There are several software and programs supports for windows 10 and it makes much easier to use. The latest windows software is windows ML and it is known as windows machine learning, it is mainly programmed with C, C++ and Java script in order to preview the machine learning models and gives quick evaluation about the trained models. The hardware changes in CPU and GPU helps to provide quick evaluation of trained models with maximum performance. The Windows ML allows the user to add Artificial Intelligence into Win32 or in Universal Windows Platform application.
How to build applications using Windows ML?
In order to build applications using Windows ML you have to follow the steps given below.
- Step 1: you need to have an ONNX model either it can be a trained ONNX model or convert any models trained in any other ML frame into ONNX using WinML tool.
- Step 2: Add that ONNX model file to the application.
- Step 3: once the file added to the application the model has to be integrated with application code.
- Step 4: once you complete integrating code then run the application on windows device.
How to get ONNX models to use in Windows ML?
The windows ML mostly evaluated trained models in ONNX format that is open neural network exchange format. This is because the ONNX format allows the user to interchange the trained models between different types of ML tools and frame works. The ONNX models can be either trained by using CNTK and PyTorch or user can get pre trained ONNX models.
In order to train the models using CNTK and PyTorch the user has to install visual studio and CNTK applications. Once installed user can choose sample by following steps
- Select file, choose option open and click on project to select the model file.
- Once the model open do right click on it and choose option set as startup project then click on run button to run the project.
- The user can train the model once the process is completed it can be viewed and edited using application
In order to get trained ONNX models you can download pre trained ONNX model from the ONNX model zoo. Then it can be trained as per user wish using services tools available in ML tools Once the model is trained it should be converted to ONNX format.
Conversion of ML models to ONNX models:
As windows ML only supports with ONNX format models so it is necessary to convert ML frame works to ONNX format. Even Windows ML allows you to convert any models trained in different framework into ONNX models using ONNXML tools. TF2ONNX option also can be used to convert different training frameworks models into ONNX model.
These conversion tools support different frameworks such as:
- Apple coreML
Apart from the above said frameworks the windows ML can change ONNX form to quantized ONNX form. Using Windows ML even floating point models can be converted to 16 bit precision models. Windows ML is not only used for conversion it can also be used to make custom ONNX operators