Data

Tools

Indexes

Applications

Experiments

Open Source Science

bbgym-rl

https://github.com/ianlienfa/bbgym-rl

Science Score: 44.0%

This score indicates how likely this project is to be science-related based on various indicators:

  • ✓
    CITATION.cff file
    Found CITATION.cff file
  • ✓
    codemeta.json file
    Found codemeta.json file
  • ✓
    .zenodo.json file
    Found .zenodo.json file
  • ○
    DOI references
  • ○
    Academic publication links
  • ○
    Academic email domains
  • ○
    Institutional organization owner
  • ○
    JOSS paper metadata
  • ○
    Scientific vocabulary similarity
    Low similarity (10.1%) to scientific vocabulary
Last synced: 6 months ago · JSON representation ·

Repository

Basic Info
  • Host: GitHub
  • Owner: ianlienfa
  • Language: Jupyter Notebook
  • Default Branch: pure
  • Size: 3.93 MB
Statistics
  • Stars: 0
  • Watchers: 1
  • Forks: 0
  • Open Issues: 0
  • Releases: 0
Created almost 4 years ago · Last pushed about 3 years ago
Metadata Files
Readme Citation

README.md

  • references

    • https://www.rapidtables.com/code/linux/gcc.html
    • https://www.mauriciopoppe.com/notes/computer-science/operating-systems/bin/make/

  • Important!!

    • The job number should be manually defined in util/config.h !
    • This is due to the usage of for performance enhancement
    • The bad news is that bitset for STL do not support dynamically change on its size
    • For the vanilla Lu algorithm, pruneOneRjSumCjLUANDSAL__Theorem1 must be on

  • Simple Usage

    • train the network first: make -j5 net ./net -d directory name
    • prepare binary for inference change to "tester" branch, and compile binary for different labeler move the binaries to current "build" directory and rename them with
    • do valiation after training and pick the model that performs best ./track -mulval 100000 ../saved_model ../case/case-small/validation > case-small.valid
    • copy the best model for testing: cp ../savedmodel/piNet41000.pt ../saved_model/inf/piNet.pt
    • do testing: make -j5 main move the resulting .pt files to /inf directory ./main -d test directory name
    • draw: change the filename for evaluation in eval.ipynb run it

  • Reproducibility

    • The torch and c++ share the same random seed, which is set in util/config.h with name "RANDOM_SEED"
    • The torch::manual_seed() is called in three places, the initialization of weight of both pi and q network and the main function that powers the training.
    • The randomness of c++ comes from the order of choosing the instances in the training process.

  • Usage

    • for inference, use dry_submit = true at submit() to decrease memory use

  • Grid Search

    • make sure the corresponding variable is defined in config.h
    • make sure the corresponding variable is defined in cmakelist.txt
    • rerun the cmake command and assign the value to the variable, ex:
    • cmake -D VHIDDENDIM='128' -D VMAXNUMCNTR='9' -D VLRPI='4e-6' -D VLR_Q='4e-5'
    • recompile the code

  • use "-f" to provide the training set path

  • by default the program requires a /validation subdirectory in the training set directory for validation purpose, user can provide a different directory by "-v"

Owner

  • Login: ianlienfa
  • Kind: user

Citation (CITATION.cff) 

cff-version: 1.2.0
message: "If you use this software, please cite it as below."
authors:
  - family-names: LIN
    given-names: EN-YEN
title: "BBGym"
version: 1.0
date-released: "Not Yet Released"

GitHub Events

Total
Last Year

Dependencies

Dockerfile docker
  • ubuntu 20.04 build