Workspace default settings

The Workspace default settings menu allows to configure default parameters for workspaces. These settings ensure consistent behavior for data processing, privacy protection, and synthetic data generation. Below is a detailed explanation of the available options.

Access the workspace default settings

Note that user should be an Owner or Editor to access Workspace Default Settings.

1. Create or open the workspace.

2. Get the workspace ID https://demo.syntho.ai/hub/<WORKSPACE-ID>/hub/?tab=Main%20hub

3. Workspace ID can be i.e. c5f5aa8b-3491-4025-b637-1ccf6e83ec24

4. Go to https://.demo.syntho.ai/engines/<WORKSPACE-ID>/global_settings/, which in this example is https://.demo.syntho.ai/engines/c5f5aa8b-3491-4025-b637-1ccf6e83ec24/global_settings/

How to modify settings

1. Access the Workspace default settings menu.

2. Modify the required values directly.

3. Save changes to apply them to the workspace.

Configuration options

Below is an overview of the default settings and their functionalities:

General settings

Parameter	Default	Possible Values	Description
seed_value	42	Any integer (e.g., 0, 42, 1234)	Sets the seed used by generators with consistent mapping enabled or inherently consistent (e.g., hash) generators. Consistent Mapping: When use_seed is true, changing seed_value will rotate the mapping so the same input maps to a different output, allowing you to periodically change synthetic outputs for security.
use_seed	false	true or false (Boolean)	If true, all generators that support consistent mapping (mock, mask, hash) will use your specified seed_value. If false, the system does not use any seed for consistent mapping, meaning changes to seed_value have no effect on the mapping scheme.

PII model settings

• Specifies the natural language processing (NLP) models for PII detection and mockers.

• Example:

  • Language code: "en" with model name "en_core_web_md".

  • Supported languages: English, Dutch, German, Japanese.

Parameter	Default	Possible Values	Description
nlp_engine_name	"spacy"	"spacy" or any other supported NLP engine name	Determines which NLP engine is used for PII scanning and detection.
models	N/A	A list of dictionaries with "lang_code" and "model_name". E.g.: [{ "lang_code": "en", "model_name": "en_core_web_md" }]	Each item defines a specific language and its associated model for PII detection.
gpu	false	true or false (Boolean)	Toggles GPU acceleration. When true, models that support GPU will run faster but require a compatible GPU setup.

For more information please see Configure to use other NLP models (limited support).

Initialization and data handling

Parameter	Default	Possible Values	Description
initialization_mode	"SCRATCH"	"SCRATCH", "APPEND", or "READ_ONLY"	Defines how the workspace is initialized. • "SCRATCH" starts an empty workspace. • "APPEND" adds data to existing tables. • "READ_ONLY" prevents any modifications.
key_generation_method	"duplicate"	"generate", "duplicate", "hash"	Determines the method for generating keys. • "generate" creates new keys. • "duplicate" copies source keys. • "hash" applies a hash function.
n_parallel_pipeline_processes	1	Any integer (e.g., 1, 2, -1 for all CPUs)	Controls the number of column-processing jobs (fitting, transforming, inverse-transforming) that run in parallel. Higher values can speed up processing but use more system resources.
default_n_training_rows	100000	Any positive integer (e.g., 10, 1000, 100000)	Sets the default number of rows used to train synthetic data models. If the input dataset exceeds this number, only the specified number of rows is used for training (unless otherwise configured).

Privacy control defaults

Parameter	Default	Possible Values	Description
default_sample_noise_ratio	0.0001	Any positive numeric value (float). • 0 < ratio ≤ 1: relative noise • > 1: absolute std. dev.	Specifies the level of noise added to synthetic data. • Between 0 and 1: noise is added as a relative ratio. • Above 1: noise is treated as an absolute standard deviation.
default_min_sample_size	5	Any positive integer value	Minimum sample size used for model training
default_cardinality_threshold	10	Any integer ≥ 1 (e.g., 5, 10, 20)	Any category in a categorical column with occurrences below this threshold is considered “rare” and gets replaced.
default_rare_category_replacement	"*"	Any string (e.g., "*" or "other")	Placeholder for rare categories to preserve privacy.
default_clip_threshold	0	Any numeric value (integer or float). • 0: no clipping • Positive number: outlier limit	Limits extreme outliers in numeric columns by “clipping” values above/below a certain threshold.

Text processing

Parameter	Default	Possible Values	Description
default_text_processor_model_settings	No single value	Same format as pii_model_settings (language models, engine, GPU)	Specifies NLP models for advanced text processing tasks (non-PII or general text analytics).
default_textpii_parallel_jobs	2	Any integer ≥ 1; -1 to use all available processors	Defines how many parallel jobs are used when scanning text for PII. Increasing the number of jobs speeds up scanning but uses more CPU resources.
default_textpii_scan_batch_size	1000	Any integer ≥ 1 (e.g., 100, 1000, 5000)	Batch size for PII detection in text columns. Larger batches can be faster but may consume more memory.

Sequence model parameters

Parameter	Default	Possible Values	Description
default_max_sequence_length	10000	Any integer ≥ 1 (e.g., 100, 1000, 10000)	Specifies the maximum sequence length for sequential data generation or processing.
default_end_of_sequence_token	-123456789.98765433	Any numeric token unlikely to appear in real data	A special marker denoting the end of a sequence, ensuring it is not confused with real data values.
default_long_sequence_threshold	10	Any integer ≥ 1 (e.g., 10, 100)	Defines a limit for the length of data sequences used in training, adjusting the longest sequences to the length of the Nth sequence. This helps prevent large sequences from overwhelming memory or computational resources.

Optimization and advanced settings

Parameter	Default	Possible Values	Description
default_ray_memory_optimization	true	true or false (Boolean)	When true, the system explicitly releases idle Ray workers between jobs, reducing memory usage. When false, workers remain alive, reducing overhead for frequent runs.
default_fast_executemany	false	true or false (Boolean)	Enables fast execution for bulk inserts.
default_drop_indexes	false	true or false (Boolean)	Temporarily drops indexes before inserting synthetic data and re-creates them afterward. Often speeds up inserts but re-building indexes can be time-consuming for large tables.

Other

Parameter	Default	Possible Values	Description
default_locale	"en"	Supports various locales, including: "en", "nl", "de", "ja"	Sets the default locale for language-based processing, such as date parsing or random text generation.
default_order_by_nr_columns	[3, 1]	A list of integers (e.g., [3, 1], [1, 2, 3])	Defines the order in which columns are processed or modeled, which can be relevant for preserving data order in AI-powered generation or for certain backends.
default_max_pending_tasks	5	Any positive integer	Defines the number of tables that can be processed in parallel when using ranked scheduling. Increasing the value can improve performance through greater concurrency, but also increases memory usage. It's recommended to adjust this setting gradually, start with the default, monitor system performance, and tune based on available memory and database connection limits to ensure a balanced and stable operation.