Quick & dirty solution to use an external mounted hard drive under Ubuntu to transparently store both Tensorflow & Keras datasets, in order to preserve fast SSD space.

tensorflow-keras-setup-dataset-paths.sh

#!/bin/bash

# Mounted Hard drive path
drive_path = /media/data/dl

# Create main dir
mkdir $drive_path

# Create folders
mkdir $drive_path/keras_datasets 
# Or "mv ~/tensorflow_datasets $drive_path/keras_datasets" if already exists
mkdir $drive_path/tensorflow_datasets 
# Or "mv ~/.keras/datasets $drive_path/tensorflow_datasets" if already exists

# Create symlinks
ln -s $drive_path/keras_datasets ~/.keras/datasets
ln -s $drive_path/tensorflow_datasets ~/tensorflow_datasets