RAM overflow while initializing graph

I have 294 Gb binary .dat file and one machine having 126 Gb of RAM.
When I execute a command
```
python -m tf_euler \
--data_dir euler \
--max_id 177416798 \
--model graphsage --mode train
```
memory overflow occurs. What can I do to process this file in chunks on my machine? Can HDFS help me to do it locally?

Start by reproducing the shown `python -m tf_euler` training command with the 294 GB `.dat` input and 126 GB of RAM. Investigate where graph initialization exhausts memory and whether chunked or HDFS-backed processing is supported; done means the dataset can be processed without memory overflow on the reported machine.

Written by the indexing model from the issue text.