Jan Vesely

the approach makes sense. Maybe we could do `detach().numpy()` to not consider tensor data type. I'm a bit wary of using a try block though. Hitting `except` is quite expensive,...

Initial triage. I've restricted the [stability_flexibility_pec_fit.py](https://github.com/PrincetonUniversity/PsyNeuLink/blob/fix/stab_flex_pec_bug/Scripts/Debug/stability_flexibility/stability_flexibility_pec_fit.py) to run `max_iterations=1`, `num_trials = 5`, `num_estimates = 1`, and only fit the gain parameter `("gain", controlModule): np.linspace(3.0, 3.0, 2), # Gain` this reduced...

Adding node execution trace doesn't show any difference. all pec invocations: ``` Running node Composition-0 Running node Task Input [I1, I2] Running node Stimulus Input [S1, S2] Running node Cue-Stimulus...

I've modified the [stability_flexibility_pec_fit.py](https://github.com/PrincetonUniversity/PsyNeuLink/blob/fix/stab_flex_pec_bug/Scripts/Debug/stability_flexibility/stability_flexibility_pec_fit.py) script to use the same seed for the initial data generation and each evaluation (using num_estimates=1) ``` +comp.nodes["DDM"].function.parameters.seed.set(1, "foobar-1") +comp.run(inputs, execution_mode=pnl.ExecutionMode.LLVMRun, context="foobar-1") ``` and both produce...

when the number of estimates is increased to 2 I can see two quire different results: ``` Evaluate results: [[([120.], [1.], [2.43]) ([120.], [1.], [0.7 ]) ([120.], [1.], [0.64]) ([120.],...

running on python3.10 produces the following results: = 2561 failed, 3486 passed, 3274 skipped, 1761 xfailed, 2179 warnings, 48 errors in 1559.37s (0:25:59) = All of the errors 2544 failures...

@dependabot recreate

> > lgtm, doesn't seem critical to have the exact numpy error text not change > > thank you. It looks like 1.22 is also hitting some numerical issues which...

Ignoring results of the x86 CI job. It can't uninstall numpy because it was installed in the restricted package set one step earlier.

Compiled and Python Parameters are now synced via shared memory. Explicit synchronization is no longer needed.