • Fixed CSV parsing issues when columns contain quotations marks.
  • IndexError is fixed when querying time series data.
  • Fixed issue for /delete predictor endpoint.
  • Fix for creating Mongo data source through HTTP.