Sneek Peek at MLOps
Hi Today, my article will be not on DevOps but on MLOps, I would like to share some of my experiences on a study I did on MLOps. Firstly we would look into What MLOPs are, then the stakeholders who are associated with this process and the benefits of opting into ML for a company handling Machine Learning driven projects.
There are different types of Ops that we have come across from time to time, starting from DevOps, DevSecOps, DataOps, and MLOps. So as a person who worked as a DevOps previously in a software project-based atmosphere, MLOps is a totally new adventure for me. Machine Learning Operations stand for MLOPs this process is enforced in places where Machine Learning or Data-Driven projects are carried out. When talking at a glance, MLOps evolved from the DevOps process which is associated with the SDLC life cycle.
When comparing DevOps and MLOps, DevOps is a linear process that includes developing the code and doing a release of the project where Continuous Integration and Continuous Delivery (CI/CD) is taken place, in the context of MLOps it has a cyclic process where code, data, and parameters are ingested to create a model and a release is done, in here also Continuous Integration and Continuous Delivery (CI/CD) is taken place but an additional stage is taken place which is the Continuous Training (CT) Process.
MLOps has three phases, which is the Design phase, Model Development phase, and operations phase. The Design phase consists of Bussiness Understanding and Isolating the problem, Requirement Engineering (Data Understanding), carrying out a feasibility study, and Use Case Identification and prioritization. Model Development is the phase where a decided ML model is created and deployed, in this phase data preparation (data cleaning, wrangling labeling), learning algorithms, Model Experimentation, Model Evaluation, Model Building, and Training, Model Serving is worked on. In this phase, the best and the optimized model is selected and extensive training should be done to obtain better results. After the model is developed, the third phase comes in place which is the Operations phase, The deployment is done and manual and repetitive tasks can be automated, Monitoring of the deployed Model and Optimizing. The model should be monitored closely to get the best out of the model.
The stakeholders in this Model Includes,
- Data Scientists
- Data Engineers
- MLOps Engineers / Platform Engineers
When talking about the benefits of the MLOps process, the main benefit is it is a cyclic process, which helps to enhance the model. The model should be monitored closely for skewed data, parameters that are used to create the model and update the model. Manual and repetitive tasks should be automated and these can optimize the productivity of the team and save the time taken to perform a certain task. Infrastructure Management is also a crucial state in operations and with the MLOps so the resource utilization can be optimized which can result in cost reduction.
In the next article, I’m hoping to write more about the benefits of MLOps, its best practices, and some tools which will be able to move forward with the MLOps process.
