Different results across platforms / python versions

#### Bug

On running the example code for bayesian optimization with `gp_minimize` (https://scikit-optimize.github.io/stable/auto_examples/bayesian-optimization.html) I get inconsistent results with different (stable) python versions and in different machines. It's not machine precision errors, these are dramatic differences.
*Correct Results:* MacOs - all python versions seem to work; Ubuntu - tested python 2.7 (scikit-learn 0.20.4, scikit-optimize 0.7.4) and 3.6 (scikit-learn 0.20.2, scikit-optimize 0.7.4) and it works
*Incorrect Results:* Ubuntu - tested python 3.7.5 and 3.8.0 and 3.8.3 (all with scikit-learn 0.23.1, scikit-optimize 0.7.4) and they give wrong results (consistently different among them) 

I'm assuming this might not be a generic problem in Ubuntu, but for me it is. My fear is if this is some problem deep down in BLAS/LAPACK algebra libraries with some ill-conditioning going on, but I'm hoping someone might have some idea of what is going on. Or maybe something simpler as the sklearn version.

#### Steps to Reproduce

1. Download the example python source code from https://scikit-optimize.github.io/stable/auto_examples/bayesian-optimization.html
2. run it with python 3.7 or 3.8 in a Ubuntu (again, I tested with 3.7.5, 3.8.0 and 3.8.3)

Platform:
Ubuntu 18.04.4 LTS 64-bit 16GB RAM 

#### Expected Results

With python 3.6.9 (same result as in the example page)
```
fun: -1.0146594077693436
    func_vals: array([ 0.03716044,  0.00673852,  0.63515442, -0.16042062,  0.10695907,
       -0.23193728, -0.60259431, -0.04943778, -1.01465941, -0.98480886,
       -0.87449015,  0.18102445, -0.10782771,  0.01197229, -0.80618926])
       models: [GaussianProcessRegressor(alpha=1e-10, copy_X_train=True,
             kernel=1**2 * Matern(length_scale=1, nu=2.5) + WhiteKernel(noise_level=0.01),
             n_restarts_optimizer=2, noise=0.010000000000000002,
             normalize_y=True, optimizer='fmin_l_bfgs_b',
             random_state=822569775), GaussianProcessRegressor(alpha=1e-10, copy_X_train=True,
             kernel=1**2 * Matern(length_scale=1, nu=2.5) + WhiteKernel(noise_level=0.01),
             n_restarts_optimizer=2, noise=0.010000000000000002,
             normalize_y=True, optimizer='fmin_l_bfgs_b',
             random_state=822569775), GaussianProcessRegressor(alpha=1e-10, copy_X_train=True,
             kernel=1**2 * Matern(length_scale=1, nu=2.5) + WhiteKernel(noise_level=0.01),
             n_restarts_optimizer=2, noise=0.010000000000000002,
             normalize_y=True, optimizer='fmin_l_bfgs_b',
             random_state=822569775), GaussianProcessRegressor(alpha=1e-10, copy_X_train=True,
             kernel=1**2 * Matern(length_scale=1, nu=2.5) + WhiteKernel(noise_level=0.01),
             n_restarts_optimizer=2, noise=0.010000000000000002,
             normalize_y=True, optimizer='fmin_l_bfgs_b',
             random_state=822569775), GaussianProcessRegressor(alpha=1e-10, copy_X_train=True,
             kernel=1**2 * Matern(length_scale=1, nu=2.5) + WhiteKernel(noise_level=0.01),
             n_restarts_optimizer=2, noise=0.010000000000000002,
             normalize_y=True, optimizer='fmin_l_bfgs_b',
             random_state=822569775), GaussianProcessRegressor(alpha=1e-10, copy_X_train=True,
             kernel=1**2 * Matern(length_scale=1, nu=2.5) + WhiteKernel(noise_level=0.01),
             n_restarts_optimizer=2, noise=0.010000000000000002,
             normalize_y=True, optimizer='fmin_l_bfgs_b',
             random_state=822569775), GaussianProcessRegressor(alpha=1e-10, copy_X_train=True,
             kernel=1**2 * Matern(length_scale=1, nu=2.5) + WhiteKernel(noise_level=0.01),
             n_restarts_optimizer=2, noise=0.010000000000000002,
             normalize_y=True, optimizer='fmin_l_bfgs_b',
             random_state=822569775), GaussianProcessRegressor(alpha=1e-10, copy_X_train=True,
             kernel=1**2 * Matern(length_scale=1, nu=2.5) + WhiteKernel(noise_level=0.01),
             n_restarts_optimizer=2, noise=0.010000000000000002,
             normalize_y=True, optimizer='fmin_l_bfgs_b',
             random_state=822569775), GaussianProcessRegressor(alpha=1e-10, copy_X_train=True,
             kernel=1**2 * Matern(length_scale=1, nu=2.5) + WhiteKernel(noise_level=0.01),
             n_restarts_optimizer=2, noise=0.010000000000000002,
             normalize_y=True, optimizer='fmin_l_bfgs_b',
             random_state=822569775), GaussianProcessRegressor(alpha=1e-10, copy_X_train=True,
             kernel=1**2 * Matern(length_scale=1, nu=2.5) + WhiteKernel(noise_level=0.01),
             n_restarts_optimizer=2, noise=0.010000000000000002,
             normalize_y=True, optimizer='fmin_l_bfgs_b',
             random_state=822569775), GaussianProcessRegressor(alpha=1e-10, copy_X_train=True,
             kernel=1**2 * Matern(length_scale=1, nu=2.5) + WhiteKernel(noise_level=0.01),
             n_restarts_optimizer=2, noise=0.010000000000000002,
             normalize_y=True, optimizer='fmin_l_bfgs_b',
             random_state=822569775)]
 random_state: <mtrand.RandomState object at 0x7f35be6c6708>
        space: Space([Real(low=-2.0, high=2.0, prior='uniform', transform='normalize')])
        specs: {'args': {'model_queue_size': None, 'n_jobs': 1, 'kappa': 1.96, 'xi': 0.01, 'n_restarts_optimizer': 5, 'n_points': 10000, 'callback': None, 'verbose': False, 'random_state': <mtrand.RandomState object at 0x7f35be6c6708>, 'y0': None, 'x0': None, 'acq_optimizer': 'auto', 'acq_func': 'EI', 'n_random_starts': 5, 'n_calls': 15, 'base_estimator': GaussianProcessRegressor(alpha=1e-10, copy_X_train=True,
             kernel=1**2 * Matern(length_scale=1, nu=2.5),
             n_restarts_optimizer=2, noise=0.010000000000000002,
             normalize_y=True, optimizer='fmin_l_bfgs_b',
             random_state=822569775), 'dimensions': Space([Real(low=-2.0, high=2.0, prior='uniform', transform='normalize')]), 'func': <function f at 0x7f35d8dc3e18>}, 'function': 'base_minimize'}
            x: [-0.35076964213550554]
      x_iters: [[-0.009345334109402526], [1.2713537644662787], [0.4484475787090836], [1.0854396754496047], [1.4426790855107496], [0.9698921802985794], [-0.4464493263345515], [-0.6474638307563569], [-0.35076964213550554], [-0.28714768066245777], [-0.2968537677230516], [-2.0], [2.0], [-1.3149517829595054], [-0.3218160663811962]]
```

![result3 6](https://user-images.githubusercontent.com/17792131/84895714-a6e99700-b09a-11ea-92c0-fd16400b5f8c.png)


#### Actual Results

With python 3.8.3

```
fun: -0.4294555822111941
    func_vals: array([ 0.03716044,  0.00673852,  0.63515442, -0.16042062,  0.10695907,
       -0.11519569, -0.07975108, -0.2971752 , -0.42763291, -0.37296444,
       -0.23640416, -0.10508651, -0.42945558, -0.28040902, -0.25062723])
       models: [GaussianProcessRegressor(kernel=1**2 * Matern(length_scale=1, nu=2.5) + WhiteKernel(noise_level=0.01),
                         n_restarts_optimizer=2, noise=0.010000000000000002,
                         normalize_y=True, random_state=822569775), GaussianProcessRegressor(kernel=1**2 * Matern(length_scale=1, nu=2.5) + WhiteKernel(noise_level=0.01),
                         n_restarts_optimizer=2, noise=0.010000000000000002,
                         normalize_y=True, random_state=822569775), GaussianProcessRegressor(kernel=1**2 * Matern(length_scale=1, nu=2.5) + WhiteKernel(noise_level=0.01),
                         n_restarts_optimizer=2, noise=0.010000000000000002,
                         normalize_y=True, random_state=822569775), GaussianProcessRegressor(kernel=1**2 * Matern(length_scale=1, nu=2.5) + WhiteKernel(noise_level=0.01),
                         n_restarts_optimizer=2, noise=0.010000000000000002,
                         normalize_y=True, random_state=822569775), GaussianProcessRegressor(kernel=1**2 * Matern(length_scale=1, nu=2.5) + WhiteKernel(noise_level=0.01),
                         n_restarts_optimizer=2, noise=0.010000000000000002,
                         normalize_y=True, random_state=822569775), GaussianProcessRegressor(kernel=1**2 * Matern(length_scale=1, nu=2.5) + WhiteKernel(noise_level=0.01),
                         n_restarts_optimizer=2, noise=0.010000000000000002,
                         normalize_y=True, random_state=822569775), GaussianProcessRegressor(kernel=1**2 * Matern(length_scale=1, nu=2.5) + WhiteKernel(noise_level=0.01),
                         n_restarts_optimizer=2, noise=0.010000000000000002,
                         normalize_y=True, random_state=822569775), GaussianProcessRegressor(kernel=1**2 * Matern(length_scale=1, nu=2.5) + WhiteKernel(noise_level=0.01),
                         n_restarts_optimizer=2, noise=0.010000000000000002,
                         normalize_y=True, random_state=822569775), GaussianProcessRegressor(kernel=1**2 * Matern(length_scale=1, nu=2.5) + WhiteKernel(noise_level=0.01),
                         n_restarts_optimizer=2, noise=0.010000000000000002,
                         normalize_y=True, random_state=822569775), GaussianProcessRegressor(kernel=1**2 * Matern(length_scale=1, nu=2.5) + WhiteKernel(noise_level=0.01),
                         n_restarts_optimizer=2, noise=0.010000000000000002,
                         normalize_y=True, random_state=822569775), GaussianProcessRegressor(kernel=1**2 * Matern(length_scale=1, nu=2.5) + WhiteKernel(noise_level=0.01),
                         n_restarts_optimizer=2, noise=0.010000000000000002,
                         normalize_y=True, random_state=822569775)]
 random_state: RandomState(MT19937) at 0x7F53B30AB340
        space: Space([Real(low=-2.0, high=2.0, prior='uniform', transform='normalize')])
        specs: {'args': {'func': <function f at 0x7f53d4635280>, 'dimensions': Space([Real(low=-2.0, high=2.0, prior='uniform', transform='normalize')]), 'base_estimator': GaussianProcessRegressor(kernel=1**2 * Matern(length_scale=1, nu=2.5),
                         n_restarts_optimizer=2, noise=0.010000000000000002,
                         normalize_y=True, random_state=822569775), 'n_calls': 15, 'n_random_starts': 5, 'acq_func': 'EI', 'acq_optimizer': 'auto', 'x0': None, 'y0': None, 'random_state': RandomState(MT19937) at 0x7F53B30AB340, 'verbose': False, 'callback': None, 'n_points': 10000, 'n_restarts_optimizer': 5, 'xi': 0.01, 'kappa': 1.96, 'n_jobs': 1, 'model_queue_size': None}, 'function': 'base_minimize'}
            x: [0.9015772522942358]
      x_iters: [[-0.009345334109402526], [1.2713537644662787], [0.4484475787090836], [1.0854396754496047], [1.4426790855107496], [1.0722894743393931], [1.104047772527582], [1.0332536162210535], [0.9564593955296274], [0.9335830910110636], [0.9616485023360357], [0.9464152096274643], [0.9015772522942358], [0.9186215206368642], [0.8568590965465104]]
```

![result3 8](https://user-images.githubusercontent.com/17792131/84896061-2d05dd80-b09b-11ea-90ba-d7f2778c338d.png)

#### Versions

Result of `import sklearn; sklearn.show_versions()`:
```
System:
    python: 3.8.3 (default, Jun 17 2020, 12:03:07)  [GCC 9.3.0]
executable: /usr/local/bin/python3.8
   machine: Linux-4.15.0-106-generic-x86_64-with-glibc2.27

Python dependencies:
          pip: 20.1.1
   setuptools: 47.3.1
      sklearn: 0.23.1
        numpy: 1.18.5
        scipy: 1.4.1
       Cython: None
       pandas: 1.0.4
   matplotlib: 3.2.1
       joblib: 0.15.1
threadpoolctl: 2.1.0

Built with OpenMP: True
```
Using: `scikit-optimize 0.7.4`

Thank you for your support! I'm very lost about what to do regarding this, as my team work requires a recent >3.6 python version and I get wrong results with those.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Different results across platforms / python versions #916

Bug

Steps to Reproduce

Expected Results

Actual Results

Versions

Metadata

Assignees

Labels

Type

Fields

Projects

Milestone

Relationships

Development

Different results across platforms / python versions #916

Description

Bug

Steps to Reproduce

Expected Results

Actual Results

Versions

Metadata

Metadata

Assignees

Labels

Type

Fields

Projects

Milestone

Relationships

Development

Issue actions