Debugging

This page explains how to debug your code without seeing the data

In v1.0.0, when any error occurs with your code at a client site, we specify the input file where the error occured and at which site. For example - data loader for site-2. But beyond that, we don't have a set system to return which exact line caused the error. This is because we want to make sure no private data is returned along with the error message. Since you cannot see the specific error that occured in detail at a client site with your code, debugging with federated learning is difficult.

As of now, to find what the error is, you will have to contact us and we will get back to you as soon as possible with the exact error message and code location. We will make this process simpler in future versions.

Preventing errors

We have created a sample jupyter notebook with dummy data similar to each client's site for you where you can run the same code in a centralized scenario before submitting on our platform. Through this Jupyter Notebook, you can verify the code will work effectively without any errors in federated learning.

In other words, most errors that will appear in a federated scenario will also appear in this jupyter notebook where you can also see the error and resolve it instantaneously.

Debugging tips

1. You can view the communication logs when an error occurs where we mention which specific part of your input caused the error and at which site. For example - data loader for site-2.

2. Feel free to contact us whenever you can't resolve the issue on your own. We understand that error handling is an important part of the process and we will get back to you with results as soon as possible.