Build Models
Last updated
Last updated
After you have loaded your data and PerceptiLabs has generated a model for you, it's time to start working with that model.
You model consists of:
On or more Input Components: these act as your data source for your inputs.
One or more Target Components: these contain your targets. You need to ensure that your predictions (last layer of your model) is connected to them.
One or more Components that connect the Input and Target to transform your data. For example, in the following model generated for MNIST dataset, a Convolution and two Dense Components make up the model and transform your Input data to predictions that are comparable with your Targets:
The following are the key customizations you can make as you iterate on your model:
Swap Components (e.g., employ Deep Learning Components like VGG16 which have been pre-trained with ImageNet data, replace the Convolution and Dense Components with a UNet Component, etc.).
Adjust Component settings in the Settings Pane. See the Component topics for descriptions of each setting.
Configure data settings. See Dataset Configuration Settings for more information.
Modify Components in a low-code manner. See Code Editor for more information.
Create your own Components. See Custom Component for more information.
For example, follow the steps below to incorporate a second Convolution Component into the MNIST model:
1) Remove the connection between the Convolution Component and the first Dense Component.
2) Move the two Dense Components and the Target to the right.
3) Click the Deep Learning menu, select Convolution, and drag it onto the model.
4) Connect the Output socket of the existing Convolution Component to the Input socket of your new Convolution Component.
5) Connect the Output socket of your new Convolution Component to the Input socket of the first Dense Component.
6) Select your new Convolution Component and configure its Feature maps setting to 16 in the Settings pane.
The updated model should now look as follows:
This example showed a few of the ways that you can modify an existing model.
Now let's learn how to train the model you have built!
