Datalore: A new web app for machine learning visualizations
The team from JetBrains has just released Datalore, a cloud-based web app for building machine learning models and creating rich visualizations in Python.
Last week, JetBrains announced the beta release for their newest creation, Datalore, a web application intended to simplify building machine learning models, create rich visualizations, and help developers with data analysis.
Working with complex data is easier than ever thanks to Datalore’s smart coding assistance, incremental computations, and built-in tools. Machine learning is all about Python and Datalore is no exception.
Datalore offers a number of tools and measures to make machine learning and data as “enjoyable and productive as possible”. Their easy to use code editor has smart code completion, inspections, quick-fixes, and easy navigation. It checks your work and speeds up the code-writing process with these automated processes.
Basically, Datalore suggests context-aware possible actions called intentions. These intentions are based on what you’ve just written, giving you relevant options for what your code should do. Then, once you’ve clicked on a particular intention, Datalore generates new code for dataset uploads, train/test splits, graph designs, and more.
This new tool also makes it simpler to see how your machine learning models change with multiple edits. Thanks to its incremental computations, you can see what happens to your predictions if you change a few variables or parameters to the models. The output on the right side of the screen always reflects your latest results, making it convenient to check your results in real time.
Datalore comes along with all the basic Python machine learning tools, including numpy, pandas, and sklearn built-in libraries. However, it also comes with two advanced built-in libraries just for Datalore. Based on the R implementation of ggplot, it comes with its own version: datalore.plot. There’s also a library that adds interactive maps to your analysis with datalore.geo_maps.
Developers can choose how many instances they need, based on the computational needs of their models. Right now, instances run from medium (4GB ram) to extra-large (61 GB ram), but JetBrains offers options if you really need to increase you available computational power.
Additionally, Datalore comes with a few basic built-in datasets and a file manager. There’s no chance of losing any data due to automatic saves. The file history ensures that you never really lose all your work, either.
You can also work remotely in real-time with access to the workbook and code-editor. Collaboration between teammates is encouraged; they can add code and write comments as well.
If you’re interested in trying out Datalore, you’ll need to sign up for a JetBrains account. However, they do have student and startup discounts, as well as an open source project license. Datalore is still under development, so all feedback is welcome via the Datalore forum. You can also check them out on Twitter @DataloreJB.©