Simply download and install the Zetane software and load previously saved ML models.
Load your test input to understand how the data changes as it passes through the layers and operators of your model.
Save time and headaches by quickly pinpointing issues and validate your model by inspecting its internal data.
Dive in your ML model to inspect the internal tensors, layer outputs, feature maps, weights and architecture. Save valuable hours and weeks by eliminating much guesswork and focusing on what matters most to achieve the performance you seek.
Analyse the outliers and operational test data to assess the operational boundaries of your ML solution. Conduct rigorous testing and evaluation of ML models to demonstrate to operational stakeholders that all risks have been identified and mitigated.
The reality is that very few people in industry understand machine learning. This leads to much misconception resulting in failed AI projects. With Zetane, you can create human understandable demonstrations of your AI solution that all stakeholders will understand. This will drive collaboration, buy-in and ensure you receive feedback early to ensure project success.
Everything. The model architecture and all the internal tensors, each layer output, feature maps, weights, biases, values and operator attributes.
After saving your Keras, Pytorch or ONNX model from your own workflow, you can load your model so that you can inspect, debug and validate it in Zetane Viewer. Using the Zetane python package, you can also use the Python-Zetane API to send models, NumPy arrays, text, images, point clouds and meshes to the Zetane Engine.
The Zetane Viewer Pro allows you to load your own input so you can identify, validate and determine how to improve your model.
By carefully encapsulating the data. At the highest level you can navigate your model architecture, zoom on specific nodes. Clicking on the tensor buttons allows you to inspect the tensor and values distributions. Then, each tensor has different types of visualization to choose from.
If the model can be saved you should be able to open it to inspect all its components. Regardless of their purpose, ML models are composed of operation and tensors which we help you access with no effort.
Freddy Lécué, Chief AI Scientist, Thales Group
Get started with the free Zetane Viewer (no email required)