Bug Report with LVGate

[jinningwang]

Hi Hantao,

There is an initialization bug with LVGate. I was developing ESAC1A, and the code was in my test branch exctest.

Below are the diagram and the code snippet.

        self.HVG = HVGate(u1=self.UEL,
                          u2=self.LA_y,
                          info='HVGate for under excitation',
                          )

        self.LVG = LVGate(u1=self.HVG_y,
                          u2=self.OEL,
                          info='LVGate for under excitation',
                          )

The LVGate should output the lower value as design. I set UEL=-999 and OEL=999, so the final output should be the input value LA_y. However, the actual output is 0. It was wired.

Could you please check the related function?

Regards,
Jinning

[cuihantao]

Thanks for reporting. This is a bug. In LVGate, we need to make sure that the Selector is evaluated before updating the equation. Inside the definition of LVGate, move the selector to before the variable, the provide discrete=self.sel to the variable. See if this helps. If not, we need to look at the input to the initialization equation. If you put a breakpoint in the initialization equation for LVG_y in ~/.andes/pycode/ESAC1A, you can check the values of the arguments used for evaluating that equation for initial value. I suspect that the values input to that initialization function is different than the final output. That is, there is a sequence issue.

…

On Sep 29, 2021, 6:32 PM -0500, Jinning Wang ***@***.***>, wrote: Hi Hantao, There is an initialization bug with LVGate. I was developing ESAC1A, and the code was in my test branch exctest. Below are the diagram and the code snippet. self.HVG = HVGate(u1=self.UEL, u2=self.LA_y, info='HVGate for under excitation', ) self.LVG = LVGate(u1=self.HVG_y, u2=self.OEL, info='LVGate for under excitation', ) The LVGate should output the lower value as design. I set UEL=-999 and OEL=999, so the final output should be the input value LA_y. However, the actual output is 0. It was wired. Could you please check the related function? Regards, Jinning — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe.

[jinningwang]

Hi Hantao,

I modified the LVGate definitions as below:

        ...
        self.sl = Selector(self.u1, self.u2, fun=np.minimum.reduce,
                           info='LVGate Selector',
                           )
        self.y = Algeb(info='LVGate output', tex_name='y', discrete=self.sl)

        self.vars = {'sl': self.sl, 'y': self.y}
        ...

However, the bug was still there.

Then I checked the pycode, the LVG_y shows:

def LVG_y_ia(HVG_y, OEL, LVG_sl_s1, LVG_sl_s0):
    return (HVG_y*LVG_sl_s0 + LVG_sl_s1*OEL)

The sequence seems correct, but the bug remains there.

[cuihantao]

Which branch has your latest code?

It fixes the issue after the modification you put.

[cuihantao]

I just pushed my changes to my develop. You can merge it with yours - if you have committed after modifying block.py, you will need manually resolve the conflict.

[cuihantao]

I placed a breakpoint at return (HVG_y*LVG_sl_s0 + LVG_sl_s1*OEL) to inspect the values.

It turns out that both LVG_sl_s0 and LVG_sl_s1 are zeros. Let me keep digging.

[jinningwang]

Sorry, I haven't pushed my code yet. Got a meeting at that time.

Thanks!

Regards,
Jinning

[jinningwang]

The bug was fixed by modifying the LVGate definition as below:

        ...
        self.sl = Selector(self.u1, self.u2, fun=np.minimum.reduce,
                           info='LVGate Selector',
                           )
        self.y = Algeb(info='LVGate output', tex_name='y', discrete=self.sl)

        self.vars = {'sl': self.sl, 'y': self.y}
        ...

Thanks for Hantao's help!