Limit forecasted values for a time series model

This tutorial teaches you how to use limits to narrow the forecasted results returned by a time series model. In this tutorial, you create two time series models over the same data, one model which uses limits and one model that doesn't use limits. This lets you compare the results returned by the models and understand the difference that specifying limits makes.

You use the new_york.citibike_trips data to train the models in this tutorial. This dataset contains information about Citi Bike trips in New York City.

Before following this tutorial, you should be familiar with single time series forecasting. Complete the Single time series forecasting from Google Analytics data tutorial for an introduction to this topic.

Required Permissions

• To create the dataset, you need the bigquery.datasets.create IAM permission.

• To create the connection resource, you need the following permissions:

  • bigquery.connections.create
  • bigquery.connections.get

• To create the model, you need the following permissions:

  • bigquery.jobs.create
  • bigquery.models.create
  • bigquery.models.getData
  • bigquery.models.updateData
  • bigquery.connections.delegate

• To run inference, you need the following permissions:

  • bigquery.models.getData
  • bigquery.jobs.create

For more information about IAM roles and permissions in BigQuery, see Introduction to IAM.

Objectives

In this tutorial, you use the following:

• The CREATE MODEL statement: to create a time series model.
• The ML.FORECAST function: to forecast daily total visits.

Costs

This tutorial uses billable components of Google Cloud, including the following:

• BigQuery
• BigQuery ML

For more information about BigQuery costs, see the BigQuery pricing page.

For more information about BigQuery ML costs, see BigQuery ML pricing.

Before you begin

1. Sign in to your Google Cloud account. If you're new to Google Cloud, create an account to evaluate how our products perform in real-world scenarios. New customers also get $300 in free credits to run, test, and deploy workloads.

2. In the Google Cloud console, on the project selector page, select or create a Google Cloud project.

   Go to project selector

3. Make sure that billing is enabled for your Google Cloud project.

4. Enable the BigQuery API.

   Enable the API

5. In the Google Cloud console, on the project selector page, select or create a Google Cloud project.

   Go to project selector

6. Make sure that billing is enabled for your Google Cloud project.

7. Enable the BigQuery API.

   Enable the API

Create a dataset

Create a BigQuery dataset to store your ML model:

1. In the Google Cloud console, go to the BigQuery page.

   Go to the BigQuery page

2. In the Explorer pane, click your project name.

3. Click more_vert View actions > Create dataset.

   

4. On the Create dataset page, do the following:

   • For Dataset ID, enter bqml_tutorial.

   • For Location type, select Multi-region, and then select US (multiple regions in United States).

     The public datasets are stored in the US multi-region. For simplicity, store your dataset in the same location.

   • Leave the remaining default settings as they are, and click Create dataset.

     

Visualize the time series you want to forecast

Before creating the model, it is useful to see what your input time series looks like.

In the following query, the FROM bigquery-public-data.new_york.citibike_trips clause indicates that you are querying the citibike_trips table in the new_yorkdataset.

In the SELECT statement, the query uses the EXTRACT function to extract the date information from the starttime column. The query uses the COUNT(*) clause to get the daily total number of Citi Bike trips.

#standardSQL
SELECT
   EXTRACT(DATE from starttime) AS date,
   COUNT(*) AS num_trips
FROM
  `bigquery-public-data`.new_york.citibike_trips
GROUP BY date

To run the query, use the following steps:

1. In the Google Cloud console, click the Compose new query button.

2. Enter the following GoogleSQL query in the query editor.

   #standardSQL
   SELECT
      EXTRACT(DATE from starttime) AS date,
      COUNT(*) AS num_trips
   FROM
     `bigquery-public-data`.new_york.citibike_trips
   GROUP BY date
   

3. Click Run. The query results similar to the following.

   

4. Use the Google Cloud console to chart the time series data. In the Query results pane, click the Chart tab. In the Chart configuration pane, choose Bar for the Chart type:

   

Create a time series model

Create a time series model, using the NYC Citi Bike trips data.

The following GoogleSQL query creates a model that forecasts daily total bike trips. The CREATE MODEL statement creates and trains a model named bqml_tutorial.nyc_citibike_arima_model.

#standardSQL
CREATE OR REPLACE MODEL bqml_tutorial.nyc_citibike_arima_model
OPTIONS
  (model_type = 'ARIMA_PLUS',
   time_series_timestamp_col = 'date',
   time_series_data_col = 'num_trips'
  ) AS
SELECT
   EXTRACT(DATE from starttime) AS date,
   COUNT(*) AS num_trips
FROM
  `bigquery-public-data`.new_york.citibike_trips
WHERE starttime > '2014-07-11' AND starttime < '2015-02-11'
GROUP BY date

The OPTIONS(model_type='ARIMA_PLUS', time_series_timestamp_col='date', ...) clause indicates that you are creating an ARIMA-based time series model. By default, auto_arima=TRUE, so the auto.ARIMA algorithm automatically tunes the hyperparameters in ARIMA_PLUS models. The algorithm fits dozens of candidate models and chooses the best one with the lowest Akaike information criterion (AIC). Additionally, because the default is data_frequency='AUTO_FREQUENCY', the training process automatically infers the data frequency of the input time series. Lastly, the CREATE MODEL statement uses decompose_time_series=TRUE by default, so both the history and forecast parts of the time series are saved in the model. You can use this information to further understand how the time series is forecasted by fetching the separate time series components such as seasonal periods.

Run the CREATE MODEL query to create and train your model:

1. In the Google Cloud console, click the Compose new query button.

2. Enter the following GoogleSQL query in the query editor.

   #standardSQL
   CREATE OR REPLACE MODEL `bqml_tutorial.nyc_citibike_arima_model`
   OPTIONS
     (model_type = 'ARIMA_PLUS',
      time_series_timestamp_col = 'date',
      time_series_data_col = 'num_trips'
     ) AS
   SELECT
      EXTRACT(DATE from starttime) AS date,
      COUNT(*) AS num_trips
   FROM
     `bigquery-public-data`.new_york.citibike_trips
   WHERE starttime > '2014-07-11' AND starttime < '2015-02-11'
   GROUP BY date
   

3. Click Run.

   The query takes approximately 17 seconds to complete, after which your model (nyc_citibike_arima_model) appears in the Explorer pane. Because the query uses a CREATE MODEL statement to create a model, there are no query results.

Forecast the time series and visualize the results

To explain how the time series is forecasted, visualize all the sub-time series components, such as seasonality and trend, using the ML.FORECAST function.

To do this, follow these steps:

1. In the Google Cloud console, click the Compose new query button.

2. Enter the following GoogleSQL query in the query editor.

   #standardSQL
     SELECT
     forecast_timestamp AS forecast_timestamp,
     history_value AS history_value,
     forecast_value AS forecast_value
     FROM ( (
        SELECT
           DATE(forecast_timestamp) AS forecast_timestamp,
           NULL AS history_value,
           forecast_value AS forecast_value
        FROM
           ML.FORECAST(MODEL bqml_tutorial.`nyc_citibike_arima_model`,
           STRUCT(365 AS horizon,
              0.9 AS confidence_level)) )
     UNION ALL (
        SELECT
           DATE(date_name) AS forecast_timestamp,
           num_trips AS history_value,
           NULL AS forecast_value
        FROM (
           SELECT
           EXTRACT(DATE
           FROM
              starttime) AS date_name,
           COUNT(*) AS num_trips
           FROM
           `bigquery-public-data`.new_york.citibike_trips
           WHERE
           starttime > '2014-07-11'
           AND starttime < '2015-02-11'
           GROUP BY
           date_name ) ))
     ORDER BY
     forecast_timestamp
   

3. Click Run. The query results similar to the following:

   

4. Use the Google Cloud console to chart the time series data. In the Query results pane, click the Chart tab:

   

The chart shows that the forecasted values for the daily total number of Citi Bike trips are negative numbers, which isn't useful. Using a model with limits instead improves the forecasted data.

Create a time series model with limits

Create a time series model with limits, using the NYC Citi Bike trips data.

The following GoogleSQL query creates a model that forecasts daily total bike trips. The CREATE MODEL statement creates and trains a model named bqml_tutorial.nyc_citibike_arima_model_with_limits. The only difference between this model and the model you created previously is the addition of the forecast_limit_lower_bound=0 option. This option causes the model to only forecast values that are greater than 0.

#standardSQL
CREATE OR REPLACE MODEL `bqml_tutorial.nyc_citibike_arima_model_with_limits`
OPTIONS
  (model_type = 'ARIMA_PLUS',
   time_series_timestamp_col = 'date',
   time_series_data_col = 'num_trips',
   forecast_limit_lower_bound = 0
  ) AS
SELECT
   EXTRACT(DATE from starttime) AS date,
   COUNT(*) AS num_trips
FROM
  `bigquery-public-data`.new_york.citibike_trips
WHERE starttime > '2014-07-11' AND starttime < '2015-02-11'
GROUP BY date

Run the CREATE MODEL query to create and train your model:

1. In the Google Cloud console, click the Compose new query button.

2. Enter the following GoogleSQL query in the query editor.

   #standardSQL
   CREATE OR REPLACE MODEL `bqml_tutorial.nyc_citibike_arima_model_with_limits`
   OPTIONS
     (model_type = 'ARIMA_PLUS',
      time_series_timestamp_col = 'date',
      time_series_data_col = 'num_trips',
      forecast_limit_lower_bound = 0
     ) AS
   SELECT
      EXTRACT(DATE from starttime) AS date,
      COUNT(*) AS num_trips
   FROM
     `bigquery-public-data`.new_york.citibike_trips
    WHERE starttime > '2014-07-11' AND starttime < '2015-02-11'
   GROUP BY date
   

3. Click Run.

   The query takes approximately 17 seconds to complete, after which your model (nyc_citibike_arima_model_with_limits) appears in the Explorer pane. Because the query uses a CREATE MODEL statement to create a model, there are no query results.

Forecast the time series by using the model with limits

1. In the Google Cloud console, click the Compose new query button.

2. Enter the following GoogleSQL query in the query editor.

   #standardSQL
     SELECT
     forecast_timestamp AS forecast_timestamp,
     history_value AS history_value,
     forecast_value AS forecast_value
     FROM ( (
        SELECT
           DATE(forecast_timestamp) AS forecast_timestamp,
           NULL AS history_value,
           forecast_value AS forecast_value
        FROM
           ML.FORECAST(MODEL `bqml_tutorial.nyc_citibike_arima_model_with_limits`,
           STRUCT(365 AS horizon,
              0.9 AS confidence_level)) )
     UNION ALL (
        SELECT
           DATE(date_name) AS forecast_timestamp,
           num_trips AS history_value,
           NULL AS forecast_value
        FROM (
           SELECT
           EXTRACT(DATE
           FROM
              starttime) AS date_name,
           COUNT(*) AS num_trips
           FROM
           `bigquery-public-data`.new_york.citibike_trips
           WHERE
           starttime > '2014-07-11'
           AND starttime < '2015-02-11'
           GROUP BY
           date_name ) ))
     ORDER BY
     forecast_timestamp
   

3. Click Run.

   

4. Use the Google Cloud console to chart the time series data. In the Query results pane, click the Chart tab:

   

The ARIMA PLUS model detects that the daily total number of Citi Bike trips is decreasing. Future forecasting values will follow this trend and give relatively smaller forecasting numbers the farther into the future you go. The chart shows that the forecasted values for the daily total number of Citi Bike trips are positive numbers, which is more useful. The model with limits detects that the daily total number of Citi Bike trips is decreasing, but it still gives meaningful forecasting values.

As this tutorial shows, the forecast_limit_lower_bound and forecast_limit_upper_bound options can help you get more meaningful forecasting values in similar scenarios to the one shown here, such as when forecasting stock prices or future sales numbers.

Delete your dataset

Deleting your project removes all datasets and all tables in the project. If you prefer to reuse the project, you can delete the dataset you created in this tutorial:

1. If necessary, open the BigQuery page in the Google Cloud console.

   Go to the BigQuery page

2. In the navigation, click the bqml_tutorial dataset you created.

3. Click Delete dataset on the right side of the window. This action deletes the dataset, the table, and all the data.

4. In the Delete dataset dialog, confirm the delete command by typing the name of your dataset (bqml_tutorial) and then click Delete.

Delete your project

To delete the project:

1. In the Google Cloud console, go to the Manage resources page.

   Go to Manage resources

2. In the project list, select the project that you want to delete, and then click Delete.
3. In the dialog, type the project ID, and then click Shut down to delete the project.

Whatsnext

• Learn how to perform multiple time-series forecasting with a single query from NYC Citi Bike trips data.
• Learn how to accelerate ARIMA_PLUS to enable forecast 1 million time series within hours.
• To learn more about machine learning, see the Machine learning crash course.
• For an overview of BigQuery ML, see Introduction to BigQuery ML.
• To learn more about the Google Cloud console, see Using the Google Cloud console.