Hyperparameter tuning with BQML

BQML allows you to fine-tune hyperparameters when building ML models through the use of CREATE MODEL statements. This process, known as hyperparameter tuning, is a commonly employed method for enhancing model accuracy by finding the ideal set of hyperparameters.

Here’s an example BigQuery SQL statement:

{CREATE OR REPLACE MODEL} model_name
OPTIONS(Existing Training Options,
   NUM_TRIALS = int64_value, [, MAX_PARALLEL_TRIALS = int64_value ]
   [, HPARAM_TUNING_ALGORITHM = { 'VIZIER_DEFAULT' | 'RANDOM_SEARCH' | 'GRID_SEARCH' } ]
   [, hyperparameter={HPARAM_RANGE(min, max) | HPARAM_CANDIDATES([candidates]) }... ]
   [, HPARAM_TUNING_OBJECTIVES = { 'R2_SCORE' | 'ROC_AUC' | ... } ]
   [, DATA_SPLIT_METHOD = { 'AUTO_SPLIT' | 'RANDOM' | 'CUSTOM' | 'SEQ' | 'NO_SPLIT' } ]...