mnist-scikit-svm-to-coreml

From a MNIST handwritten digit classifier to Apple's Core ML format

This tutorial goes through:

1. Loading MNIST handwritten image data
2. Training scikit-learn Support Vector Machine (SVM) model with the handwritten digit data
3. Converting scikit-learn SVM model into Apple's Core ML format
4. Loading the coreml model and testing

Setting Up Environment

We use coremeltools to convert scikit-learn model to coreml format: https://pypi.python.org/pypi/coremltools

You can use the following commands to create your environment. You just need to replace < projectname > with the name you prefer for the environment. Pay attention that venv works only with python version 3. I used python 3.6

$ python -m venv projectname
$ source projectname/bin/activate
(venv) $ pip install  jupyter
(venv) $ pip install ipykernel
(venv) $ ipython kernel install --user --name=projectname
(venv) $ pip install -r requirements.txt

To update coremltools:

pip install -U coremltools

Now you can launch jupyter notebook to go through the notebook.

jupyter-notebook 

After you have done with running jupyter notebook, you can deactivate the environment:

deactivate

Core ML format

If you have Xcode 9 installed on your system, you can see that "digit_recognizer.mlmodel" is recognized and market with the following icon.

You can drag and drop the following model in a Xcode project to include and use the model in your app. Or directly click on the open to see the model specs:

Notes

1. coremltools install command gave an error in the first try; however, running the command second time fixed the issue.

2. When trying to convert scikit-learn svm model to core ml model, you might get the following error:

RuntimeError: Got non-zero exit code 72 from xcrun. Output was: xcrun: error: unable to find utility "coremlcompiler", not a developer tool or in PATH

In order to fix it, you need to install Xcode 9 (beta) from Apple developer site and and set the path:

xcode-select --print-path (check the version of xcode currently used)
sudo xcode-select --switch /Applications/Xcode-beta.app/Contents/Developer (set the path to xcode-beta)