SimonBlanke /
Gradient-Free-Optimizers
| Total Complexity | 4 |
| Total Lines | 45 |
| Duplicated Lines | 0 % |
| Changes | 0 | ||
| 1 | # Author: Simon Blanke |
||
| 2 | # Email: [email protected] |
||
| 3 | # License: MIT License |
||
| 4 | |||
| 5 | |||
| 6 | import numpy as np |
||
| 7 | from scipy.stats import norm |
||
| 8 | |||
| 9 | |||
| 10 | def normalize(array): |
||
| 11 | num = array - array.min() |
||
| 12 | den = array.max() - array.min() |
||
| 13 | |||
| 14 | if den == 0: |
||
| 15 | return np.random.random_sample(array.shape) |
||
| 16 | else: |
||
| 17 | return ((num / den) + 0) / 1 |
||
| 18 | |||
| 19 | |||
| 20 | class ExpectedImprovement: |
||
| 21 | def __init__(self, surrogate_model, position_l, xi): |
||
| 22 | self.surrogate_model = surrogate_model |
||
| 23 | self.position_l = position_l |
||
| 24 | self.xi = xi |
||
| 25 | |||
| 26 | def calculate(self, X_sample, Y_sample): |
||
| 27 | mu, sigma = self.surrogate_model.predict(self.position_l, return_std=True) |
||
| 28 | # TODO mu_sample = self.surrogate_model.predict(X_sample) |
||
| 29 | mu = mu.reshape(-1, 1) |
||
| 30 | sigma = sigma.reshape(-1, 1) |
||
| 31 | |||
| 32 | # with normalization this is always 1 |
||
| 33 | Y_sample = normalize(np.array(Y_sample)).reshape(-1, 1) |
||
| 34 | |||
| 35 | imp = mu - np.max(Y_sample) - self.xi |
||
| 36 | Z = np.divide(imp, sigma, out=np.zeros_like(sigma), where=sigma != 0) |
||
| 37 | |||
| 38 | exploit = imp * norm.cdf(Z) |
||
| 39 | explore = sigma * norm.pdf(Z) |
||
| 40 | |||
| 41 | aqu_func = exploit + explore |
||
| 42 | aqu_func[sigma == 0.0] = 0.0 |
||
| 43 | |||
| 44 | return aqu_func[:, 0] |
||
| 45 |
