Nixtla’s StatsForecast integration offers univariate time series forecasting models. StatsForecast uses classical methods such as ARIMA, rather than deep learning. Models train very quickly and generalize well, so are unlikely to overfit. Models also perform well on short time series, where deep learning models may be more likely to overfit.

You can learn more about its features here.

​
How to bring StatsForecast Models to MindsDB

Before creating a model, you will need to create an ML engine for StatsForecast using the CREATE ML_ENGINE statement:

CREATE ML_ENGINE statsforecast
FROM statsforecast;

Once the ML engine is created, we use the CREATE MODEL statement to create the StatsForecast model in MindsDB.

CREATE MODEL model_name
FROM data_source
  (SELECT * FROM table_name)
PREDICT column_to_be_predicted
GROUP BY column_name, column_name, ...
ORDER BY date_column
WINDOW 12 -- model looks back at sets of 12 rows each
HORIZON 3 -- model forecasts the next 3 rows
USING
  engine = 'statsforecast',
  model_name = 'model',
  frequency = 'X',
  season_length = 1,
  hierarchy = ['column'];

The following parameters can be used while creating the StatsForecast model:

  • model_name is an optional parameter that lets users specify one of the models from this list, which otherwise is chosen automatically.
  • frequency is an optional parameter that defines the frequency of data such as daily, weekly, monthly, etc. Available values include “H”, “M”, “MS”, “Q”, “SM”, “BM”, “BMS”, “BQ”, “BH”.
  • season_length is an optional parameter that defines the length of the season depending on frequency. For instance, season_length defaults to 12 if frequency is set to M (months).
  • hierarchy is an optional parameter that may improve prediction accuracy when the data has a hierarchical structure. See more here.

To ensure that the model is created based on the StatsForecast engine, include the USING clause at the end.

​
Example

Let’s go through an example of how to use Nixtla’s StatsForecast with MindsDB to forecast monthly expenditures.

Please note that before using the StatsForecast engine, you should create it from the MindsDB editor, or other clients through which you interact with MindsDB, with the below command:

CREATE ML_ENGINE statsforecast
FROM statsforecast;

You can check the available engines with this command:

SHOW ML_ENGINES;

If you see the StatsForecast engine on the list, you are ready to follow the tutorials.

​
Tutorial using SQL

In this tutorial, we create a model to predict expenditures based on historical data using the StatsForecast engine.

We use a table from our MySQL public demo database, so let’s start by connecting MindsDB to it:

CREATE DATABASE mysql_demo_db
WITH ENGINE = 'mysql',
PARAMETERS = {
    "user": "user",
    "password": "MindsDBUser123!",
    "host": "samples.mindsdb.com",
    "port": "3306",
    "database": "public"
};

Now that we’ve connected our database to MindsDB, let’s query the data to be used in the example:

SELECT *
FROM mysql_demo_db.historical_expenditures
LIMIT 3;

Here is the output:

+------------+----------+-------------+
| month      | category | expenditure |
+------------+----------+-------------+
| 1982-04-01 | food     | 1162.6      |
| 1982-05-01 | food     | 1150.9      |
| 1982-06-01 | food     | 1160        |
+------------+----------+-------------+

The historical_expenditures table stores monthly expenditure data for various categories, such as food, clothing, industry, and more.

Let’s create a model table to predict the expenditures:

CREATE MODEL quarterly_expenditure_forecaster
FROM mysql_demo_db
  (SELECT * FROM historical_expenditures)
PREDICT expenditure
GROUP BY category
ORDER BY month
HORIZON 3
USING ENGINE = 'statsforecast';

Please visit our docs on the CREATE MODEL statement to learn more.

Please note that the WINDOW clause is not required because StatsForecast automatically calculates the best window as part of hyperparameter tuning.

The ENGINE parameter in the USING clause specifies the ML engine used to make predictions.

We can check the training status with the following query:

DESCRIBE quarterly_expenditure_forecaster;

One of the pros of using the StatsForecast engine is that it is fast - it doesn’t take long until the model completes the training process.

Once the model status is complete, the behavior is the same as with any other AI table – you can query for batch predictions by joining it with a data table:

SELECT m.month as month, m.expenditure as forecasted
FROM mindsdb.quarterly_expenditure_forecaster as m
JOIN mysql_demo_db.historical_expenditures as t
WHERE t.month > LATEST
AND t.category = 'food';

Here is the output data:

+----------------------------+-----------------+
| month                      | forecasted      |
+----------------------------+-----------------+
| 2017-10-01 00:00:00.000000 | 10256.251953125 |
| 2017-11-01 00:00:00.000000 | 10182.58984375  |
| 2017-12-01 00:00:00.000000 | 10316.259765625 |
+----------------------------+-----------------+

The historical_expenditures table is used to make batch predictions. Upon joining the quarterly_expenditure_forecaster model with the historical_expenditures table, we get predictions for the next quarter as defined by the HORIZON 3 clause.

Please note that the output month column contains both the date and timestamp. This format is used by default, as the timestamp is required when dealing with the hourly frequency of data.

MindsDB provides the LATEST keyword that marks the latest training data point. In the WHERE clause, we specify the month > LATEST condition to ensure the predictions are made for data after the latest training data point.

Let’s consider our quarterly_expenditure_forecaster model. We train the model using data until the third quarter of 2017, and the predictions come for the fourth quarter of 2017 (as defined by HORIZON 3).

​
Tutorial using MQL

In this tutorial, we create a model to predict expenditures based on historical data using the StatsForecast engine.

Before we start, visit our docs to learn how to connect Mongo Compass and Mongo Shell to MindsDB.

We use a collection from our Mongo public demo database, so let’s start by connecting MindsDB to it from Mongo Compass or Mongo Shell:

> use mindsdb
> db.databases.insertOne({
        'name': 'mongo_demo_db',
        'engine': 'mongodb',
        'connection_args': {
            "host": "mongodb+srv://user:MindsDBUser123!@demo-data-mdb.trzfwvb.mongodb.net/",
            "database": "public"
        }
   })

Now that we’ve connected our database to MindsDB, let’s query the data to be used in the example.

> use mongo_demo_db
> db.historical_expenditures.find({}).limit(3)

Here is the output:

{
  _id: '63fd2388bee7187f230f56fc',
  month: '1982-04-01',
  category: 'food',
  expenditure: '1162.6'
}
{
  _id: '63fd2388bee7187f230f56fd',
  month: '1982-05-01',
  category: 'food',
  expenditure: '1150.9'
}
{
  _id: '63fd2388bee7187f230f56fe',
  month: '1982-06-01',
  category: 'food',
  expenditure: '1160'
}

The historical_expenditures collection stores monthly expenditure data for various categories, such as food, clothing, industry, and more.

Let’s create a model to predict the expenditures:

> use mindsdb
> db.predictors.insertOne({
       name: 'quarterly_expenditure_forecaster',
       predict: 'expenditure',
       connection: 'mongo_demo_db',
       select_data_query: 'db.historical_expenditures.find({})',
       training_options: {
          timeseries_settings: {                
              order_by: ['month'],               
              group_by: ['category'],
              horizon: 3           
          },
          engine: 'statsforecast'
      }  
  })

Please visit our docs on the insertOne statement to learn more.

Please note that the window clause is not required because StatsForecast automatically calculates the best window as part of hyperparameter tuning.

The engine parameter in the training_options clause specifies the ML engine used to make predictions.

We can check the training status with the following query:

> db.models.find({
      name: 'quarterly_expenditure_forecaster'
  })

One of the pros of using the StatsForecast engine is that it is fast - it doesn’t take long until the model completes the training process.

Once the model status is complete, the behavior is the same as with any other AI collection – you can query for batch predictions by joining it with a data collection:

> db.quarterly_expenditure_forecaster.find({
      "collection": "mongo_pred_01.historical_expenditures",
      "query": {"category": "food"}
  }).limit(3)

By default the forecasts are made for month > LATEST.

Here is the output data:

{
  _id: '63fd2388bee7187f230f58a5',
  month: 2017-10-01T00:00:00.000Z,
  category: 'food',
  expenditure: 10256.251953125
}
{
  _id: '63fd2388bee7187f230f58a4',
  month: 2017-11-01T00:00:00.000Z,
  category: 'food',
  expenditure: 10182.58984375
}
{
  _id: '63fd2388bee7187f230f58a3',
  month: 2017-12-01T00:00:00.000Z,
  category: 'food',
  expenditure: 10316.259765625
}

The historical_expenditures collection is used to make batch predictions. Upon joining the quarterly_expenditure_forecaster model with the historical_expenditures collection, we get predictions for the next quarter as defined by the horizon: 3 clause.

Please note that the output month column contains both the date and timestamp. This format is used by default, as the timestamp is required when dealing with the hourly frequency of data.

MindsDB provides the latest keyword that marks the latest training data point. In the where clause, we specify the month > latest condition to ensure the predictions are made for data after the latest training data point.

Let’s consider our quarterly_expenditure_forecaster model. We train the model using data until the third quarter of 2017, and the predictions come for the fourth quarter of 2017 (as defined by horizon: 3).

​
StatsForecast + HierarchicalForecast

The StatsForecast handler also supports hierarchical reconciliation via Nixtla’s HierarchicalForecast package. Hierarchical reconciliation may improve prediction accuracy when the data has a hierarchical structure.

In this example, there may be a hierarchy as total expenditure is comprised of 7 different categories.

SELECT DISTINCT category
FROM mysql_demo_db.historical_expenditures;

Here are the available categories:

+-------------------+
| category          |
+-------------------+
| food              |
| household_goods   |
| clothing          |
| department_stores |
| other             |
| cafes             |
| industry          |
+-------------------+

Spending in each category may be related over time. For example, if spending on food rises in October 2017, it may be more likely that spending on cafes also rises in October 2017. Hierarchical reconciliation can account for this shared information.

Here is how we can create a model:

CREATE MODEL hierarchical_expenditure_forecaster
FROM mysql_demo_db
  (SELECT * FROM historical_expenditures)
PREDICT expenditure
GROUP BY category
ORDER BY month
HORIZON 3
USING
  ENGINE = 'statsforecast',
  HIERARCHY = [‘category’];

The CREATE MODEL statement creates, trains, and deploys the model. Here, we predict the expenditure column values. As it is a time series model, we order the data by the month column. Additionally, we group data by the category column - the predictions are made for each group independently (here, for each category). The HORIZON clause defines for how many rows the predictions are made (here, for the next 3 rows).

You can use the DESCRIBE [MODEL] command to check for details:

DESCRIBE hierarchical_expenditure_forecaster.model;

On execution, we get:

+------------+-----------+---------------+--------------+
| model_name | frequency | season_length | hierarchy    |
+------------+-----------+---------------+--------------+
| AutoARIMA  | MS        | 1             | ["category"] |
+------------+-----------+---------------+--------------+

Predictions with this model account for the hierarchical structure. The output may differ from the default model, which does not assume any hierarchy.

Was this page helpful?

NeuralForecastTimeGPT