MLOps

a mix of roles, many of them new to the IT landscape

The overarching theme I see is that organizations immediately jump to the mind set of "we need a data scientist." You should look for smart generalists with data versatility, as opposed to specialists in the beginning.

CDO, Data Science, Data Engineer, DevOps, Cloud Engineer

You need people that understand how AI works at a high level (engineers, math or statistics background)

Data science is a team sport - multiple roles are required (data scientists and analysts, data engineers, ML developers, ML architects, DevOps and operations teams)

Data visualization. People that can visualize data in charts and graphs.

And how do you find people with that skill set? CS and applied mathematics double majors are in short supply!

Data science teams don’t want to worry about the underlying infrastructure systems, configuration, and operations – they just want the ability to quickly get access to the tools and data they need to develop their models, continuously iterate, build their data pipelines

Business understanding: how to use AI and what it can do for your business

Organizing across traditional functional boundaries and creating a new team structure

multiple roles. Data Engineers, Data Scientists, ML Architects, Data Operations.

First and foremost, a person who is a great listener and is able to break the problem/request into smaller actionable statements in a simple spoken language

@jameskobielusThis is why HPE is investing in universities to train this new skill http://www.uh.edu/data-science-institute/

@CalvinZito Not the same skill set as an "HPEStorageGuy", but I'm learning.

https://xconomy.com/texas/2018/11/02/hpe-gives-10m...

@CalvinZito you need some "training"

Combination of business, IT, developer, data skills and security members.. Diversity in roles and skills will be key to address biases.

@fvha As a starting point, I grabbed a "Big Data and AI" Great Courses DVD set from my public library. I didn't make it very far!

Then a team that can translate English language requests to more programmable tasks, visualization experts who can help visualize the inputs and outputs to specific tasks, Data Scientists to build models to represent the data and possible outcomes.

@fvha This is the key. Not all Data Scientists can be SMEs is all areas. But a keen observation and effective communication with business users makes them better suited for the job

Right skills may be outside the organisation. There are experts with AI project, use case experience and technical depth that can complement data teams #crowchat @HPE_TechSvcs

ML and Data Science are about extracting knowledge from a given dataset. It all starts with data, you need lots of data and it has to be meaningful data. You can't expect good results if the data isn't any good.

the technology and tools are interesting, but organizing teams and processes for successful business outcomes is key

Do I have permission to use the data?

@JeffGMKE CrowdChat is an open community platform - you can embed and link the entire event or individual pieces

How does one measure or quantify the business value?

Understanding the business outcome you are hoping to achieve. I've seen all too often organizations try to jump right in without a clear cut goal. Is it improving customer experience? Predicting future outcomes? Improving sales? Marketing? Maybe all of the above?

Starts with a team. Having the right team in place. Providing the necessary toolset with minimum friction. Secure access to data

@fvha Not to put you on the spot but tell everyone how long you've been involved in this space?

In a recent whitepaper from Futurum 5 key considerations are listed with recommendations @jameskobielus, #HPEAI, #MLOps https://www.hpe.com/us/en/resources/services/explo...

@MikeLeone_IT And the more granular the goals the better. You want to be set up for success to show you're making incremental progress with established metrics to measure successful progression.

@stu I was talking about "the data" from the data science perspective. One of the considerations is...do I have permission to use the data?

Then a place to store artifacts and reuse them without having to create word documents for tracking

@CalvinZito My background is in Speech Recognition. Collecting speech data is more than a hobby.

@jameskobwielus HPE has the framework to be able to help cusomters define theiur success https://www.hpe.com/us/en/services/consulting/big-...

At a high level there are 3 key considerations: 1. business problem - what are you looking to improve 2. Technology - data, tools, infrastructure 3. people - do I have the right skills and how to train employees to use AI

@fvha and did that for over 16 years.

@jameskobielusLastly taking models to production in a meaningful way.

@fvha So you were doing #MLOps before anyone put a name on it!

@fvha yes, well before it was called MLops

@fvha ... when you are working on stuff and there is no name yet for it you know you are in early

Data scientists often get frustrated because they have to spend a lot of time getting access to environments & data and have to install tools and libraries for each project from scratch. Self-service flexible environments with secure & governed access to enterprise data is key

@csnow_data_uk True that, I am currently in the phase of getting access to data!

@razatweets you could not be more right about training the employees to use AI, or to help them ask the right questions about the huge data they have been collecting so far!

In many ways they extremely adjacent

While DevOps is more about marrying the developers and IT operations folks, ML Ops builds on DevOps and incorporates some additional personas, such as data engineers and data scientists.

MLops is building on top of the DevOPS concepts. However, MLops focusses on the ML lifecycle which is a pipeline that relies on continuous feedback and improvement.

In order to scale models and the implementation of models and to quickly train, infer and re-run models...one has to do this with a DevOps framework

For ML Ops specifically, it's heavily centered on providing intelligence derived from data to a specific business objective, goal, or use case, and enabling developers to consume models and utilize the insights for rapid innovation.

Training a model and reiterating through the cycle until you achieve the acceptance criteria of the model. Devops doesn’t have this requirement or iterative step.

ML Lifecycle is very iterative with a heavy dependency on data. MLOps leverages DevOps processes, however the iterative nature of the ML Lifecycle and the dependency on data should be taken into account. Changes in underlying data will drive retraining decisions

This video talks about operationalizing #MLOps
https://www.youtube.com/watch?v=8p1kP5bg6Hw

Data science and ML are about more than just the code, it’s also about the data and the models - that’s one of the fundamental differences.

A great resource for ML logistics https://mapr.com/ebook/machine-learning-logistics/

The ML lifecycle is a very iterative and continuous process (e.g. retraining the models to ensure accuracy and reduce model drift or model decay) - some of the requirements are quite different than traditional DevOps.

@jameskobielus

Devops concepts such as reproducibility and consistency between stages is very much the same in ML Ops. It just goes beyond deployment into monitoring and retraining and redeployment.

But it is possible to apply CI/CD best practices and bring the speed and agility of DevOps to ML Ops. Last week HPE introduced our new ML Ops software solution focused on addressing these challenges and bringing a DevOps approach to the ML lifecycle: https://bit.ly/2lKOnBO

Here’s a great interview with @NandaVijaydev about applying DevOps to machine learning: https://youtu.be/O98sNtMBaR8

The MLops lifecycle is more iterative than traditional DevOps, especially in the model training loop.

From Big Data to Machine Learning to Deep Learning – The progress of AI http://highfens.com/2019/01/02/big-data-machine-le...