`FloatingPointError: underflow encountered in exp` in `optimizer.maximize()`

[JaydenTeoh]

Describe the bug A clear and concise description of what the bug is. Using bayesian-optimization==1.4.3 and scipy==1.10.1 results in FloatingPointError: underflow encountered in exp

Here is the code snippet I am testing on:

def black_box_function(x, y):
    return -x ** 2 - (y - 1) ** 2 + 1

pbounds = {'x': (2, 4), 'y': (-3, 3)}

optimizer = BayesianOptimization(f=black_box_function, pbounds=pbounds, verbose=2, random_state=7)

optimizer.maximize(init_points=45, n_iter=20)

This is the error log, seems like when Queue is empty, the self.suggest(util) function gets called which calls the kernel in gaussian process and that's where the exponential underflow occurs:

Traceback (most recent call last):
  File "/common/home/users/j/jxteoh.2023/.conda/envs/dcd/lib/python3.8/site-packages/bayes_opt/bayesian_optimization.py", line 305, in maximize
    x_probe = next(self._queue)
  File "/common/home/users/j/jxteoh.2023/.conda/envs/dcd/lib/python3.8/site-packages/bayes_opt/bayesian_optimization.py", line 27, in __next__
    raise StopIteration("Queue is empty, no more objects to retrieve.")
StopIteration: Queue is empty, no more objects to retrieve.

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/common/home/users/j/jxteoh.2023/.conda/envs/dcd/lib/python3.8/runpy.py", line 194, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/common/home/users/j/jxteoh.2023/.conda/envs/dcd/lib/python3.8/runpy.py", line 87, in _run_code
    exec(code, run_globals)
  File "/common/home/users/j/jxteoh.2023/algos/bw/train.py", line 196, in <module>
    stats = train_runner.run(j)
  File "/common/home/users/j/jxteoh.2023/algos/bw/envs/runners/adversarial_runner.py", line 815, in run
    agent_info = self.agent_rollout(
  File "/common/home/users/j/jxteoh.2023/algos/bw/envs/runners/adversarial_runner.py", line 719, in agent_rollout
    level_sampler.update_with_rollouts(agent.storage)
  File "/common/home/users/j/jxteoh.2023/algos/bw/level_replay/level_sampler.py", line 271, in update_with_rollouts
    self._update_with_rollouts(rollouts, score_function)
  File "/common/home/users/j/jxteoh.2023/algos/bw/level_replay/level_sampler.py", line 651, in _update_with_rollouts
    _, seed_idx = self.update_seed_score(actor_index, seed_t, score, max_score, diversity_t, num_steps)
  File "/common/home/users/j/jxteoh.2023/algos/bw/level_replay/level_sampler.py", line 276, in update_seed_score
    score, seed_idx = self._partial_update_seed_score_buffer(actor_index, seed, score, diversity, num_steps, done=True)
  File "/common/home/users/j/jxteoh.2023/algos/bw/level_replay/level_sampler.py", line 327, in _partial_update_seed_score_buffer
    seed_idx = self._next_buffer_index                  # retrieve the seed_idx with min support
  File "/common/home/users/j/jxteoh.2023/algos/bw/level_replay/level_sampler.py", line 372, in _next_buffer_index
    return self.sample_balanced_weights_w_diversity().argmin()     
  File "/common/home/users/j/jxteoh.2023/algos/bw/level_replay/level_sampler.py", line 906, in sample_balanced_weights_w_diversity
    self.balanced_diversity_coef_support = self.get_balanced_diversity_weightage(np.mean(weights), np.mean(diversity_weights))
  File "/common/home/users/j/jxteoh.2023/algos/bw/level_replay/level_sampler.py", line 196, in get_balanced_diversity_weightage
    self.balanced_weight_optimizer.maximize(init_points=45, n_iter=20)
  File "/common/home/users/j/jxteoh.2023/.conda/envs/dcd/lib/python3.8/site-packages/bayes_opt/bayesian_optimization.py", line 281, in maximize
    x_probe = self.suggest(util)
  File "/common/home/users/j/jxteoh.2023/.conda/envs/dcd/lib/python3.8/site-packages/bayes_opt/bayesian_optimization.py", line 190, in suggest
    self._gp.fit(self._space.params, self._space.target)
  File "/common/home/users/j/jxteoh.2023/.conda/envs/dcd/lib/python3.8/site-packages/sklearn/gaussian_process/_gpr.py", line 304, in fit
    self._constrained_optimization(obj_func, theta_initial, bounds)
  File "/common/home/users/j/jxteoh.2023/.conda/envs/dcd/lib/python3.8/site-packages/sklearn/gaussian_process/_gpr.py", line 622, in _constrained_optimization
    opt_res = scipy.optimize.minimize(
  File "/common/home/users/j/jxteoh.2023/.conda/envs/dcd/lib/python3.8/site-packages/scipy/optimize/_minimize.py", line 696, in minimize
    res = _minimize_lbfgsb(fun, x0, args, jac, bounds,
  File "/common/home/users/j/jxteoh.2023/.conda/envs/dcd/lib/python3.8/site-packages/scipy/optimize/_lbfgsb_py.py", line 305, in _minimize_lbfgsb
    sf = _prepare_scalar_function(fun, x0, jac=jac, args=args, epsilon=eps,
  File "/common/home/users/j/jxteoh.2023/.conda/envs/dcd/lib/python3.8/site-packages/scipy/optimize/_optimize.py", line 332, in _prepare_scalar_function
    sf = ScalarFunction(fun, x0, args, grad, hess,
  File "/common/home/users/j/jxteoh.2023/.conda/envs/dcd/lib/python3.8/site-packages/scipy/optimize/_differentiable_functions.py", line 158, in __init__
    self._update_fun()
  File "/common/home/users/j/jxteoh.2023/.conda/envs/dcd/lib/python3.8/site-packages/scipy/optimize/_differentiable_functions.py", line 251, in _update_fun
    self._update_fun_impl()
  File "/common/home/users/j/jxteoh.2023/.conda/envs/dcd/lib/python3.8/site-packages/scipy/optimize/_differentiable_functions.py", line 155, in update_fun
    self.f = fun_wrapped(self.x)
  File "/common/home/users/j/jxteoh.2023/.conda/envs/dcd/lib/python3.8/site-packages/scipy/optimize/_differentiable_functions.py", line 137, in fun_wrapped
    fx = fun(np.copy(x), *args)
  File "/common/home/users/j/jxteoh.2023/.conda/envs/dcd/lib/python3.8/site-packages/scipy/optimize/_optimize.py", line 76, in __call__
    self._compute_if_needed(x, *args)
  File "/common/home/users/j/jxteoh.2023/.conda/envs/dcd/lib/python3.8/site-packages/scipy/optimize/_optimize.py", line 70, in _compute_if_needed
    fg = self.fun(x, *args)
  File "/common/home/users/j/jxteoh.2023/.conda/envs/dcd/lib/python3.8/site-packages/sklearn/gaussian_process/_gpr.py", line 276, in obj_func
    lml, grad = self.log_marginal_likelihood(
  File "/common/home/users/j/jxteoh.2023/.conda/envs/dcd/lib/python3.8/site-packages/sklearn/gaussian_process/_gpr.py", line 546, in log_marginal_likelihood
    K, K_gradient = kernel(self.X_train_, eval_gradient=True)
  File "/common/home/users/j/jxteoh.2023/.conda/envs/dcd/lib/python3.8/site-packages/sklearn/gaussian_process/kernels.py", line 1704, in __call__
    K = (1.0 + K + K**2 / 3.0) * np.exp(-K)
FloatingPointError: underflow encountered in exp

[till-m]

Running on bayesian-optimization==1.4.3, scikit-learn==1.2.2 and scipy==1.10.0 I do not get this error. What's your scikit-learn version?

[JaydenTeoh]

@till-m I am using scikit-learn==1.2.2 as well. Not sure if it matters but I am running it on a nvidia GPU cluster

[till-m]

Not sure if it matters but I am running it on a nvidia GPU cluster

I guess it could, though this would be above my paygrade. Can you check if the limits of numerical precision are somehow lower on your system then they would usually be?

[till-m]

Closing this for now -- feel free to reopen if the issue persists and you think it is due to this package.