Changes in HeatWave 9.0.1-u1 (2024-09-02, General Availability)

• HeatWave AutoML

• HeatWave GenAI

• HeatWave Lakehouse

• HeatWave MySQL

HeatWave AutoML

• HeatWave AutoML now supports semi-supervised learning for anomaly detection. This type of machine learning algorithm uses a specific set of labeled data along with unlabeled data to detect anomalies. (WL #16373)

• HeatWave AutoML now supports topic modeling. This is an unsupervised machine learning technique that is capable of scanning a set of documents, detecting word and phrase patterns within them, and automatically clustering word groups and similar expressions that best characterize the documents. (WL #16275)

HeatWave GenAI

• HeatWave GenAI now supports multiple languages, which lets you use the HeatWave GenAI APIs in non-English languages. This includes ingestion of documents written in languages other than English, vector and similarity search, and querying these documents by giving prompts in the same language.

  For more information, see Languages. (WL #16475)

• HeatWave GenAI now supports large language model (LLM) inference batch processing, which lets you run the HeatWave GenAI routines on multiple queries, in parallel, across different nodes in the HeatWave cluster. Thus, it improves the LLM inference performance and throughput while keeping the inference latency of every request unchanged.

  For more information, see ML_GENERATE_TABLE, ML_RAG_TABLE, and ML_EMBED_TABLE. (WL #16399)

• The HeatWave GenAI ML_RAG routine now includes new filtering options, exclude_vector_store and exclude_document_name, which let you exclude specific vector store tables or documents from context retrieval. (WL #16401)

• You can now use incremental load to refresh vector store tables. For more information, see Loading Data Incrementally into the Vector Store Table. (WL #15817)

• Auto Parallel Load for unstructured data provides a new format option setting, auto_unstructured, which lets you ingest multiple files with different unstructured data formats in a single load.

  Additionally, for ingesting Microsoft PowerPoint and Microsoft Word files into the vector store, Auto Parallel Load introduces the following new format aliases: pptx and docx.

  For more information, see Ingesting Files Using Auto Parallel Load. (WL #16433)

HeatWave Lakehouse

• HeatWave Lakehouse now supports timestamp_format as a dialect parameter to customize the format for columns of the DATETIME and TIMESTAMP data types. You can also set the formats for specific columns by using the ENGINE_ATTRIBUTE option. Formats set at the column level override formats set with the dialect parameter. For more information, see Lakehouse External Table Syntax. (WL #16425)

• Some HeatWave Lakehouse error messages include the URL to the external object. The URL can contain up to 1024 characters, and exceed the error message limit of 512 characters. The truncated error message will now include a MySQL command to access the full error message.

  It is now possible to filter out HeatWave Lakehouse warning messages. The console Total Warnings will include both the displayed and the filtered warnings count. See: HeatWave Lakehouse Error Messages.

  HeatWave Lakehouse now supports max_error_count. (WL #16141)

HeatWave MySQL

• • GROUP_CONCAT() function now supports the CUBE, ROLLUP, and WITH ROLLUP options.

  • COUNT(DISTINCT) function now supports the CUBE, ROLLUP, and WITH ROLLUP options.

  (WL #16365)

• HeatWave MySQL now supports the following JSON functions:

  • JSON_INSERT()

  • JSON_REMOVE()

  • JSON_REPLACE()

  • JSON_SET()

  • JSON_ARRAY_APPEND()

  • JSON_ARRAY_INSERT()

  • JSON_MERGE_PATCH

  • JSON_MERGE_PRESERVE

  • JSON_MERGE()

  • JSON_SCHEMA_VALID()

  • JSON_SCHEMA_VALIDATION_REPORT()

  (WL #16345)