Gretel-synthetics

⚙️ When using a Record Factory watchdog thread, if the thread throws an exception, cease generating records

⚙️ Cache a configurable amount of invalid record details in the `RecordFactory` which can be accessed and inspected as needed

⚙️ Support using a thread to monitory synthetic record generation progress. This lets the status of invalid records be checked as well while generating.

⚙️ Adds a new `validation_split` bool option to the configuration object. Enabled by default. When enabled, a fraction of the training data is used as validation data. Use of a validation set is recommended as it helps prevent over-fitting and memorization. When enabled, 20% of data will be used for validation.

Docs:

https://gretel-synthetics.readthedocs.io/en/latest/api/config.html#gretel_synthetics.config.BaseConfig.validation_split

---

⚙️ When using the `RecordFactory` class to create synthetic records, the `generate_all()` method will buffer records so they can be assembled into a DataFrame. A new `callback` option is available, that when set to a callable, will receive a periodic object containing data generation metrics.

Docs:

https://gretel-synthetics.readthedocs.io/en/latest/api/batch.html#gretel_synthetics.batch.GenerationProgress

Automatically select character-based tokenization over SentencePiece if `vocab_size` is set to zero.