Bayesian Training Guide

Prev Next

This guide covers everything you need to train a bayesian_network model and use it to generate synthetic data with the same characteristics as your source data.

Section

What it covers

1. Training

Step-by-step: create the activity and train a model

2. Engine

What a Bayesian network is, how it works and when to choose it

3. Hyperparameters

Activity settings and every bayesian_network parameter, with tuning advice

4. Model

What training produces, how to read the log and when to retrain

5. Column Inclusion / Exclusion Guide

Which columns to train on and which to leave out

6. Generate and Validate

Generating data, checking it against the source and fixing issues

New to Bayesian networks? Follow section 1 first. It links to the other sections wherever you need more detail.


The example dataset

New to machine learning? Load our ML Sample Dataset - Lending into your database first, then work through this guide with the CUSTOMER table on its own, then add the other tables once you're comfortable.

The examples in this guide use the CUSTOMER table from a synthetic consumer-lending dataset. It has 12,000 rows, one per customer, and 11 columns:

Column

Type

What it holds

customer_id

Text

Unique identifier, e.g. C000001

open_date

Date

When the customer relationship started

tenure_months

Whole number

Months since open_date (0–127)

age_band

Category

Six bands, 18-24 to 65+

income_band

Category

Five bands. About 7% of values are missing

employment_cd

Category

FULLTIME, PARTTIME, SELFEMP, CONTRACT, RETIRED

region_cd

Category

Eight regions across Ireland and the UK

channel_cd

Category

BRANCH, ONLINE or BROKER

credit_score

Whole number

473–850

credit_limit

Decimal

1,000–40,000, in steps of 500

dti_ratio

Decimal

Debt-to-income ratio, 0.05–0.65

It mixes categories, whole numbers and decimals, includes missing values, and has strong relationships between columns. For example, customers with higher credit scores have higher credit limits. Those are the things you'll check the synthetic data for.


1. Training

Before you start, you need:

  • A connection to your source database, scanned into the Data Dictionary

  • A Database Definition built from that connection

Create the data activity

1. Create a Machine Learning data activity. Go to Data Activities and create a new activity of type Machine Learning.

2. Add a name and description. Use a name you'll recognise later, such as Customer – Bayesian. In the description, note the table and engine you're training.

3. Select a Database Definition and rule set. Select your Database Definition, then choose New Rule Set. If you already have a training rule set for this data, choose Existing Rule Set instead.

Configure the activity

4. Select the source connection. On the Configuration tab, select the Source Connection the training data is read from.

5. Leave Find Rules off. It adds significant training time, and on a first run you want to confirm that everything completes. See Activity settings.

6. Set Imputation Strategy to Random Sample. It keeps each column's real proportions and works for every column type. See Activity settings.

7. Set Generative Engine to bayesian_network. See 2. Engine for why you'd choose it.

8. Set the engine parameters. Choose a struct_learning_search_method to suit your data. See Search methods for the options. The example uses tree_search, with the other values in Example values.

Choose tables and columns

9. Select your tables. On the Tables tab, check the table(s) you want and click Add Selected. Start with one table. It's the quickest way to confirm everything works before you add related tables.

10. Finish the wizard. Click Next, then Finish. Your new data activity opens.

11. Run Analyze (optional). From the Actions panel, choose Analyze. It confirms the data can be read and suggests post-processing rules, such as the valid range of each numeric column. Relationship suggestions only appear when your rule set includes related tables. See Analyze.

12. Set which columns are active. Open your training rule set and expand your table. Make repeated information inactive, and any identifier that no other table points to. In the example, the CUSTOMER table is trained on its own, so customer_id and open_date are inactive. The table's primary key is still read during training, even when it's inactive, so the platform can keep track of rows. If you train related tables together, keep the key columns that link them active. See 5. Column Inclusion / Exclusion Guide.

13. Check the table-level engine (optional). Each table in the rule set uses the engine from the Configuration tab unless you set a different one for it. To use a different engine or different parameters for one table, edit the table and set them there. A table-level engine overrides the Configuration tab for that table.

Train

14. Train the model. Go back to the dashboard and select Train Model.

15. Check the log. Shortly after training starts, open the job log and find the line beginning Fit internal model. Confirm it says bayesian_network and shows the parameters you set. If it doesn't, stop the job and check steps 7 and 13. See Reading the training log.

When training finishes, the model is attached to the activity and its output files are available to download. See 4. Model.


2. Engine

What a Bayesian network is

A Bayesian network describes your data as a set of columns that depend on each other. It has two parts:

  • The structure is a graph showing which columns depend on which. For example, it might learn that credit_limit depends on income_band.

  • The probabilities say how likely each value of a column is, given the values of the columns it depends on. For example, how likely each credit limit is for a customer in the 100K+ income band.

Training happens in two stages:

  1. Structure learning. The engine searches for the graph that best fits your data. The search method is set by struct_learning_search_method. See Search methods.

  2. Probability learning. For each column, it counts how often each value appears alongside each value of the columns it depends on.

To generate a row, the engine works through the graph. It picks a value for each column at random, using the probabilities for the values already picked. The result is a new row that follows the same patterns as your real data, but isn't a copy of any real row.

How numbers are handled

A Bayesian network treats each distinct value of a column as one possible outcome. A numeric column with 338 different values, such as credit_score in the example, is learned as 338 outcomes.

When generating, the engine picks one of those values, then adds a small random amount controlled by encoder_noise_scale. Whole-number columns, such as tenure_months and credit_score, are converted back to whole numbers when the output is written.

How many distinct values your columns have affects which search method suits your data. See Search methods.

How bayesian_network compares with the other engines

bayesian_network

tvae

ctgan

Type

Probability graph

Neural network (autoencoder)

Neural network (two competing networks)

Best for

Small to medium tables where you want to see the relationships

Small to medium tables with mixed categories and numbers

Imbalanced data with many categories

Training stability

Good

Good

Can be unstable

Speed (platform rating)

8/10

7/10

6/10

Can you see what it learned?

Yes, as a graph of dependencies

No

No

When to choose bayesian_network

Choose bayesian_network when:

  • Your table is mostly categories, codes and bands, with some numbers

  • You want a fast first result, or you're training on CPU

  • You want to see which columns depend on which

Choose something else when:

  • Your table is large, with many columns or many distinct values per column. Try tvae or ctgan.

  • One or more category columns are very imbalanced (a rare value in under 1% of rows) and must still appear. Try ctgan.

Global engine vs table-level engine

You can set the engine in two places:

  • Configuration tab: the default engine for the activity.

  • Table level, in the training rule set: the engine for that table.

The table-level setting wins. Leave it blank to use the Configuration tab engine, or set it to use a different engine for one table. Confirm which engine each table used in the log (see Reading the training log).

When you train more than one table, keep the key columns that link the tables active: each foreign key, and the primary key it points to. They're what lets each child row link to the right parent. The engine from the Configuration tab applies to every table unless a table has its own engine set, so check the log to confirm every table uses bayesian_network.

To learn how the tables relate, the platform copies some of the parent table's columns into each child table before training. They appear in the log with names beginning __cond__, for example __cond__268335_CUSTOMER__credit_score in LOAN_ACCOUNT. Numeric parent columns are grouped into ranges first (BIN_0, BIN_1 and so on).

Correlations: keep only a few per table

A correlation tells the engine to look at a value in the parent table when it creates a child row. For example, "when you create a loan, look at the customer's credit score". Correlations are set on the definition, in the Data Dictionary, not in the training rule set. Every training run on that definition uses the correlations that are switched on there, so check them before you train. Each active correlation adds its parent column to the child table as an extra input, called a conditioning column.

Keep correlations selective. When you run Analyze, at most 2 correlations per table are switched on by default: the 2 with the highest strength. The others are created but left switched off. You can switch more on yourself, but a few strong correlations work better than many. They keep training light, and give the engine enough rows to learn each pattern well.

The engine groups the real rows by every combination of parent values and learns each group. The number of groups multiplies with every conditioning column: N columns with k values each gives k^N groups, but the number of training rows stays the same. With a few strong columns, each group has plenty of rows to learn from.

In the example, two conditioning columns on ACCOUNT_BEHAVIOUR, such as status_cd and principal_amt, give about 20 groups of roughly 1,000 rows each.

When you choose which correlations to keep:

  • Keep the strongest. They capture the real structure in your data.

  • Count parent columns, not correlations. Two correlations from the same parent column add only one conditioning column.

  • Switch off correlations whose parent column is inactive in the training rule set. They add nothing.

  • Drop duplicates that carry the same information. In the example, default_flag and status_cd had identical strengths, because status_cd already has a DEFAULT value.

When you generate related tables, switch on Use Sequential ID for each table's primary key. Every key is then unique, every child row links to a parent row, and each parent has a similar number of children to the source. See Generate data.

CPU vs GPU

bayesian_network doesn't use a GPU, and runs well on CPU. If the log includes CUDA libraries not found, you're on CPU. This makes no difference to bayesian_network.

Do I need a target column?

No. A Bayesian network learns how all the columns relate to each other, so no single column is treated as the answer.


3. Hyperparameters

A hyperparameter is a setting you choose before training. It controls how the model learns, as opposed to what it learns from the data. Changing a hyperparameter means you need to retrain.

Activity settings

These two settings are on the Configuration tab and apply to the whole activity, whichever engine you use.

Find Rules. Searches for mathematical formulas linking your columns, such as one column being another divided by a third. It adds significant time to training. Leave it off for a first run, and turn it on later if you suspect columns are calculated from each other.

Imputation Strategy. Decides how missing values are filled before training.

Option

How it fills a gap

Use for

Random Sample

A value drawn from the column's existing values

Any column. Recommended default

Mode

The most common value

Categories

Least Frequent

The least common value

Rarely useful

Mean

The average

Numbers only

Median

The middle value

Numbers only

Zero

0

Numbers where missing genuinely means zero

Random Sample keeps each column's proportions closest to the real data. Whichever option you choose, the model also records which values were originally missing, in an extra column called <column>__missing_indicator. This lets it learn the pattern of what's missing as well as the values.

bayesian_network parameters

Structure learning

Parameter

Example value

What it does

struct_learning_search_method

tree_search

How the engine searches for the structure. See Search methods

struct_learning_score

bic

How a candidate structure is scored. bic favours simpler structures with fewer links. k2 doesn't penalise extra links, so it can build a larger structure. Only used by search methods that allow several dependencies per column

struct_max_indegree

4

The most columns any one column can depend on. Only used by search methods that allow several dependencies per column

struct_learning_n_iter

1000

The most search steps allowed. Only used by search methods that allow several dependencies per column

Data encoding and generation

Parameter

Example value

What it does

encoder_max_clusters

10

Maximum number of clusters used when encoding numeric columns. Higher values keep finer detail in numeric columns. Lower values make training lighter and faster

encoder_noise_scale

0.1

How much random variation is added to generated numeric values. At 0, generated values repeat the exact values from the source. Higher values add more variety, but drift further from the real values

compress_dataset

Off

Reduces the data before training. Leave off, so the output can be compared with the source

sampling_patience

500

Number of attempts the engine makes to produce a valid row before giving up. Only matters if generation returns fewer rows than you asked for

Search methods

The search method decides how the engine finds which columns depend on which. Each method suits a different kind of data, so the best choice depends on your table.

Method

How it works

Suits data where

Trade-offs

tree_search

Links every column into a tree. Each column depends on at most one other column

Columns have many distinct values, such as detailed numbers, or the row count is modest

Fast. Every column is linked, but each column depends on only one other. Ignores struct_learning_score, struct_max_indegree and struct_learning_n_iter

mmhc

Runs a statistical test on each pair of columns to find which are related, then builds a structure where a column can depend on several others

Most columns are categories, codes, flags or bands with a small number of values each, there are many rows, and real patterns involve several columns together

Can capture patterns across several columns, and leaves unrelated columns unlinked. Needs enough rows for each combination of values to find a relationship

hillclimb

Starts with no links, then repeatedly makes the single change that most improves the score. A column can depend on several others

Most columns are categories, codes or bands, and there are plenty of rows

Can capture patterns across several columns. Uses struct_learning_score, struct_max_indegree and struct_learning_n_iter. Can stop at a good structure rather than the best one

pc

Uses statistical tests on groups of columns to decide which are linked. A column can depend on several others

Mostly categories and many rows, or a single table where numeric relationships matter

Can capture patterns across several columns. Needs enough rows for each combination of values to find a relationship

exhaustive

Scores every possible structure and keeps the best

Tables with very few columns

Finds the best-scoring structure. The number of possible structures grows very quickly with each column, so it's suited to a handful of columns

Search method and score combinations. The score (struct_learning_score) only applies to methods that allow several dependencies per column.

Combination

Best for

What it means for your data

tree_search

A single table with detailed numeric columns, or a quick first result

Every column is linked to one other, so relationships between pairs of columns are kept closely. Patterns that depend on several columns at once come out weaker

mmhc + bic

Mostly categories, codes or bands, with many rows and patterns across several columns

Columns with no clear relationship are left unlinked. Needs enough rows for each combination of values; detailed numeric columns can leave too few, so fewer columns are linked

hillclimb + bic

Mostly categories, codes or bands, including related tables

Builds a compact structure that keeps category proportions and links between categories. Relationships between detailed numeric columns are kept best when the numbers are grouped into bands

hillclimb + k2

Smaller tables where you want the engine to find as many links as it can

Builds a larger structure with more links. Training takes longer and needs more resources, especially with columns that have many distinct values

pc

Mostly categories with many rows, or a single table where relationships between numeric columns matter

Can keep patterns across several columns, including numeric relationships. Needs enough rows for each combination of values

exhaustive

A handful of columns

Finds the best-scoring structure. Time grows very quickly with each extra column, so keep the column count small, for example by training only the columns you need

Choosing for your own data. Compare methods on your own table:

  1. Look at your columns. Note how many distinct values each has, and whether important patterns involve more than two columns together.

  2. Train with one method, generate, and compare the output with the source. See Compare with the source.

  3. Retrain with another method, keeping every other setting the same, and set the same Random seed when you generate.

  4. Compare the two outputs on the relationships and missing values you care about, and on training time.

Methods that allow several dependencies per column, such as hillclimb, find more numeric relationships when detailed numbers are grouped into bands in the source. Raising struct_max_indegree and struct_learning_n_iter also lets them find more relationships.

Example values

The values used for the example table (around 12,000 rows, a mix of category and numeric columns). Use them as a starting point and adjust for your data.

Parameter

Value

Notes

struct_learning_search_method

tree_search

The method used for the example. See Search methods to choose for your data

struct_learning_score

bic

Ignored by tree_search. Used by methods that allow several dependencies per column

struct_max_indegree

4

Ignored by tree_search

struct_learning_n_iter

1000

Ignored by tree_search

encoder_max_clusters

10

Default

encoder_noise_scale

0.1

Default

compress_dataset

Off

Keeps the training data unchanged

sampling_patience

500

Default

For related tables, lower values such as struct_max_indegree 2 and encoder_max_clusters 5 keep training lighter.

Tuning: which setting to change

Change one setting at a time and compare each run against the source.

What you see in the output

Try

Most relationships are missing: the correlation heatmap is almost blank

Check the method in the log and try another. If the method allows several dependencies per column, try raising struct_max_indegree and struct_learning_n_iter, or group detailed numbers into bands. See Search methods

Missing values appear about twice as often as in the source

The model hasn't linked the column to its missing-value indicator. Try another search method

A pattern that depends on several columns is weaker than in the source

With tree_search, each column depends on only one other. Try hillclimb or pc, or apply it with a post-processing rule

Generated numbers repeat the source's exact values too closely

Increase encoder_noise_scale slightly

Generated numbers drift too far from realistic values

Lower encoder_noise_scale

Generation returns fewer rows than you asked for

Increase sampling_patience

Training is slow

Lower encoder_max_clusters or struct_max_indegree, group detailed numbers into bands, train fewer tables together, or try another search method


4. Model

What a trained model is

A trained model is the result of a training job, attached to your data activity. It contains everything the engine learned from your data. It takes a snapshot of the training rule set when the job starts, so changes you make to the rule set afterwards don't affect a model that's already trained.

You use the model by selecting it in a generation rule set (see 6. Generate and Validate).

Each new training run attaches its model to the activity in place of the previous one. Before you generate, check that the activity and the generation rule set point at the model you want.

Output files

When training finishes, open the trained model from the training rule set. The trained model page shows:

  • Configuration files: training.json and metadata.json, which you can download.

  • Location: the folder on your host server where the model files are saved.

  • Engine parameters: the engine settings the model was trained with, including the search method.

metadata.json is the most useful configuration file: it shows the query used to load the training data, so you can confirm which columns were trained. See Training output files for what each file contains.

Reading the training log

Open the job log shortly after training starts and check these lines.

Engine and parameters:

Fit internal model 'bayesian_network' for table '<table>' with params: {...}

Confirm the engine is bayesian_network and the parameters match what you set.

Columns and types:

type_metadata_mapping: {'CUSTOMER': {'integer_cols': [...],
'float_cols': [...], 'categorical_cols': [...]}}

Check that each column is classed as you'd expect. A number stored as text, for example, would appear under categorical_cols. The table's primary key is always listed, even when it's inactive, because the platform uses it to keep track of rows.

Category encoding:

VERIFY [CUSTOMER][channel_cd]: [('BRANCH', 0), ('BROKER', 1), ('ONLINE', 2)]

Each category is listed with the number it's stored as. Check for unexpected values, such as a misspelling or a stray blank. Missing values appear as -1.

Missing values: columns with gaps get an extra <column>__missing_indicator column. In the example, you'd see income_band__missing_indicator.

Structure learning: after Fit internal model, the log can be quiet for a while as the engine searches for the structure. This is normal.

Finished:

Synthcity relational model trained and saved.

Messages you can ignore:

Message

Why it's safe

UserWarning: Clustering metrics expects discrete values...

Raised while tree_search measures how strongly columns are related. The calculation still runs

Error importing TabularGoggle: No module named 'dgl'

An unrelated engine failing to load. It doesn't affect bayesian_network

CUDA libraries not found

No GPU. bayesian_network doesn't use one

SKIPPING Category Mapping for <key>: ratio too high

The primary key has a different value in almost every row, so it isn't mapped to categories

Checking what the model learned

To check the model, generate data and compare it against the source. The quickest check is the correlation heatmap from the generation report: if the source has strong relationships between numeric columns but the heatmap is almost blank, the model hasn't learned them. See Compare with the source.

When to retrain

Situation

Retrain?

You changed an engine parameter

Yes

You made columns active or inactive

Yes

The source data has changed

Yes

The model missed a relationship or distribution

Yes, with adjusted settings

Values break a fixed rule (range, whole numbers, date order)

No, add a post-processing rule

You want more or fewer rows

No, set the row count at generation


5. Column Inclusion / Exclusion Guide

Every active column is something the model has to learn. A Bayesian network compares columns with each other during structure learning, so each extra column adds work, and columns with many distinct values add the most.

Exclude these

Columns calculated from another column. If one column is worked out from another, keep one of them. The model would otherwise have to learn the calculation, and will get it slightly wrong. Example: tenure_months is calculated from open_date, so open_date is inactive. If you need both, recalculate the second one with a post-processing rule.

Columns that give away another column's value. A column set directly from another column's value. For example, a status column that is set to DEFAULT whenever a default flag is 1. It adds nothing, and it can hide the real relationships.

Free text. Notes, comments and descriptions. Each distinct text is treated as a separate category, so the engine can't produce new text and may repeat real text. Mask or generate these separately.

Personal details. Names, emails, phone numbers and addresses. Excluding them keeps real personal details out of the model. Generate them with generator functions.

Unique reference numbers that aren't keys. Any column other than a key with a different value in every row. There's no pattern to learn from it. Generate it with a post-processing rule or generator function instead.

Include these

Columns with the patterns you want to keep. Any column involved in a relationship you care about must be active, on both sides of that relationship. Example: credit_score and credit_limit both stay active, so the model can learn that they rise together.

Columns with missing values. Keep them. The model learns where values are missing as well as what they are, and the pattern often carries information. Example: income_band is missing more often for BROKER customers. Keeping the column keeps that pattern.

Low-cardinality categories. Codes, bands and flags with a manageable number of values are what a Bayesian network learns best.

Keys that link tables. When you train related tables together, keep each foreign key and the primary key it points to active. In the example, that's customer_id in CUSTOMER and LOAN_ACCOUNT, and account_id in LOAN_ACCOUNT and ACCOUNT_BEHAVIOUR. They're what lets each child row link to the right parent.

Think before including these

Column type

Guidance

Dates

The platform splits a date column into parts (year, month, day, hour and so on) before training, so the engine can learn patterns such as busier months or weekdays. This happens by default for date columns with at least 100 distinct values. A date column with fewer distinct values is treated as a category. If a date can be recalculated from a number of days or months, you can train the number instead and recalculate the date with a post-processing rule

Two columns that say nearly the same thing

Keep one, and recalculate the other with a post-processing rule

Categories with hundreds of values

Each value adds training time. Group rare values into an OTHER category in the source, or exclude the column

Numbers with hundreds of distinct values

Affect which search method suits your data. Grouping them into bands helps methods that allow several dependencies per column. See Search methods

Columns filled by a generator function

Exclude from training. They're filled at generation time

Quick checklist

Column

Active?

Why

Primary key that another table points to (multi-table)

✓

Keeps parent–child links

Foreign key (multi-table)

✓

Keeps parent–child links

Unique reference number that isn't a key

✗

No pattern to learn

Category, code or band

✓

What a Bayesian network learns best

Number

✓

Unless calculated from another column

Date

✓ / ✗

Train it, or train a number and recalculate the date

Calculated from another column

✗

Recalculate with a post-processing rule

Gives away another column's value

✗

Adds nothing, hides real relationships

Free text

✗

Can't be generated; may repeat real text

Personal details

✗

Use generator functions

Column with missing values

✓

The pattern of what's missing is information


6. Generate and Validate

Generate data

1. Create a generation rule set. Select your trained model in its configuration.

2. Review the rule set. The table and column settings are copied from training. Leave post-processing rules out for now, so you can see what the model produces on its own.

Edit each table in the generation rule set:

  • Default Count. Set the number of rows. If you trained related tables, set it on the parent table only and leave it empty for child tables. Their row counts come from the parent, following the number of children per parent in the source.

  • Use Sequential ID. Switch this on for each table's primary key, select the key in Column, and enter a Start ID, such as 1. Each row then gets a unique key, numbered in order. Don't switch it on for foreign keys: they're filled from the parent rows, which keeps the links between tables. Use Use Composite Sequential ID instead for a primary key made of more than one column.

Sequential IDs are numbers. If your keys have a format, such as C000001, add a custom_python rule to apply it. See Improve the output.

3. Generate a small sample. Select Generate and enter a small row count, such as 500. Check that the job completes, the right columns are filled, and data types and formats are what you expect.

Large outputs. You can generate millions of rows. Check a small sample first, then scale up.

4. Generate a full run. Generate again with the same number of rows as your source table (12,000 in the example). Matching the row count makes the comparison fair. The job log shows the row count and where the output file was saved. See Generation reports.

The Generate Data dialog has two further options:

  • Random seed. Enter a whole number to get the same output every time you generate with the same settings. This makes runs easy to compare: any difference comes from the model or the rules, not from chance. Leave it blank for different output each run.

  • Generate report charts. Saves charts next to the output file, with an index.html page to view them all. See Generation reports.

The generation log contains many lines like Probability values don't exactly sum to 1. Differ by: -2.22e-16. Adjusting values. These are tiny rounding corrections and are safe to ignore.

Each generation from the same model overwrites the previous output file. Download the output before you generate again if you want to keep it.

Generation reports

Each generation saves reports next to the output files. They compare the generated data with the source, so they're the quickest place to start checking it.

Report

What it shows

generated_executive_summary.json

An overall score and grade, a score for each table, the weakest columns and any risks found

generated_evaluation_report.json

How closely each column matches the source, how well relationships between columns are kept, privacy checks, and whether each check passed

generated_integrity_report.json

Duplicate or missing primary keys, child rows whose parent is missing, values outside the source range and unknown categories

generated_coverage_report.json

How much of each column's range and categories the generated data covers

generated_interpretation_report.md

How to read the reports and charts, and a plain-English verdict

Start with the executive summary. Then check the integrity report for key problems, and the relationship check (correlation_mean_abs_diff) in the evaluation report.

Compare with the source

Download the generated data from your host server and compare it with the source data. The reports cover most of the checks below.

Compare on:

  • Distributions. Categories should appear in similar proportions. Numeric columns should have similar ranges, averages and shape.

  • Rare values. Rare categories and extreme values should still appear, in roughly the same amounts.

  • Relationships. Columns that move together in the source should still move together. In the example, credit limit should still rise with credit score.

  • Missing values. Each column should have a similar share of missing values. About 7% of income_band values should be empty.

  • Valid values. Every value should be possible: known categories, sensible ranges, and business rules that still hold.

  • Links between tables. If you generated more than one table, every child row should point to a real parent row, and each parent should have a similar number of children.

  • Similarity to real records. No synthetic row should be a copy of a real one.

Leave inactive columns and generator-filled columns out of the comparison.

What to expect. A close match, not an exact one. Proportions and missing-value rates are usually within a few percent of the source, and relationships between columns can come back a little softer. A small difference across many columns is normal. If one column or relationship is far off, improve the output as described below.

The model learns patterns, not rules. Formats and business rules, such as credit_limit always being a multiple of 500, or a loan never being larger than the customer's credit limit, are applied with post-processing rules. See Improve the output.

Improve the output

There are two ways to improve the output. You can use either or both.

Retrain the model if the model missed something, such as a weak relationship. See Tuning: which setting to change. Also check that the relevant columns were active.

Add post-processing rules to fix things the model can't be expected to learn, such as fixed steps, key formats or calculated columns. Rules are applied after the model has run, so you can add them and generate again without retraining. Rules run in the order they're listed, so put a rule that recalculates a value after any rules that fix its inputs.

Problem in the output

Rule

Values outside the real range

range_clipping

Values that should be in fixed steps, such as multiples of 500

custom_python

Values that should be calculated from other columns

product_rule, custom_python, sql_alignment

Keys that need a format, such as a prefix

custom_python

Dates in the wrong order

date_sequence, sequential_event_gate

Child rows pointing to parents that don't exist

foreign_key_sync, relational_alignment

Synthetic rows too similar to real ones

privacy_similarity_score

Duplicate primary keys, or keys that should be numbered in order

Use Sequential ID on the table in the generation rule set

An identifier column you excluded from training

custom_python, or Use Sequential ID on the table

For the example, these rules give an output that matches the source format:

#

Rule

Settings

1

custom_python

df['credit_limit'] = (df['credit_limit'] / 500).round() * 500

2

custom_python

df['customer_id'] = 'C' + (df.index + 100001).astype(str)

3

custom_python

m = 2026 * 12 + 7 - df['tenure_months'].round().astype('int64'); df['open_date'] = (m // 12).astype(str) + '-' + (m % 12 + 1).astype(str).str.zfill(2) + '-01'

Rule 2 creates new identifiers from C100001 upwards, so none match a real customer. Rule 3 recalculates open_date from tenure_months, counting back from the dataset's snapshot date (August 2026). It's accurate to the month, so every date falls on the 1st.

See Post-processing rules reference for every rule and its settings.

Repeat until you're happy

Retrain and regenerate until the output meets your needs. Change one thing at a time and compare each run against the source, so you can see which change made the difference.