Error in multistart with ls optimizers

[Mv77]

Hi! this is Mateo from last year's DSE Summer School. I'm starting to use estimagic and it is just, well, magical! Thank you!

I've run into this small bug.

Bug description

Least squares algorithms appear not to work with multistarts. I think the issue is that some updating step of multistarts is looking for the best criterion value in a list where the least-squares algo has stored the vector of residuals instead.

To Reproduce

import estimagic as em
import numpy as np

def loss(params):

    resid = np.array([params['x'], params['y']])
    out = {
        "root_contributions": resid,
        "contributions": np.square(resid),
        "value": np.sum(np.square(resid)),
    }
    return out

est = em.minimize(
    criterion=loss,
    params={'x': -1.0, 'y': 1.0},
    lower_bounds={'x': -10, 'y': -10},
    upper_bounds={'x': 10, 'y': 10},
    algorithm='scipy_ls_trf',
    multistart=True,
    multistart_options={
        "n_samples": 3,
        "share_optimizations": 1.0
    }
)

print(est)

Expected behavior

I'd expect the above code to perform three starts.

Screenshots/Error messages

Instead, it raises the following error after the first start

Traceback (most recent call last):
  File "/home/mvg/em_fail.py", line 16, in <module>
    est = em.minimize(
  File "/usr/local/Anaconda3/lib/python3.9/site-packages/estimagic/optimization/optimize.py", line 403, in minimize
    return _optimize(
  File "/usr/local/Anaconda3/lib/python3.9/site-packages/estimagic/optimization/optimize.py", line 766, in _optimize
    raw_res = run_multistart_optimization(
  File "/usr/local/Anaconda3/lib/python3.9/site-packages/estimagic/optimization/tiktak.py", line 168, in run_multistart_optimization
    state, is_converged = update_convergence_state(
  File "/usr/local/Anaconda3/lib/python3.9/site-packages/estimagic/optimization/tiktak.py", line 471, in update_convergence_state
    if valid_new_y[best_index] <= best_y:
ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

System

• OS: Ubuntu 18.04

[janosg]

Hi Mateo (@Mv77), thanks a lot for this very nice bug report. It helped me to localize the error quickly.

I will fix this in #411 and add your example as a unit test. We will merge today and make a new release. The new version will be on conda-forge tomorrow or on Monday.

[Mv77]

Awesome; thank you @janosg