Our team has worked on a lot of new features in the MindsDB Stack. Take a look below to find all of the new and cool features available in the past several releases from July.
One of the great new features that we have in MindsDB 2.2.0 is the MySQL integration. Now, you can train and query machine learning models straight from the MySQL database. Stay tuned about new tutorials for it. The additional new features are:
- New columns added to SQL queries. Now MindsDB can return the maximum and minimum target value and also an explanation in each query.
- Improvements to the configuration, which means users can dynamically edit the configuration sent to mindsdb server.
- Fix issues around authentication in mindsdb server.
- Major refactoring around unit and integration tests.
- Bug fixes for timestamp transformation, SSL certificates, SQL queries with special characters, no UTC timestamp, null inserts, and major improvement around ClickHouse integration.
We did a lot of new improvements in the 0.32.0 version of Lightwood:
- Lightwood has new text encoders that shall improve the speed when encoding text.
- To ensure better quality we have new tests added for recurrent neural network encoders.
- Fix issue with Multihot encoder to handle None values.
- Added new Multicategorical encoder and support for multicategorical output.
- Improvements and code refactoring around the datasources.
- Improved how Lightwood encodes images.
- Added Multi-label prediction which means Lightwood can now predict a list of categories.
- Installation fixes for MacOsx and fix for non-ascii alphanumeric strings tokens.
The following are all of the new cool features around mindsdb_native 2.1.2 version:
- Improvements in the way that MindsDB calculates regression scores.
- New logic that will improve the data cleanup and eliminate the redundant data from training and testing datasets.
- Improve memory consumption of MindsDB when large datasets are used with the new features in DataTransfomer.
- MindsDB now has two new datasources for connecting to the Microsoft SQL Server and MongoDB.
- New interface for predict/learn functionality. Some of the arguments that are deprecated in the new interface are:
- Removed when parameter. Now when_data can be a dict, more consistent with learn having a single from_data argument.
- Renamed unstable_parameters_dict to advanced_args.
- Improvements to the text data analysis to distinguish between short and rich text.
- New Functional interface that supports utility functions around the models as analyze_dataset, load model, delete model, download model, export storage.
- Minor bug fixes around null values parameters, datasource parameters, confusion matrix and outliers in the data analyzer.
In the MindsDB Scout we are working on a new cool feature that will allow the Scout users to integrate Scout with Databases. That means with few clicks you can train and query machine learning models from the databases. This feature is still in beta and it will be released in the next few weeks. Other than that we did new improvements as:
- Visual improvements of the connection terminal to distinguish between errors and information messages.
- New text stats visualizations where users can visualize the occurrences of words in the dataset.
- New heat map that displays the confusion matrix
- Improvements in the datasource quality and added outliers in the histogram charts.
- Design improvements of the notifications, icons, charts and fonts.
- Added percentages to the data quality visualizations for variables occurrences.
- New calendar components for time-series features.
- Fix a bug related to the remote login options
- Fix issue with timestamp when querying models
To get additional information about MindsDB check out our documentation or enroll in the Machine Learning Basics using MindsDB free online course.
We are happy to have you contribute to MindsDB. You can join our community or check out the MindsDB’s repositories on GitHub, and if you have any questions feel free to reach out to us.
We want to say special thanks to people who already contributed to MindsDB in July: