generatezdt.py

"""Create synthetic datasets for surrogate training."""
# %%
import pickle as pk
from os import getcwd

import numpy as np
import pandas as pd
from optproblems import wfg, cec2005, cec2007, dtlz, zdt
from pyDOE import lhs
from scipy.stats.distributions import norm

# %% ZDT
problems = {
    "ZDT1": zdt.ZDT1,
    "ZDT2": zdt.ZDT2,
    "ZDT3": zdt.ZDT3,
    "ZDT4": zdt.ZDT4,
    "ZDT6": zdt.ZDT6,
}
folder = getcwd() + "/zdtdatasetscsv"
# %%
num_var = {
    "ZDT1": 30,
    "ZDT2": 30,
    "ZDT3": 30,
    "ZDT4": 10,
    "ZDT6": 10,
}
num_obj = 2
num_samples = [
    100,
    150,
    200,
    250,
    300,
    350,
    400,
    500,
    600,
    700,
    800,
    900,
    1000,
    1200,
    1500,
    2000,
]
distribution = ["uniform", "normal"]
noise = [False]
missing_data = [False]
# %% Parameters for normal distribution and noise
means = [0.4, 0.6]
stdvs = [0.25, 0.2]
noise_mean = 0
noise_std = 0.1
miss_fraction = 0.1
# %% Creation of datasets
for problem in problems:
    print(problems[problem])
    objective = problems[problem]()
    for num_sample in num_samples:
        var_names = ["x{0}".format(x) for x in range(num_var[problem])]
        obj_names = ["f1", "f2"]
        filename = (
            folder + "/" + problem + "_" + str(num_var[problem]) + "_" + str(num_sample)
        )
        filename_noisy = (
            folder + "noisy/" + problem + "_" + str(num_var[problem]) + "_" + str(num_sample)
        )
        var = lhs(num_var[problem], num_sample)
        obj = [objective(x) for x in var]
        data_uniform = np.hstack((var, obj))
        data_noisy = data_uniform
        data_uniform = pd.DataFrame(data_uniform, columns=var_names + obj_names)
        # pk.dump(data_uniform, open((filename + "uniform.p"), "wb"))
        data_uniform.to_csv(filename + "_uniform.csv", index=False)
        for i in range(num_var[problem] + num_obj):
            data_noisy[:, i] = data_noisy[:, i] + np.random.normal(
                noise_mean, noise_std, num_sample
            )
        data_noisy = pd.DataFrame(data_noisy, columns=var_names + obj_names)
        # pk.dump(data_noisy, open((filename + "uniform_noisy.p"), "wb"))
        data_noisy.to_csv(filename_noisy + "_uniform_noisy.csv", index=False)
        var_norm = var
        for i in range(2):
            var_norm[:, i] = norm(loc=means[i], scale=stdvs[i]).ppf(var[:, i])
        var_norm[var_norm > 1] = 1
        var_norm[var_norm < 0] = 0
        obj_norm = [objective(x) for x in var_norm]
        data_normalized = np.hstack((var, obj))
        data_noisy = data_normalized
        data_normalized = pd.DataFrame(
            data_normalized, columns=var_names + obj_names
        )
        # pk.dump(data_normalized, open((filename + "normal.p"), "wb"))
        data_normalized.to_csv(filename + "_normal.csv", index=False)
        for i in range(num_var[problem] + num_obj):
            data_noisy[:, i] = data_noisy[:, i] + np.random.normal(
                noise_mean, noise_std, num_sample
            )
        data_noisy = pd.DataFrame(data_noisy, columns=var_names + obj_names)
        # pk.dump(data_noisy, open((filename + "normal_noisy.p"), "wb"))
        data_noisy.to_csv(filename_noisy + "_normal_noisy.csv", index=False)