Ask function outputting repeated points

[boyarkin]

Hello!
I'm trying to solve black-box optimization task and that task is specific: domain of objective function is quite small, so when suggestion point x is not in domain, f(x) returns nan or any big value like 2.72489e+19 or larger.

My code:

import nevergrad as ng

def obj_function(x):
    return black_box.get_function_value(x.tolist())

def solve():
    initial_point = black_box.get_initial_point()

    # init optimizer
    param = ng.p.Array(shape=(parametrization_val,)).set_standardized_data(initial_point)
    ng_optimizer = ng.optimizers.registry['BFGS'](parametrization=param, budget=1000)

    for _ in range(ng_optimizer.budget):
        x = ng_optimizer.ask()
        obj_v = obj_function(*x.args, **x.kwargs)
        ng_optimizer.tell(x, obj_v)

    recommendation = ng_optimizer.provide_recommendation()
    return recommendation.value

The main problem, that nevergrad optimizer suggest the same point many times until the budget ends. So my log looks like:

Suggestion: 
2.34383, 2.23239, -1.48685, 0.742839, -1.16699, 1.04315, -1.00391, 0.0158264, -0.808701, -0.876145, 0.376727, 0.323922 ...
Cost (R): 2.72489e+19
______________________________
Suggestion: 
2.34383, 2.23239, -1.48685, 0.742839, -1.16699, 1.04315, -1.00391, 0.0158264, -0.808701, -0.876145, 0.376727, 0.323922 ...
Cost (R): 2.72489e+19

.... ~150 repetitions of information above...
______________________________
Suggestion: 
0.784922, -0.0693755, -0.0383782, 0.652343, -0.0737021, 0.119224, 0.627094, 0.904481, 0.53831, -0.528335, 0.29518 ...
Cost (R): 2.72489e+19

.... ~150 repetitions of information above...

and so on.

So it can't fins solution and recommendation just return initial point as the best result. So I can't understand why nevergrad suggest repeated points and how to avoid this?
Thank you