Flow Forecast Training Dockerfiles

This repository contains Dockerfiles needed to run Flow Forecast as a distributed Weights and Biases hyper-parameter sweep.

1. Make a sweep YAML file of the hyperparameters you want to search for on your FF model (example in the example_sweep directory). The program in the sweep file should be run_flow.py.
2. Get your WANDB_SWEEP_ID. For this you need to run wandb sweep --project sweeps_demo config.yaml
3. Make an environment file In your environment file there should be the following items:

• WANDB_API_KEY
• WANDB_SWEEP_ID (this should be the full project/sweep)
• BASIC_CONFIG_PATH
• ENVIRONMENT_GCP (option)

1. Run docker run --env-file .env_file aistream1/flow_sweep

Your base config path file should have the JSON file with the default sweep values.