MLSteam Model SDK

💡This example demonstrates how to use MLSteam Model SDK to download an encrypted model for prediction. It has been verified in an MLSteam lab running from a ubuntu:20.04 image.

Create a new model version

  1. Create a new folder (previously known as dataset) in MLSteam.

  2. Upload the model.zip file into folder and extract it.

  3. Create a new model named stock in MLSteam.

  4. Within the stock model, create a new packaged or encrypted version named v1.

    • models: models/ (folder)

    • hooks: hooks/ (folder)

    • manifest: manifest.json

Install required SDK packages

If the last step of install-themisdev fails in JupyterLab or if you are not using Ubuntu, you may install the package according to the official instructions instead.

%pip install -U pip
%pip install -U setuptools
%pip install -U mlsteam-model-sdk
!mlsteam-model-cli install-themisdev -p ubuntu

Install required model packages

  1. Replace the following pip-requirements path by the actual file location.

%pip install -U -r examples/tensorflow_stockpred/requirements.txt

Initialize SDK

  1. Replace the value of --default_project_val by the actual project name (format: PROJECT_OWNER_NAME/PROJECT_NAME).

  2. Change --default_model_val also if you use another model name.

!rm -rf ~/.mlsteam-model-sdk
!mlsteam-model-cli init \
    --default_project_type name \
    --default_project_val "admin/test" \
    --default_model_type name \
    --default_model_val "stock"
!cat ~/.mlsteam-model-sdk/cfg.ini

from mlsteam_model_sdk.sdk.model import Model

sdk_model = Model()

sdk_model.download_model_version(version_name='v1')

mv = sdk_model.load_model_version(version_name='v1')

inputs = [50.0]*80
outputs = mv.predict(inputs=inputs)
print(outputs)

mv.models