Building the MLOps Environment to Automate the ML Pipeline for a Logistics Company
– The accelerated operation of machine learning models for forecasting Yamato Transport’s workload makes it possible to continuously develop new functions and improve the accuracy of the ML models –
ExaWizards Inc., a company that develops AI-enabled services to solve social issues (Headquarters: Minato-ku, Tokyo; Representative Director & President: Ko Ishiyama; hereafter, “ExaWizards”), announced today that ExaWizards successfully automated the workflow to create machine learning models: Data Collection–Data Preprocessing–Learning–Forecasting (hereafter, “ML pipeline”) by building an MLOps (*1) environment for Yamato Transport Co., Ltd. (Headquarters: Chuo-ku, Tokyo; Representative Director, Executive Officer and President: Yutaka Nagao; hereafter, “Yamato Transport”).
The ML pipeline, with each of its process automated, enabled Yamato Transport to realize the acceleration of its monthly model operation and continuously improve its machine learning models. This is an exceptionally rare case of MLOps employed by a large company, although it has been already introduced to some IT start-up companies, including EC ventures. Based on this achievement, ExaWizards will continue to support businesses, including large companies, in adopting machine learning and operating ML models.
☑︎The background behind the development of the MLOps environment and the effect of the automated ML pipeline
Yamato Transport must create machine learning models every month, which are the bases for the company’s forecasts of the workload of its approx. 6,500 delivery centers for several months ahead in order to realize the effective and optimal allocation of its management resources and fair pricing appropriate for market needs. There are several machine learning models that are operated on a monthly basis and a variety of tasks in the workflow to create such models. Among these tasks are preparing monthly transaction data and master files, rewriting config files, manually executing programs, etc. Yamato Transport has agilely conducted a PoC, which eventually has put a heavy workload on its employees for a long time. Also, the operating schedule, including reporting the forecast results to the Business Department, has been planned in a short-term perspective, which has made it difficult to recreate machine learning models or reanalyze the forecast.
In order to solve the above problems, ExaWizards automated the workflow that had been run manually on a monthly basis by building the MLOps environment (ML pipeline): Data Collection–Data Preprocessing–Learning–Forecasting.
As a result, the automated ML pipeline introduced to Yamato Transport accelerated the monthly operation of machine learning models, and enabled the company to have a more flexible operating schedule, which led to a significant reduction in person-hours for the operation. In addition, the automated program testing in the ML pipeline created the new monthly cycle that starts with effective operation and ends with development. This new cycle not only stabilized the operation of machine learning models but also made it possible to continuously develop new functions and improve the accuracy of machine learning models.
☑︎ Improve the operation of the source code necessary to make the ML pipeline automatically run
Yamato Transport must version control source code files for its machine learning models. However, the company was faced with a problem where no one could easily identify which source code was the latest version since the timing and procedure to register source codes on its version control tools heavily relied on individual skills. Especially, when several development projects were simultaneously ongoing, more time and effort were required for the preparation of the source codes used for the operation of machine learning models in the next month due to the source code control becoming more complicated.
In order to solve this problem, ExaWizards improved the version control method of source codes by referring to Git flow (*2) and standardized the operation method of the version control that had relied on individual skills. Also, the company put in place a development/operational environment in which several vendors can simultaneously use the same source code for development by consolidating its version control tools to GitHub.
As a result of these efforts, the roles of a source code were categorized into “Master (Main),” “Operation in Production Environment (Release),” “Development and Unit Testing (Feature),” and “Integration Testing (Develop),” which brought about innovation to the operation workflow of source codes. The new operation workflow consists of Process 1: Monthly Operation in Production Environment; Process2: Function Development + Testing in Verification Environment; and Process 3: Source Code Update for Next Month’s Operation in Production Environment. The revised workflow stabilized the operation of source codes.
☑︎ Yamato Transport’s Commitment to DX
In “YAMATO NEXT 100” announced in January 2020 as well as “One Yamato 2023,” the company’s medium-term management plan disclosed in January 2021, Yamato Transport clearly stated that it would promote the optimal allocation of its management resources and fair pricing based on the workload forecast through data analysis by switching to data-driven management. The company also specified that it would realize the enhanced productivity of its front-line employees such as delivery drivers by aggressively promoting these initiatives further, aiming to generate a business growth through the increased time and number of contact points with customers, as well as other goals achieved.
ExaWizards supports Yamato Transport in improving the company’s future sales margin per employee through the MLOps.
ExaWizards will continue to promote the implementation of AI in society going forward while supporting businesses in tackling the management challenges essential to being successful in the industry and maximizing customer values through the introduction of machine learning as well as the operation of ML models.
Comment by Mr. Norihiko Nakabayashi, Executive Officer, Yamato Transport Co., Ltd.
The MLOps that was successfully introduced to Yamato Transport enabled us to stabilize the operation of machine learning models while continuously developing ML models and improving its accuracy. We would like to utilize the values of machine learning for our business by combining this MLOps platform with various digital services.
*1 MLOps, short for Machine Learning Operations, refers to the model development–implementation–operation cycle intended to standardize and streamline the continuous operation of machine learning models in the customer’s operating environment.
*2 Git flow is a branch (source code control unit)-enabled development method offered by Git and a control method of source codes developed by several individuals.
[About ExaWizards Inc.] https://exawizards.com/
With the mission of “Solving social issues through Artificial Intelligence for future generations,” we are developing and commercializing AI products in various fields, such as nursing care, medical care, HR, robotics, finance, and cameras, in order to solve industry and society-wide issues identified from individual company issues, while working on solving issues in each department and company-wide use of AI. Our members include AI engineers, software and hardware engineers, strategy consultants, UI/UX designers, domain experts in nursing care and other fields, researchers, policy experts, and other cross-disciplinary personnel. In Japan’s super-aging society, we are developing our business with a thorough understanding of the needs and issues in each field.
<Inquiries about this release>
Public Relations Department, ExaWizards Inc.