Developers and data scientists often work together in teams, but our objectives and approaches to data and ML couldn’t be more different: from how we reason about data, why we collect it, and the tools we use, to where AI actually fits in our workflow.
In this Ask the Expert event, you will be able to chat with Zan and ask your questions about AI and ML development, DataRobot APIs, etc. On this interesting and important topic, Zan is available to answer your questions.
|Zan Markan is a Developer Advocate at DataRobot. He has spent over a decade working in technology, in a variety of roles and enterprises—as a developer, manager, and in developer relations.
He is passionate about educating developers and enabling them to be successful with whichever technology they are using.
This Ask the Expert event is now closed.
Thank you Zan for being a terrific event host!
Let us know your feedback on this event, suggestions for future events, and look for our next Ask the Expert event coming soon.
Hi @zmarkan , thanks for being around and hope you are taking good care of yourself during the covid-19 challenging times.
My question to you is, what was the most advanced API functionality (ie. required most amount of coding) that you executed with the DataRobot Python/R API?
Hi @a_shoop, and thank you for your question.
As I don't have extensive Python / R experience, I went straight off piste and used the REST APIs directly with Node.JS.
I've used the APIs to make predictions in real-time and monitor my deployed ML model. Even without the use of dedicated SDKs neither required more than a few lines of code and API calls.
I plan to share one example project in the next few days
This question came through before we opened it up - from Mladen:
From what I've seen DataRobot can definitely be used (and is being used) to solve pretty much any ML use-case out there. From finance to manufacturing, healthcare, retail, and everything in between. I found a pretty good overview here: https://www.datarobot.com/use-cases/
Would DataRobot be a good solution for more general ML, or would it perform well with practically every problem ML can solve?
Hi @nico1e. And thank you for your question.
It's really great that you asked about a dedicated developer section in the community, as we're just preparing to launch it.
Our plan is to create the best resource fo developers wishing to learn ML and AI skills and technologies, and adopt them into their toolsets, as well as to give the developer community the chance to influence our developer experience going forward.
We're still a few days away from launching it, but we've prepared a short questionnaire where you can register your interest, tell us what you would like to see, and we can notify you when it's available.
You can find the form at the following link.
Hope this helps!
Thank you Zan this is wonderful!
A follow-up question, What is the hardest or most challenging thing when working with ML and AI as a developer?
Hi @ErnieZ, thank you for your question.
The most challenging thing when working with ML and AI as a developer to me is the step of gathering and preparing data for training - what will later produce our ML model.
A training dataset requires a column designated as the target - that is the value you want to predict, and is not the data that you will have at the time of making predictions.
Therefore, a training dataset might require you to combine multiple sources of data so it contains this target as well.
Other than that, it used to be that training the model itself was also a quite a time consuming process, but I find it pretty straightforward with DataRobot nowadays.
Hope this answers your question!
To my knowledge most things in the DataRobot UI are supported with APIs and libraries and vice versa, but there might be a few minor discrepancies with features not available in the APIs and libraries yet.
We don’t have a list to share, but if you have a concrete example in mind then our support team will be able to answer you this.
We are working on a set of demos and example projects. I'll be publishing the first one very soon - in the next few days.
I will update this response when I will have posted it and let you know!
I've had some exposure to TensorFlow and PyTorch through research for different ml projects and discussions with my colleagues. How does DataRobot differ from them?
Both TensorFlow and PyTorch are individual open source libraries and frameworks for machine learning (deep learning to be precise).
DataRobot's Automated Machine Learning product will use multiple libraries and frameworks in addition to TensorFlow to train models, in order to suggest you which one works best on your dataset, and your use-case.
I checked out that link and see a lot of other good info in your wiki too. thanks for clarifying and also sending along that link. I'll be busy reading it all for awhile !
One of the most common applications is what we call a Model Factory.
The idea here is that you build your own automation layer on-top of DataRobot.
For example, you might need to build several hundred propensity models for different products to use inside a large scale product recommendation or lifetime customer value application. You can use the DataRobot APIs to automate the building of these large number of projects and models, and then automatically deploy and integrate them.
Does that make sense?
DataRobot's MLOps automatically monitors models when you deploy them into production.
One of the ways models are being monitored is for data drift - it's checking for whether the data you are using to make predictions is starting to deviate from the data that you used to train the model. Models with a high measure of data drift are considered less reliable.
As a developer there are many ways and reasons why you could consider using machine learning technologies in your apps.
If you're working on a team with data scientists, then you can integrate the models that they build into your applications, or build visualizations to monitor the models you have deployed in production.
If you're not working alongside data scientists you can also create models yourself, with the data you are already likely collecting - for logging or analytics for example.
And if you're looking for use-cases themselves, I would recommend you check out the answer I gave a few days ago to another community member about some interesting use-cases: https://community.datarobot.com/t5/ai-ml-general-discussions/ask-the-expert-ai-and-ml-for-developers...
Hope this helps!