graph embedding

[ruochen99]

Add back the graph embedding script that was removed from the last PR. To generate workload embeddings for the tasks, run

python3 graph_embedding.py

and in mlp_model.py or xgb_model.py, replace

tmp_task_embedding = get_workload_embedding(task.workload_key)

with

tmp_task_embeddings = pickle.load(open("task_embeddings.pkl", 'rb'))
tmp_task_embedding = tmp_task_embeddings[json.loads(task.workload_key)[0]]

and modify the dimensions accordingly. Note that I include two different embedding models in this script, but they are not super useful for improving the model performance. Including here to provide some food for thought.