A constant flow of enhancements to the Alchemite™ machine learning (ML) software ensures that key workflows are always getting that little bit easier and faster. Two recent examples are the extension of functionality that enables users to add domain knowledge when training ML models and a new copy-paste feature when updating data. You can see these and other features in the demo video below.
In the video, Daniel Yanes from the Intellegens Science Team, runs through a typical workflow for improving the predictive accuracy of a model. Alchemite™ is often used in this manner – first, learning from whatever data you have available to build a model and then supporting you to figure out what experiments to do next in order to both move towards your research targets and improve your ML model quality. Higher quality models improve your understanding through more accurare insight into relationships within the data and enable you to identify formulations that hit your targets, faster.
Here’s a little more information on the two new features demonstrated in the video:
- Categorical values in calculated columns. When training an ML model, performance can be improved if you give Alchemite more information about the inputs to the system and their relationships. One way to encode this domain knowledge is to tell Alchemite to automatically compute a new column of input data (a so-called ‘calculated column’) using other inputs. The latest enhancement enables this feature to use categorical values – such as names or labels. In the demonstration, it is used to tell the model what types of cation it might consider to be large or small.
- Cut-and-paste in data editing. Now you can paste not just single values but whole ranges of cells between applications like Excel and the data tables in Alchemite™, greatly speeding up the manipulation or addition of data.
More flexibility when adding domain knowledge to your modelling and slicker data entry – these are the sorts of small improvements we like to see at Intellegens!
For more information on this and other ML workflows, contact us.