Build an AI model
In the Modeling view, click New experiment and name the experiment. Remember to use a good naming strategy for your versions. Version 1, Version 2, Version 3, etc., becomes a little hard to decipher after a while.
Create ready-to-run experiment with the Experiment wizard
The goal of the Experiment wizard is to make it easy to create a ready-to-run experiment, where all necessary parameters are preset. Select input data, snippet, and weights, and you’re good to run.
The idea is to make it easy for DL-non-savvy and platform-non-savvy users to get started quickly on the platform. The wizard takes advantage of the available info, provides a good starting point, and makes sure the user gets a good result with almost no effort.
When all choices are made, a new experiment will be constructed that already has the input, and target blocks initialized with the specified features, has the loss function set up suitably, and is initialized with pre-trained weights (or no weights).
The first step is to select a dataset. The latest saved dataset and version will be autoselected. Change if you want to.
Based on the input dataset, the wizard will propose a Training subset and a Validation subset. Most often, these will be the subsets that are named something close to Training or Validation, pretty straight forward.
Choose a snippet
Based on the information in the dataset, the wizard selects features that look like input and target, as well as the problem type you’re trying to solve.
If you want, set all weights trainable by checking the Weights trainable (all blocks) box.
The Problem type, e.g., Image regression or Single-label text classification, will decide which snippets you can choose.
Example: With text as input data and Problem type Single-label text classification, pick the BERT snippet in case their input feature looks like text.
Example: With images in size 224x224 or above as input data and Problem type Image regression, pick the EfficientNet snippet.
All your choices decide how the final experiment will look like and what parameters will be prepopulated. You can obviously always adjust and override the suggested snippet, and pick something else, or choose to proceed with a blank experiment instead.
If the snippet has pretrained weights, then click the Initialize weights tab and select Weight initialization and, if you want, set all weights trainable by checking the Weights trainable (all blocks) box.
Note that third party terms may apply if you use pretrained weights.
In the Modeling view
When you’ve created an experiment with the Experiment wizard, a ready-to-run model will populate the Modeling canvas. Everything needed will be set, input and target feature, weights (if you wanted those), loss function, activation, runtime settings, i.e., batch size, learning rate, number of epochs.
You can still proceed as normal, edit/add/remove anything on the Modeling canvas or override all the existing choices, delete the snippet, and add a new one.
Create blank experiment
If you feel that you know exactly what you’re doing, you can click on Create blank experiment. Then a new experiment will open up with a clean blank Modeling canvas.
Add a block or a snippet to an experiment
The new block will be connected if you select an unconnected block on the canvas and then click to add a new block. Then the new block will appear under the previous block on the canvas. A connection between the two blocks is created automatically.
The new block will be unconnected if no block has been selected on the canvas. Then drag the new block to where you want it in the model. Connect the new block to another block by using hold-and-drag between the connecting points.
If needed, remove a connector or block by selecting it and then pushing the delete or backward key on the keyboard or clicking the Delete button in the GUI.
Set block parameters
Select the newly added block. Set the parameters for the block in the Inspector. The Information pop-up will give you information on which setting you need to set.