BoLib

This is a library to work with bayesian optimization (bo). It is designed to use bo without transforming the data manually.

Here is an example where the computer guesses your pet and your pet's age. If you don't have a pet, just pretend you have a 7 years old cat. This example is also in the examples folder as simple.py.

import sys
import os
import random

root_path = os.path.abspath('..')
sys.path.append(root_path)


from bolib.Dimension import DiscreteDimension, NumericDimension
from bolib.Bo import Bo
from bolib.ComputeSpace import ComputeSpace

petelements = ["cat", "dog","bird", "ape", "rabbit"]
pet     = DiscreteDimension(elements=petelements ,name = "pet")
age     = NumericDimension(min=1, max=10, name="age")

RANK_MAX = 5
RANK_MIN = 0

ranking = [ NumericDimension(min=RANK_MIN, max=RANK_MAX, name="Rating") ]
dimensions = [pet, age]
compSpace = ComputeSpace(x = dimensions, y = ranking)

optimizer = Bo(compSpace)
nguess = [random.random() for _ in dimensions]
print("I will guess your pet, for this I will tell you the animal, and an age.")
print(f"You rate how close I am on a scale from {RANK_MIN} to {RANK_MAX}.")
print("A good metric is, if the animal is correct, give 2 points, then")
print("add 1 point if the year is close,")
print("add 2 points if my guess 1 year off,")
print("add 3 points if my year is correct.")
input("ready?")

g = compSpace.denormalize(nguess)

while True:
  guess = compSpace.denormalize(nguess)
  guess[1] = age.new(round(guess[1].value)) 
  print(guess.to_dataframe()[['name', "value"]])
        
  try:
    feedback = float(input(f"Enter rating [{RANK_MIN} = fits not at all, {RANK_MAX} = fits very well]: "))        
  except:
    break
  if feedback == 5:
    print(f"You have a {guess[0].value} that is {guess[1].value} years old.")
    break
  compSpace.add_value(guess, feedback)
          
  nguess = optimizer.infer()

philipphock/bolib

BoLib