Setup script to run t-SNE on server

[vu-minh]

• code should be run in parallel (using python multiprocessing with joblib)
• how to store the embedding result (pickle with joblib.dump)

[vu-minh]

• Find a way to use C++ or multi-core implementation.
• Current issue: when run in parallel, each process calculate pairwise distances matrix, that lead to OutOfMemory error.
• We can try:
• [ ] Reduce number of parallel jobs
• [ ] Using C++ implementation and still run job in parallel
• [ ] Using multi-cores implementation and run in sequential