Fix 2 issues

[TimSchim]

Description

HPO with GPTuner is not working with current versions of SciPy

Issue 1:

Starting from SciPy v1.11.0 the shape of argument x0 for scipy.optimize.minimize() should be of shape (n,). The current NNI implementation assumes shape (1, n). Related issues:

• https://github.com/aristoteleo/dynamo-release/issues/578

Issue 2:

scipy.optimize.minimize() returns a scipy.optimize.OptimizeResult object. The current NNI implementation assumes the "fun" item to be an array, when the type should actually be a float. Related issues:

• #4978 (closed but not fixed)

Test Options

• [x] fast test
• [ ] full test - HPO
• [ ] full test - NAS
• [ ] full test - compression

Checklist

• [ ] test case
• [ ] doc

How to test