Default solver in parametric learner

[MarcosCarreira]

Simulating a process with negative elements in the adjacency matrix:

kernel_adjacency = np.array([[[0.1], [-0.05]], [[-0.02], [0.08]]]) kernel_decays = np.array([1.]) baseline = np.array([1., 0.9]) run_time = 10000 hawkes = SimuHawkesSumExpKernels( adjacency=kernel_adjacency, decays=kernel_decays, baseline=baseline, end_time=run_time, seed=1398) n_realizations = 100 multi = SimuHawkesMulti(hawkes, n_simulations=n_realizations) multi.simulate() multi_timestamps = multi.timestamps

And learning:

hawkes_param_learner = HawkesSumExpKern(kernel_decays) hawkes_param_learner.fit(multi_timestamps) baseline_param_learner = hawkes_param_learner.baseline adjacency_param_learner = hawkes_param_learner.adjacency

With the default solver does not retrieve the negative elements; only the option solver='bfgs' retrieves the correct adjacencies. Why not make it the default solver?