Skip to content

Instantly share code, notes, and snippets.

@hvy
Last active January 22, 2021 01:57
Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save hvy/4ef02ee2945fe50718c71953e1d6381d to your computer and use it in GitHub Desktop.
Save hvy/4ef02ee2945fe50718c71953e1d6381d to your computer and use it in GitHub Desktop.
import matplotlib.pyplot as plt
import numpy as np
from scipy.stats import norm
from scipy.special import erfcinv
import optuna
def objective(trial):
# Suggest from U(0, 1) with Optuna.
x = trial.suggest_float("x", 0, 1)
# Inverse transform into normal.
y0 = norm.ppf(x, loc=0, scale=1)
# Inverse transform into lognormal.
y1 = np.exp(-np.sqrt(2) * erfcinv(2 * x))
return y0, y1
if __name__ == "__main__":
n_objectives = 2 # Normal and lognormal.
study = optuna.create_study(
sampler=optuna.samplers.RandomSampler(),
# Could be "maximize". Does not matter for this demonstration.
directions=["minimize"] * n_objectives,
)
study.optimize(objective, n_trials=10000)
fig, axs = plt.subplots(n_objectives)
for i in range(n_objectives):
axs[i].hist(list(t.values[i] for t in study.trials), bins=100)
plt.show()
@hvy
Copy link
Author

hvy commented Jan 22, 2021

Resulting plots.

out

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment