Home - News ( United States | United Kingdom | Italy | Germany ) - Football scores

Showing content from https://docs.databricks.com/aws/en/machine-learning/automl/automl-covariate-forecast below:

AutoML: Improve forecasting with covariates (external regressors)

This article shows you how to use covariates, also known as external regressors, to improve AutoML forecasting models.

Covariates are additional variables outside the target time series that can improve forecasting models. For example, if you're forecasting hotel occupancy rates, knowing if it's the weekend could help predict customer behavior.

In this example, you:

1. Create a randomized time-series dataset.
2. Perform basic feature engineering work.
3. Store the dataset as a FeatureStore table.
4. Use the FeatureStore as covariates in an AutoML forecasting experiment.

This example uses randomly generated time series data for hotel occupancy rates in January 2024. Then, use AutoML to predict the occupancy_rate for the first day of February 2024.

Run the following code to generate the sample data.

Python

df = spark.sql("""SELECT explode(sequence(to_date('2024-01-01'), to_date('2024-01-31'), interval 1 day)) as date, rand() as occupancy_rate FROM (SELECT 1 as id) tmp ORDER BY date""")
display(df)

Use the sample dataset to feature engineer a feature called is_weekend that a binary classifier of whether or not a date is a weekend.

Python

from pyspark.sql.functions import dayofweek, when

def compute_hotel_weekend_features(df):
  ''' is_weekend feature computation code returns a DataFrame with 'date' as primary key'''
  return df.select("date").withColumn(
      "is_weekend",
      when(dayofweek("date").isin( 1, 2, 3, 4, 5), 0) 
      .when(dayofweek("date").isin(6, 7), 1) 
  )
hotel_weekend_feature_df = compute_hotel_weekend_features(df)

To use covariates on AutoML, you must use a Feature Store to join one or more covariate feature tables with the primary training data in AutoML.

Store the data frame hotel_weather_feature_df as a Feature Store.

Python

from databricks.feature_engineering import FeatureEngineeringClient
fe = FeatureEngineeringClient()

hotel_weekend_feature_table = fe.create_table(
  name='ml.default.hotel_weekend_features', 
  primary_keys=['date'],
  df=hotel_weekend_feature_df,
  description='Hotel is_weekend features table'
)

note

This example uses the Python FeatureEngineeringClient to create and write tables. However, you can also use SQL or DeltaLiveTables to write and create tables. See Work with feature tables in Unity Catalog for more options.

Use the feature_store_lookups parameter to pass the Feature Store to AutoML. feature_store_lookups contains a dictionary with two fields: table_name and lookup_key.

Python

hotel_weekend_feature_lookup = {
  "table_name": "ml.default.hotel_weekend_features", 
  "lookup_key": ["date"]
}
feature_lookups = [hotel_weekend_feature_lookup]

note

feature_store_lookups can contain multiple feature table lookups.

Use the following code to pass the features_lookups to an AutoML experiment API call.

Python

from databricks import automl
summary = automl.forecast(dataset=df, target_col="occupancy_rate", time_col="date", frequency="d", horizon=1, timeout_minutes=30, identity_col=None, feature_store_lookups=feature_lookups)

• AutoML Python API reference
• Work with feature tables in Unity Catalog

RetroSearch is an open source project built by @garambo | Open a GitHub Issue

Search and Browse the WWW like it's 1997 | Search results from DuckDuckGo

HTML: 3.2 | Encoding: UTF-8 | Version: 0.7.4