This article was co-written by @sean_mealin and @augustazhang.
SAS Visual Data Mining and Machine Learning makes it easy for users to build deep learning models that uncover insights from all types of data, including images, text, audio, and time series data. You may be familiar with the deepLearn action set, which supports training and deploying deep neural networks, recurrent neural networks, and convolutional neural networks in both SAS and open-source languages.
Now, users have access to even more deep learning capabilities, thanks to the new deep learning model zoo (dlModelZoo) action set available with the SAS Viya 2022.09 stable release.
Using the new action set, customers can take advantage of more powerful models as well as use models compiled to TorchScript. The new action set brings the power of PyTorch into the Viya environment, while hiding the messy mechanics of deep learning behind the scenes.
What does the new action set do?
Just like the deepLearn action set, dlModelZoo allows users to both train and score deep learning models. Customers can use models out-of-the-box, or they can use transfer learning to adapt a pretrained model to their own data.
Why should I use the new action set?
The dlModelZoo action set includes all the models you know and love from the deepLearn action set, plus some new models that are even more powerful. New models include YOLO V5, ENet, and more. Classic models such as ResNet, VGG, and UNet are also available.
Unlike other action sets, dlModelZoo uses PyTorch models, allowing third-party models to be imported and used. That means it is possible to build a model in PyTorch, convert it to a TorchScript representation, and import it to Viya for training and scoring.
What’s happening to the deepLearn action set?
The deepLearn action set is still supported. While it will receive critical updates and bug fixes, the newer models will be exclusively available using the dlModelZoo action set. Models built using the dlModelZoo action set cannot be used with other action sets, and vice versa.
What about DLPy?
Currently, DLPy only supports the deepLearn action set, and this support will be maintained for the foreseeable future. But don’t worry! Over the next few months, it will receive updates to work with the dlModelZoo action set, too. That means you’ll be able to access all the new deep learning capabilities from Jupyter notebooks.
Where can I learn more about dlModelZoo?
The dlModelZoo documentation is a good place to learn more and see some examples. Over the next couple of months, additional blog posts and other content will be coming out, which will take deep dives into the new action set. Stay tuned!
