[PYTHON] Time series data prediction by AutoML (automatic machine learning)

Here are the steps to learn a time-series predictive regression model using Azure Machine Learning's AutoML (Automatic Machine Learning).

AutoML (Automated ML) is a function that automatically performs processes such as feature engineering, model selection, and hyperparameter selection on the Azure Machine Learning side by simply inputting data and basic settings. See below for details. What is Automatic Machine Learning (AutoML)

Below is the sample code and data used in this article. : robot: Sample code (github) : beers: [Sample data (for training) --1992-2016 beer production time series data](https://github.com/Azure/MachineLearningNotebooks/blob/master/how-to-use-azureml/automated- machine-learning / forecasting-beer-remote / Beer_no_valid_split_train.csv) : beers: [Sample data (for testing) --2017-2018 beer production time series data](https://github.com/Azure/MachineLearningNotebooks/blob/master/how-to-use-azureml/automated- machine-learning / forecasting-beer-remote / Beer_no_valid_split_test.csv)

Based on the following training data (1992-2016), we will forecast the demand for beer after 2017. If you have CSV data with two columns, date and numeric data (such as sales), you can try the same procedure.

Creating Azure Machine Learning

1. Access to the Azure portal (https://ms.portal.azure.com)
2. Press the menu in the following order
3. [+ Create Resource] 2. [AI + Machine Learning] 3. [Machine Learning]
4. Set the following items as appropriate

• Workspace name: Optional
• Resource group: Optional
• Location: Eastern Japan
• Workspace Edition: Enterprise (Choose Enterprise instead of Basic to use Auto ML) After the above settings, click the [Confirm and Create] button to start deployment.

Launch Azure Machine Learning Studio

From the ML screen, click [Launch Now] in Azure Machine Learning Studio (Alternatively, go to https://ml.azure.com and open the Azure Machine Learning studio screen directly)

Creating a compute instance

1. Integrated development screen Open [Computing] of Azure Machine Learning studio
2. Select the Compute Instance tab and click + New
3. Set the following items as appropriate

• Computing name: Optional
• Virtual machine type： CPU (Central Processing Unit)
• Virtual machine size: Default (Standard_DS3_v2)
• 4 vCPU core is sufficient

After the above settings, click the [Create] button to start deployment.

Creating a training cluster

1. Integrated development screen Open [Computing] of Azure Machine Learning studio Select the [Compute clusters] tab and click [+ New]
2. Set the following items as appropriate

• Computing name: Optional
• Virtual machine type： CPU (Central Processing Unit)
• Virtual machine size: Default (Standard_DS3_v2)
• 4 vCPU core is sufficient
• Minimum number of nodes: 0
• Maximum number of nodes: 1

After the above settings, click the [Create] button to start deployment.

Upload sample code & data

1. From ml.azure.com, access the integrated development screen Azure Machine Learning studio.
2. Click Notebook, click Upload Folder, and zip downloaded from github (https://github.com/yahanda/time-series-forecasting-with-auto-ml) Upload the unzipped folder.

Run nodebook

1. Click beer-forcasting-automated-ml.ipynb, then click Jupyter> Edit with Jupyter
2. Kernel settings
3. Specify the Python environment used by Jupyter
4. Select Python 3.6 – AzureML
5. Execute cell (There are 3 ways to execute cell)
6. Shortcut: Ctrl + Enter (recommended)
7. Select Run Cells from the Cell bar
8. Select [> Run]

Details of what you are doing can be found in Notebook Please refer to the explanation of.

When run to the end, the best model has a very high accuracy with an average absolute error rate (MAPE) of 6.3%.

Model MAPE:
0.06370896673493383

Model Accuracy:
0.9362910332650661

It is a plot of the predicted value and the actual value.

(Check model accuracy)

You can also check the information of various models that AutoML tried from the [Automatic ML] menu of Azure Machine Learning studio via GUI.

• Overview of the best model

• List of models AutoML has tried

• Model accuracy

• Importance of explanatory variables

reference

• Tutorial: Predict taxi fare using automatic machine learning
• Automatically train time series forecast model