Results

After starting the model training, you will be redirected to the “Results” tab.

Results Tab

The “Results” tab includes:

Model Growth Chart
Presents iterations of growth and model construction on a single graph. With this feature, it is possible to compare, choose, and download models with differences of up to kilobytes in size and hundredths of a percent in accuracy, making it possible to embed these models in the most resource-constrained devices. (Watch the video to learn more)

Model Growth Chart

Training progress
Displays the progress of model creation. You can stop training anytime by pressing the “Stop” button if you are satisfied with the current models. Once the training is finished or stopped, you will be able to view the training log by clicking on the “Log” button that appears next to the progress bar.

Training Progress

Processed data
You can see what data the model was trained on, taking into account the additional settings you have chosen, for example, features created from Feature Extraction. By clicking on the "Data Analytics" button, you can access an overview of the analytics charts based on the processed data. (Read more)

Processed Data

Model Info Area
Displays information related to the model that has been selected in the model growth chart.

Info area contains:

The target metric value and its name. Two types of metrics are displayed: cross-validation and holdout (provided the holdout dataset was used prior to training). To switch between different metric types, use the switch located in the chart area.

The total footprint of the selected model. Here you can find information on the estimated usage of SRAM and FLASH memory by the model, inference engine, and signal processing. Estimated values are provided for the target hardware selected. (these values are provided as an example). After completing the training, C libraries will be prepared for a wider range of hardware types, and source code will be available for enterprise plans.

The “C Library” button for the selected model, that allows users to download the resulting model and code for inference on the device.

Model Info Area

C library

The archive contains the following folders and files:

artifacts – contains models converted to various formats, as well as an executable file for predictions on the desktop.

neuton – contains libraries for embedding, Cortex M0, Cortex M4, Cortex M33, and STMicro ISPU are supported.

neuton-generated – contains information necessary for the correct operation of libraries.

LICENSE – contains the possibilities and restrictions on the use of Neuton’s intellectual property.

README – contains instructions for leveraging libraries.

We don't recommend that you modify any files in the archive. Unsupervised changing of files may cause errors in model inference.

C library Archive

Analytics Tools

Model Quality Diagram
Simplifies the process of evaluating the model quality. (Read more)

Feature Importance Matrix (FIM)
Helps to understand how each feature impacts the model. (Read more)

Confusion Matrix (for classification tasks only)
Shows the number of correct and incorrect predictions based on the validation data. (Read more)

Analytics Tools