Welcome to the MindsDB’s summary of all of the new features and bug fixes we did in August. The new releases around MindsDB’s stack are ready for download. To get more info read below!
The new feature that we are really happy about is the integration with PostgreSQL. Training Machine Learning Models straight from the PostgreSQL database can be done starting from the 2.3.0 version of MindsDB. Apart from the new database integration, there are a lot of new features related to the integrations with ClickHouse, MariaDB and MySQL databases. The short summary of those are:
- New /config/all_integrations endpoint that lists all integrations and the data associated with them.
- New integrations health check endpoint /integrations/<name>/check (where <name> is the integration's id) that allows users to check if MindsDB can connect to the database.
- New database integration datasource that can be added through /datasources/<name> endpoint.
- New parameters added to the /config as database_name and date_last_updated.
- Fixed issue with provided passwords in config as integers.
- Improvements in creating datastore directory, so when an error happens the datastore will not be created.
- Fixed the issue with the wrong storage_path. Now, the provided --config storage_path shall be used as mindsdb_native storage.
- Fixed the permissions error in Windows.
- Improvements in the error handling around routes and added more detailed information as a response.
The latest version of MindsDB Native 2.5.0 introduces new features as:
- Improvements to user-provided null values as column values.
- New Histogram logic that would simplify and improve the word computation.
- New column importance formula that will resolve the negative values in column importance.
- New data_split_indexes argument to Predictor.learn().
- Added guess_probability to stats. This new key contains a probability of randomly guessing the category.
- Improvements to model training that can now fail and the appropriate error will be returned on an attempt to predict. Also, the training can fail if the accuracy is worse than random.
- Fixed issue with array columns that were encoded as time series.
- New stats for tags analysis.
- Small optimizations features added to the time series encoder.
- Optimized and simplified the "self-aware" network.
- Cleaner training loop logic.
- General performance improvements with some datasets that we have had bad performance before.
MindsDB JS SDK
The latest available JS SDK on npm is the 0.9.3 version that contains bug fixes around create_datasource and check_database endpoints.
Starting from this month Lightwood, MindsDB and MindsDB native are using Github Actions to automate our build, test, deploy workflows.
Download the latest Scout version and check out some of the newly available features/bug fixes:
- Improvements to the bar chart visualizations of the outliers in the data analysis.
- Fixed issues with data analysis on the datasets that contain null values.
- Fixed a bug with the histogram in the (occurrences of a variable in the dataset) that doesn’t render all data.
- Visual improvements to the predictor results and alert messages of the Scout forms.
- Fixed error with sample_margin_of_error. Now, this flag can be used in advanced parameters.
- Fixed issues with timeouts when analyzing the uploaded dataset.
- Minor bug fixes around the responsiveness of the messages around the Dataset section.
There is a new feature that we are working on in the past few months in the MindsDB Scout which is currently in Beta version, where you can connect directly to the databases from the Scout. That means you can create a datasource, train a model and query it from the Scout. To check that join our beta testers or reach out to us through the community forum.
We are keeping the last part of our product updates, to say Big Thanks to our great community that is making the MindsDB products better by trying and testing the latest features and also reporting bugs and suggestions to them: