PerceptiLabs is proud to announce the first major Silver release of our visual machine learning (ML) modeling tool, PerceptiLabs 0.11. Not only does this release candidate include a number of significant new features, functionality, and UI improvements, it also offers more stability than our past beta versions. Through these enhancements, PerceptiLabs is becoming the GUI for TensorFlow–you drag and drop the components, and PerceptiLabs generates the visualizations and TensorFlow code for you.
A Visual API for TensorFlow
One of the key aspects of PerceptiLabs is that it builds a TensorFlow model for you behind the scenes. As you drag and drop components in the Modeling Tool, PerceptiLabs automatically creates the underlying TensorFlow code, effectively wrapping that code inside of visual components so that you can easily visualize your model.
To support this functionality and streamline the modeling process, our latest release provides a number of exciting enhancements. Let's take a closer look at what this release has in store.
Improvements to the Modeling Tool
If you've been experimenting with our previous versions, the first major change you'll notice is the layout of the Modeling Tool (formerly referred to as the Workspace editor). Gone are the component icons which have now been replaced with drop-down menus that allow you to easily identify components by name. Once placed into your model, each component is clearly marked with its name and its input and output connectors. And once configured, each component displays an optional live preview directly in the Modeling Tool to show how it has transformed its input.
As you add components to your model and connect them, PerceptiLabs' new Model Autoconfig feature automatically applies "good" hyperparameter settings to help you get to a "good" model faster. In addition, you can now more easily access these settings by simply selecting a component and viewing them in the Settings pane. This eliminates the need to navigate to and from the former Settings popup for each component, and streamlines the general workflow.
We should also mention that a number of new components have been added including Random, GAN, and Object Detection, while some of the Training components have been modified to make it easier to implement different types of models.
PerceptiLabs' Model Hub allows you to import, export, and manage your models. The ModelHub includes a Status column displaying the current training status of your models, and also lists other PerceptiLabs users who are collaborating on that model.1
The code editor, which is now accessed through the Settings pane, has been enhanced with a console window to help with debugging component code.
In addition, there are other new code-related features as well:
- Jupyter Notebook Export: see all of the code for your model via the Notebook view and optionally export your code as a Jupyter Notebook file.
- GitHub Model Export: export your model and data files directly to your GitHub user account from PerceptiLabs to share with others.
On the community side you can now access our new forum through which you can post technical questions, get help, and engage with other community members. Our existing Slack channels will now be used for more general conversations.
Looking Forward to a new way Forward
We're very excited about this release as we feel PerceptiLabs is maturing into a powerful ML modeling tool and GUI for TensorFlow, that can be used by experts or beginners. It truly is a new way to build ML models.
Ready to check it out? Download and run it now:
Finally, we'd like to invite you to watch our live presentation of this latest release. Details TBD, but check back for updates or keep an eye on our social media accounts for more info.
1Cross collaboration is available in the Docker and Enterprise versions of PercepiLabs.