Frequently Asked Questions

How can I use Flow Forecast without Weights and Biases?

We don’t recommend using FF without Wandb as Wandb helps you manage hyper-parameter sweeps and track your model results. Wandb is free to use and you can make runs private. Not using Wandb will NOT make things simpler but rather more complicated. However, if this is what you truly desire then you can just not input the Wandb API key when prompted. You can also specify Wandb as false in your configuration file.

I keep getting the IndexError: index 0 is out of bounds for axis 0 with size 0 when performing inference error?

This means that the selected datetime is likely not in your dataset. Please ensure the selected datetime is in your dataset. For more information on how CSVs should be formatted for inference see this issue.

What model should I start with?

In general we recommend starting with DA-RNN as it is fairly simple and requires less parameters than some of the more complex models. It typically provides strong baseline and the training time for it is not that long. Transformer models may achieve better performance but they often require much better parameter searches and fine-tuning.

Why do you use configuration files?

We use configuration because the facilitate easily trackable and reproducible experiments.

Can I use the underlying Python API?

Yes it is to import models and use functions in the underlying Python.

Where can I find information on how to add models?

See our page on adding models.

I have some other question. Where can I ask it?

You can use Stack Overflow or if it is a specific issue open an issue on our GitHub repo.