Skip to content

Latest commit

 

History

History
98 lines (61 loc) · 3.01 KB

model_visualizer.md

File metadata and controls

98 lines (61 loc) · 3.01 KB

Model Visualizer

This describes how to view an ML model in an interactive webpage using the MLTK's commands/APIs.

_Any_  `.tflite` or `.h5` model file will work with the model visualizer.  
i.e. The model file does _not_ need to be generated by the MLTK to view the model.

Quick Reference

Overview

Model visualization allows for viewing how the various layers of a model are connected. Model visualization is enabled using the Netron machine learning model viewer, a tool for viewing models in an interactive webpage.

- Model visualization is done _entirely_ in the local web-browser. The model is _not_ uploaded to any remote servers
- Models may also be viewed by dragging and dropping a `.tflite` or `.h5` model file on the [http://netron.app](https://netron.app) webpage

Command

Model visualization from the command-line is done using the view operation.

For more details on the available command-line options, issue the command:

mltk view --help

Example 1: View Keras model

In this example, we view the trained .h5 model file in the image_classification model's archive.

NOTE: The model graph will appear in your web-browser.

mltk view image_classification

Example 2: View Tensorflow-Lite model

In this example, we view the trained .tflite model file in the image_classification model's archive.

NOTE: The model graph will appear in your web-browser.

mltk view image_classification --tflite

Example 3: View external Tensorflow-Lite model

The given model need not be generated by the MLTK. External models are also supported by the view command.

NOTE: The model graph will appear in your web-browser.

mltk view ~/workspace/my_model.tflite

Example 4: View model before training

Training a model can be very time-consuming, and it is useful to view a model before investing time and energy into training it. For this reason, the MLTK view command features a --build flag to build a model and view it before the model is fully trained.

In this example, the image_classification model is built at command-execution-time and this file is opened in the viewer. Note that only the model specification script is required, it does not need to be trained first.

NOTE: The model graph will appear in your web-browser.

mltk view image_classification --tflite --build

Python API

Model visualization is accessible via the view_model API

Examples using this API may be found in view_model.ipynb