Looking for what’s new at MindsDB? Check out all of the new and improved features added in April.
In the latest MindsDB 1.17.1 version we have:
- Fixed some issues around parsing the input data, solving the cause of previously reported crashes (mainly affecting datasets with dates and/or time series columns)
- Cleaned up our model analysis component. It should now be (literally) 10 times as fast and the new column importance scores that it yields should also be a more accurate reflection of the column’s actual importance.
The latest Lightwood 0.23.0 version introduced the following features:
- Lightwood will now include a so-called “confidence interval” for every numerical prediction it makes. We encourage users to use this instead of the prediction itself. Perfectly predicting a continuous quantity (like a number) is basically impossible, so it makes more sense for us to give an interval for the prediction (e.g. from 0.9 to 1.1, which is equivalent to 1 +/- 0.1).
- Added a fancy recurrent neural network that will handle sequential numerical inputs (aka timeseries.) This should improve accuracy and speed up training time for any time series data.
- Made CPU -> GPU and GPU -> CPU transfer work in some situations where it was not behaving properly before. All models should now be transferable from CPU to GPU and vice versa.
- Improved our automated quality control and made deployment easier by removing various dependencies.
MindsDB Cloud Server
We have added a new feature to the Where to Run option. Now, the Scout users can create a Cloud account and connect to our MindsDB Cloud Server. They can also use the Single Sign On functionality by login to MindsDB Cloud Server through their GitHub accounts.
The Query Dashboard provides powerful analytical insights about the prediction. The improved sections will make each prediction more explainable and will help with decision making.
- Explanation: visually displays the target value in a pie chart alongside the confidence for it. All of the values of the features used for this prediction have also been added.
- Additional information: this section displays which columns would likely yield more accurate predictions, so the users can include them in the query.
- Extra Insights: additional information about the prediction e.g. what could the model predict if some feature values were missing.
MindsDB Covid19 application
We think that non-aggregated granular geodata, which is per case data, should be available for researchers and others wanting to help. One particular approach that we have decided to tackle is using data based on symptoms in locations that are more granular than citywide. This can help determine likely locations where there are COVID-19 infections in people that have not necessarily been tested, but may be asymptomatic and spreading the virus unknowingly.
With this goal in mind, we have developed a simple system where people can report symptoms anonymously and their location down to the neighborhood. The data gathered is anonymous and will be made public to help our communities, health care system, scientists, and innovators make informed decisions. You can contribute to it by starting the census at https://covid.mindsdb.com/
Nelson Mandela once said Education is the most powerful weapon which you can use to change the world. We at MindsDB strongly believe in free education, that’s why we have created MindsDB Education. The first free course that we are providing is Machine Learning Basics Using Mindsdb. This course is an introduction to the basics of Machine Learning, as well as a guide for using Mindsdb to solve problems using Machine Learning. To check the course syllabus or enroll in the course visit education.mindsdb.com.
We are constantly curating new benchmark datasets for evaluating the performance of MindsDB. Check some of the examples at mindsdb-examples. Feel free to contribute with new examples and datasets.