Datalayers + TimesFM 2.5 Example

This example repository accompanies the Datalayers + TimesFM 2.5 hands-on blog post. It demonstrates a complete minimal workflow:

1. Create a time-series table in Datalayers.
2. Load CSV data into the table.
3. Read data through FlightSQL.
4. Run forecasting with TimesFM 2.5.
5. Export forecast CSV and visualization.

Environment Requirements

• Python >= 3.10
• A running Datalayers instance (with FlightSQL enabled). We recommend running Datalayers on docker.
• Network access from this project to your Datalayers service

Required Python Dependencies

Install the required libraries in your environment:

pip install \
  pandas \
  matplotlib \
  flightsql-dbapi \
  kagglehub

Notes:

• It's required to install timesfm2.5 directly from source code.

Quick Start

1. Create database/table:

# Run sql in scripts/init_db.sql against your Datalayers SQL endpoint
# We provide `dlsql` to connect Datalayers server with following commands:
dlsql -h localhost -P 8360 -u admin -p public
# After entering the cli, you can create database using `CREATE DATABASE <dbname>`, and `exit` to exit.
# If you are not changed to the target database, use `use <dbname>` to enter your database.

2. Load CSV data:

python scripts/load_csv_via_flightsql.py \
  --csv /path/to/your/dataset.csv \
  --table <tablename> \
  --host localhost \
  --port 8360 \
  --user admin \
  --password public \
  --db <dbname>

3. Run forecasting and visualization:

python example/visualization.py

4. Check outputs:

• outputs/timesfm25_forecast.csv
• outputs/timesfm25_visualization.png

This setup is intended for experimentation and learning.