Source code for deephyper.skopt.sampler.hammersly

# -*- coding: utf-8 -*-
""" Inspired by
import numpy as np
from .halton import Halton
from import Space
from .base import InitialPointGenerator
from sklearn.utils import check_random_state

[docs]class Hammersly(InitialPointGenerator): """Creates `Hammersley` sequence samples. The Hammersley set is equivalent to the Halton sequence, except for one dimension is replaced with a regular grid. It is not recommended to generate a Hammersley sequence with more than 10 dimension. For ``dim == 1`` the sequence falls back to Van Der Corput sequence. References ---------- T-T. Wong, W-S. Luk, and P-A. Heng, "Sampling with Hammersley and Halton Points," Journal of Graphics Tools, vol. 2, no. 2, 1997, pp. 9 - 24. Parameters ---------- min_skip : int, default=-1 Minimum skipped seed number. When `min_skip != max_skip` and both are > -1, a random number is picked. max_skip : int, default=-1 Maximum skipped seed number. When `min_skip != max_skip` and both are > -1, a random number is picked. primes : tuple, default=None The (non-)prime base to calculate values along each axis. If empty, growing prime values starting from 2 will be used. """ def __init__(self, min_skip=0, max_skip=0, primes=None): self.primes = primes self.min_skip = min_skip self.max_skip = max_skip
[docs] def generate(self, dimensions, n_samples, random_state=None): """Creates samples from Hammersly set. Parameters ---------- dimensions : list, shape (n_dims,) List of search space dimensions. Each search dimension can be defined either as - a `(lower_bound, upper_bound)` tuple (for `Real` or `Integer` dimensions), - a `(lower_bound, upper_bound, "prior")` tuple (for `Real` dimensions), - as a list of categories (for `Categorical` dimensions), or - an instance of a `Dimension` object (`Real`, `Integer` or `Categorical`). n_samples : int The order of the Hammersley sequence. Defines the number of samples. random_state : int, RandomState instance, or None (default) Set random state to something other than None for reproducible results. Returns ------- np.array, shape=(n_dim, n_samples) Hammersley set. """ rng = check_random_state(random_state) halton = Halton( min_skip=self.min_skip, max_skip=self.max_skip, primes=self.primes ) space = Space(dimensions) n_dim = space.n_dims transformer = space.get_transformer() space.set_transformer("normalize") if n_dim == 1: out = halton.generate(dimensions, n_samples, random_state=rng) else: out = np.empty((n_dim, n_samples), dtype=float) out[: n_dim - 1] = np.array( halton.generate( [ (0.0, 1.0), ] * (n_dim - 1), n_samples, random_state=rng, ) ).T out[n_dim - 1] = np.linspace(0, 1, n_samples + 1)[:-1] out = space.inverse_transform(out.T) space.set_transformer(transformer) return out