Skip to content

Latest commit

 

History

History
78 lines (65 loc) · 2.8 KB

changelog.md

File metadata and controls

78 lines (65 loc) · 2.8 KB

Changelog

23.11.2022

  • Updates pfp
  • NOTE! File Paths in pfp_runner are hardcoded. Error will occure. Change manually
  • Top above will be fixed in the next days

22.11.2022

  • Implemented Simulation Driven Runner
  • Old Runner is now Algorithm Driven Runner
  • Implemented Use Case pfp (Maurice BA)
  • Implemented inequality parameter constraint (sum(w*p) >= c)
  • NOTE: TOP 1-3 are not finished

07.10.2022

  • rounding bug fixed in mrp
  • reduced stochastic propability in mrp_sim (got to eval if promising)

05.10.2022

  • wording change: batch_* -> trial_*
  • small change cma-es. Skipps initial now and generates {trial_size} arms directly
  • turbo state updates correctly now
  • reduced tolerance of turbo to force exploration
  • increased penalty cost rate from 0.25 to 0.33 (reason: at 50D best value found at sl of 0.78)

04.10.2022

  • set surrogate model für turbo and gpei as experiment parameter (gsheet)
  • identify best trial fixed
  • Bug at CMA-ES fixed. Now finds minimum instead of maximum

30.09.2022

  • Brute Force with logs
  • log every new best point

28.09.2022

  • MRP: reduced propability of delay and reduction (stochastic)
  • MRP: bug fix at releases. Now it works as it should :)
  • MRP: added function to get data of every class instance (csv)

23.09.2022

  • wording: configs -> experiment
  • fixed solver and simulation
  • configs folder will be created if not exists
  • further minor changes

21.09.2022

  • wording changes at mrp_runner (inventory -> stock, demand -> orders)
  • Implemented new BOMs etc. in Gsheet

16.09.2022

  • Sobol Runner
  • Brute Force Runner
  • MRP Solver OOP

15.09.2022

  • MRP Simulation OOP

14.09.2022 (Phil)

  • acquisition value implementiert
  • feature importance hinzugefügt

13.09.2022 (Phil)

  • Algorithmus Runner Klasse erstellt, aktuelle Runner erben davon
  • MRP stochastic_method via Sheets konfigurierbar
  • MRP method discrete -> "Tail" hinzugefügt
  • changelog file hinzugefügt (offensichtlich)
  • laden der Sheets aus der main heraus, dazu als sysarg "load" eintragen
    • Habe Funktion geschrieben zum checken der sysargs
    • Genereller Aufbau der sysargs: main.py [experiment_id] [opt:replication] [opt:"load"]
    • aber auch möglich main.py [opt:"load"] [experiment_id] [opt:replication]
  • wenn num_init/n_init -1 dann num_init = 2*dim (definiert im Algorithmus Runner)
  • cmaes implementiert (WIP und hartes theoretisches Defizit meinerseits, muss lernen was die Parametrisierungen bedeuten und diese ggf. implementieren...)
  • Update Gsheet Felder u_stochastic_method, a_sigma0 (Hyperparameter cma-es)
  • algo runner Funktion "get_technical_specs" für experiment json Informationen. sm and acqf im SubRunner definieren.
  • *_stepwise files gelöscht
  • requirements.txt upgedated (cma)

general note

  • Changes einfach chronologisch (neueste oben) eintragen mit Datum und Name (siehe oben)