How can I make GraphSAGE produce the same node classification results for each run on the same dataset?

[kakaroa]

Describe the bug

*I am using GraphSAGE to train and predict multiple times on the same dataset, but the results are slightly different each time, and the random seeds are all set.

To Reproduce

Steps to reproduce the behavior:

1. Go to '...'
2. Execute '...'
3. See error

Observed behavior

Describe or paste the exact output that demonstrates the bug (for example, the stack trace if it is an exception).

Expected behavior

A clear and concise description of what you expected to happen.

Environment

Operating system: for example: macOS, Ubuntu

Python version: for example: 3.7.2

Package versions: for example: stellargraph==0.8.3, tensorflow==2.0.0

The following Python code can automatically compute the data for this section (for example: run the code in a Jupyter notebook cell, or, on macOS, copy the code and run pbpaste | python - in a terminal):

import subprocess, platform, sys
pkgs = subprocess.Popen(["pip", "freeze"], stdout=subprocess.PIPE, stderr=subprocess.STDOUT).communicate()[0].decode("utf-8")
print("Operating system: `{}`\nPython version:\n```\n{}\n```\nPackage versions:\n<details>\n\n```\n{}\n```\n\n</details>".format(platform.platform(), sys.version, pkgs))

Additional context

Add any other context about the problem here. For example:

Data used: if this occurs with a specific dataset, which dataset and how to acquire it

Screenshots: if applicable, add some screenshots to explain the behaviour