PerceptiLabs
Search…
Training View
The Training View is displayed when training begins (by clicking Run in the Modeling Tool), and provides a number of stats about how the model is performing during training:
The following are the key controls in the Training View's tool bar:
- 1.Back to Model: Returns to the Modeling Tool.
- 2.Pause: pauses training.
- 3.Stop: stops training.
Downloading Stats
Most of the panes contain a download button that generates a screenshot of that tab's/pane's data:
A pane's download button can be used to download a screenshot of that specific pane.
This can be useful for grabbing snapshots of different statistics (e.g., when you pause training at a certain point).
Input Tab
The Input tab displays a visualization of the current data sample:
Target Tab
The Target tab shows a number of panes with vital statistics including loss and accuracy over all epochs, prediction versus ground truth (for the current data sample), and the predictions made so far for each classification (you want to strive for a single color per category):
Tip: You can toggle the display of specific data within a graph (e.g., validation data) by clicking on the respective item in the graph's legend.
Map and View Box
At the bottom of the Statistics View are the Map and View Box areas:
Map and View Box
The Map displays the model and the View Box displays real-time statistics about the training. When you click on individual components in the Map, the View Box displays statistics for the selected component, allowing you to analyze your model on a more granular level during training.
The View Box contains the three tabs described in the following sub sections.
Weights & Output Tab
The Weights & Output tab displays the Component's output on the right, and the weights used to create that output on the left. When training models, watch out for weights which are 0 or those which not changing.
Weights & Output Tab
Bias Tab
The Bias tab displays the bias for the current data sample:
Bias Tab
Gradients Tab
The Gradients tab displays the gradient descent for the model:
Gradients Tab
Last modified 6mo ago