In most cases, humans still have to teach machines how to interpret data. With the Datalore web application, the developers at JetBrains want to make this step at least a little easier. The application has been available in open beta since February and is intended to make creating calculations simple and collaborative. Among other things, it relies on intelligent suggestions and incremental calculation.
Datalore can be used to analyze and visualize data. With this tool, developer JetBrain aims to expand its own product range in the field of machine learning and deliberately relies on Python. According to JetBrain, it is pursuing a clear vision with these products, namely: “to make development as enjoyable and productive as possible for everyone”. The Czech developers want to achieve this with a number of features that make the processing of data in Datalore particularly easy. In addition to an intelligent code editor, tools for incremental calculation and machine learning are also on board.
Code editor with clairvoyance
One feature that the programmers are particularly proud of is the code editor. This not only makes coding easier by making predictions, but also creates the predictions depending on the context. This means that the editor’s suggestions are based in real time on what was last written as code. From this, the editor generates records for uploads, graph design and other functions.
Calculations are not a house of cards
Calculating machine learning models sometimes requires a lot of fine-tuning. Datalore aims to address this fact by having the program follow dependencies between different calculations. This means that when the data of a calculation is changed, Datalore tries to minimize the recalculations in dependent calculations. In this way, changes should become visible more quickly and the output should always show the most current version of a visualization.
Off to the library
For the data analysis in Python basics like NumPy, Pandas and Sklearn must not be missing, for this Datalore offers two own libraries. One is datalore.plot, which is based on ggplot, and the other is datalore.geo_maps. With the latter interactive map material is to be implemented in analyses. The maps are based on OpenMaps. In addition, built-in datasets such as Iris, MNIST or Titanic are included in Datalore, which can be supplemented with own .csv files using a file manager. Changes in the respective datasets are saved automatically, and thanks to a version control system, changes should be able to be undone without any problems.
Interactive with each other and the data center
Since Datalore is a pure web application, workbooks and code editors can be used simultaneously by several team members. In addition, exchange is possible via a commenting system. Both these and all other functions of the application run on computational resources that depend on the amount of data used in each case. At the bottom of the browser window, the resource usage can be continuously read, both the CPU load and free memory are displayed.
For larger applications such as deep learning algorithms, JetBrains also offers larger computing power packages.
