TLSQL is a system designed to simplify machine learning workflows on structured tabular data. It translates SQL-like statements into standard SQL queries and structured learning task descriptions, enabling data scientists and engineers to focus on model development instead of writing complex SQL or manually managing datasets.
TLSQL works seamlessly with relational databases, data warehouses, and data lakes, enabling end-to-end table-based ML workflows.
PREDICT VALUE: Specifies the test set.TRAIN WITH: Specifies the training set.VALIDATE WITH: Specifies the validation set.
The TLSQL Workflow
The PREDICT statement specifies the target column for prediction and the task type (classification or regression). This statement defines the test set - the data for which you want to make predictions. The WHERE clause in this statement filters which rows are included in the test set.
PREDICT VALUE(column_selector, TASK_TYPE)
FROM table
[WHERE conditions]CLF: Classification task - predicts discrete categoriesREG: Regression task - predicts continuous values
PREDICT VALUE(users.Age, CLF)
FROM users
WHERE users.Gender='F' The TRAIN statement specifies which columns and tables to use for training data, along with optional filtering conditions. This statement defines the dataset used to train your machine learning model.
TRAIN WITH column_selector
FROM table1, table2, ...
[WHERE conditions]TRAIN WITH (users.*, movies.*, ratings.*)
FROM users, movies, ratings
WHERE users.Gender='M' AND users.userID<3000The VALIDATE statement is used to specify the validation data, and its syntax is similar to that of PREDICT, with only slight differences in the beginning part.
VALIDATE WITH column_selector
FROM table
[WHERE conditions]VALIDATE WITH (users.Age)
FROM users
WHERE users.Gender='M' and users.userID>3000=,!=,<>,<,<=,>,>=
AND,OR,NOT
IN (value1, value2, ...): Check if value is in listBETWEEN value1 AND value2: Range check
Install TLSQL from the tlsql root directory:
cd tlsql
pip install -e .This will install TLSQL in development mode, allowing you to make changes to the code without reinstalling.
import tlsql
# Convert a single TLSQL statement
result = tlsql.convert("PREDICT VALUE(users.Age, CLF) FROM users WHERE users.Gender='F'")
print(result.statement_type) # 'PREDICT'
print(result.target_column) # 'users.Age'
print(result.task_type) # 'CLF'
print(result.sql) # generated SQLCheck out the examples directory for more usage examples:
examples/sql_conversion.py: Basic TLSQL to SQL conversion examplesexamples/bridge_demo.py: BRIDGE model training with TLSQLexamples/tl_workflow.py: TLSQL workflow with three-level logic
@article{chen2026tlsql,
title={TLSQL: Table Learning Structured Query Language},
author={Chen, Feiyang and Zhong, Ken and Zhang, Aoqian and Wang, Zheng and Pan, Li and Li, Jianhua},
journal={arXiv preprint arXiv:2601.14109},
year={2026}
}