Visualize model graph tensors using Visualizers
Neural networks are made up of components that process generic, undefined data. In order to make sense of the data, Tensorleap enables you to visualize the data contained in any node, with the exception of the Optimizer, using a Visualizer node. You can visualize the input(s), output(s) (prediction), ground truth(s), and intermediate tensors. When running Sample Analysis, the visualizations related to each Visualizer node is shown on the left panel.
When performing Sample Analysis, error analysis algorithms analyze the sample's error throughout the model and project it to each input. This will generate various heat-maps highlighting different features. Therefore, at least one connection of the Dataset Node (model-graph input) must be connected to a Visualizer.
You can write your own Custom Visualizers in the Integration Script (see Visualizer Function), or use one of the following out-of-the-box Decoders:
- HorizontalBar - visualize numeric data as horizontal bar with labels
- Image - visualize 2D image
- Graph - visualize numeric data as a graph
- Numeric - presents a number or an array of numbers
- Text - tokens representation of text (for token-to-text Visualizer, use the Decoder Function with your tokenizer)
- ImageMask - visualize an image and an overlay mask (commonly used in image segmentation)
- TextMask - visualize text and an overlay mask (commonly used in text segmentation)
To add a Visualizer to the model graph, follow the following steps:
- 1.On the Network view, right-click anywhere and select Visualizer from the menu.
- 2.When the Visualizer node appears in the Network view, click it to open a panel to the right. From here, you can set the
Selected Visualizerto one of the out-of-the-box Decoders, or custom Decoders, set in the Integration Script (see Visualizer Function).
- 3.Connect the Dataset Node output you to the Visualizer node. (at least one visualizer must be connected)
- 4.(optional) - Add and connect an additional Visualizer nodes to selected nodes' outputs.
This example demonstrates how to add visualizers to an image classification model. An image decoder is added to the input, and two
HorizontalBarvisualizers are added to the prediction and ground-truth in order to compare the two: