A workflow defines how your model will be trained -
Typically, this will require hundreds lines of machine learning codes. On Datature's platform, however, workflows are built visually, without any codes, using our graphical user interface. We will then transpile the visual workflow into working and bug-free codes for you. To see how this fits into our overall work pipeline, look at Datature Pipeline.
The process is simple -
Build a workflow
By dragging and dropping some building blocks, which we refer to as modules, on the workflow canvas, to define the settings for your Dataset Split, Data Augmentations, and Model Selection.
No, changing the workflow after the the run has been started will not affect the run. You can monitor training process and then delete the training even if it is ongoing.
You can have multiple workflows but only one workflow on each canvas. Having multiple workflows can help with experimentation and version control, as well as simultaneous testing of different setups. Simply create a new workflow for each of your experimentation and version control.
Check for error messages. Make sure the modules are connected to each other, the workflow cannot be run if there are 'floating' modules on the canvas. Contact Us if you require support on getting your workflow to run.
Updated 4 months ago