Ludwig (program for training of AI models)

2019: Emergence in a public access

In February, 2019 the Uber company released the tool open source of Ludwig which is intended for training and a training of the systems of artificial intelligence. At the same time for use of technology it is not necessary to write a program code.

According to the VentureBeat edition with reference to the statement of Uber, Ludwig became result of a two-year company performance over optimization of development of AI systems for practical projects. Service of the order of the taxi uses this tool for such tasks as extraction given from the scanned images of car driver licenses, determination of destination points interesting to clients during their correspondence with drivers and also forecasting of delivery dates of food.

Uber was released by the open tool training AI systems without writing any code

Ludwig is unique thanks to the capability to facilitate understanding of deep learning for nonspecialists and to accelerate cycles of iterations as for experienced developers of models of machine learning, and for researchers, says in the blog of Uber. — Through Ludwig experts and researchers can simplify process of creation of prototypes and data processing that it was possible to focus on development of architecture of deep learning, but not on preparation of Big Data.

Ludwig provides a set of command prompt utilities for training, testing of models and  obtaining forecasts and also the Python API interface allowing to train or load AI model and to use it for obtaining forecasts on the basis of new data sets.

Uber added the standard visualization tools helping users to understand performance of their models of deep learning and to compare their possibilities of forecasting to Ludwig.

For  the distributed training of Ludwig it is capable to use Uber Horovod framework supporting parallel operation of many video cards and  computers. The solution is constructed based on TensorFlow framework.^[1] 

Notes