Intent

• Showcase Azure Machine Learning AutoML time series capabilities.
• Dataset is sourced from here, and contains daily energy consumption data for 2020 across different East Coast counties and districts.
• These scripts do not extend into deploying the model through a web service or managed online endpoint. This is geared to understanding the workflows around training and prediction through the Python SDK.

Notes

• Python version: python=3.7
• A pre-condition is to specify a sub.env file with SUB=<your subscription id> in the root of the folder.
• A Makefile shows the steps to create the infrastructure, run the pipeline and then use the best model to generate predictions.
• Though the requirements.txt exists to install all the needed libraries, the various stages of requiring various installs are shown below:
  • pip install azureml-core and pip install python-dotenv (before running clusters.py)
  • pip install pandas and pip install azureml-dataset-runtime (before running datasets.py)
  • pip install azureml.train, pip install azureml.train.automl and pip install azureml.pipeline (before running create_pipeline.py)
  • pip install matplotlib (before running metric_evaluation.py)

Final Prediction