Practical forecasting with AutoML

In this tutorial, we will first learn how to set up forecasting problems with AutoML in the benign, clean-data case. Then we'll look at some of the more complex cases that come up in real world datasets and how to shape the data so that AutoML forecasting will eat it.

Where and when

Currently scheduled for Kalaloch (35/2615) on Jun 6, 2019 from 1pm to 3pm. Times and places sometimes change, verify with the official MLADS agenda.

Lesson plan

"Planning is useful though the plans are normally useless."

The plan for the tutorial (90 minutes) is:

15 minutes – set up with AutoML (clone this repo, AzureML notebooks repo, solve hiccups).
30 minutes – go through the standard forecasting notebooks
15 minutes - [slides] common data problems, how to diagnose and solve them
30 minutes – Demonstrations of data wrestling
30 minutes - schedule slip, questions, coffee breaks

Prerequisites and setup

You should already have these:

Bring a laptop with Python 3.6 and git client installed
Have an Azure subscription (see below if you don't have one)

Setting up a Notebook VM

The Notebook VM has the MachineLearningNotebooks repo already set up.

Go to the Portal and create an ML Workspace
Go to Notebook VMs and create one. This will take about 15 minutes.
Connect to the Jupyter notebook endpoint of the VM
Open a new terminal
`cd <you_username>
git clone https://github.com/singliar/MLADS2019
Run GroceryForecast.ipynb

Setting up AutoML on local

Clone the Machine Learning Notebooks repo.
Open a shell or command prompt window, go to /how-to-use-azureml/automated-machine-learning and execute the automl_setup script appropriate for your platform (Win, Linux, Mac). Many packages will be installed (5 minutes on good network).
A browser window with Jupyter will open. Ctrl+C the ipykernel in the terminal.
conda activate azure_automl
pip install binpacking
Re-start jupyter in the root directory of the repo (two folders up) with jupyter notebook
Open the setup notebook configuration.ipynb.
Make sure to use the azure_automl kernel.
Transfer your subscription id, resource group name and workspace you want to use into the second code cell of the notebook.
Run cells according to instructions.

Setting up this repo

One day, these tools, and much more, will become part of AutoML. Until they do, they live here.

Clone this repo
Copy config.json from your AutoML repo into this directory
Start a new jupyter notebook here and run the notebook.

How to get a free Azure subscription

Create a Microsoft account at https://outlook.com (skip if you already have an account @outlook.com, @Hotmail.com, or @live.com)
Use your Microsoft account to get the free Azure subscription https://azure.microsoft.com/en-us/free/

You get 200$ Azure credits expiring in 30 days
Credit card needed for identification purpose. You will not be charged even after 30 days or the 200$ credits are used

pawarbi / MLADS2019