I'm playing around with tuning serial gradient descent, and fairly often it seems to fail with this error:
ERROR:pyrogue.Device.SerialGradientDescent.AMCc.FpgaTopLevel.AppTop.AppCore.SysgenCryo.Base[0].CryoChannels.SerialGradientDescent:int too big to convert
Traceback (most recent call last):
File "/usr/local/src/rogue/python/pyrogue/_Process.py", line 100, in _run
self._process()
File "/tmp/fw/rogue_MicrowaveMuxBpEthGen2_v1.1.0.zip/python/CryoDet/DspCoreLib/CryoDetCmbHcd/_SerialGradientDescent.py", line 97, in _process
dx = calcGrad( centerFreqVar, freqErrorVar, etaPhaseVar, freq[channel] + currDf, initialStep, numAverages ) # center difference
File "/tmp/fw/rogue_MicrowaveMuxBpEthGen2_v1.1.0.zip/python/CryoDet/DspCoreLib/CryoDetCmbHcd/_SerialGradientDescent.py", line 28, in calcGrad
centerFreqVar.set( centerFrequencyMHz + df )
File "/usr/local/src/rogue/python/pyrogue/_Variable.py", line 839, in set
varFuncHelper(self._linkedSet,pargs,self._log,self.path)
File "/usr/local/src/rogue/python/pyrogue/_Variable.py", line 868, in varFuncHelper
return func(**args)
File "/tmp/fw/rogue_MicrowaveMuxBpEthGen2_v1.1.0.zip/python/CryoDet/DspCoreLib/CryoDetCmbHcd/_CryoChannel.py", line 94, in <lambda>
linkedSet = lambda var, value, write: var.dependencies[0].set(int(round((value*2**23./freqSpanMHz))), write=write),
File "/usr/local/src/rogue/python/pyrogue/_Variable.py", line 334, in set
raise e
File "/usr/local/src/rogue/python/pyrogue/_Variable.py", line 324, in set
self._block.set(self, value)
File "/usr/local/src/rogue/python/pyrogue/_Block.py", line 362, in set
ba = var._base.toBytes(value)
File "/usr/local/src/rogue/python/pyrogue/_Model.py", line 128, in toBytes
ba = value.to_bytes(byteCount(self.bitSize), self.endianness, signed=True)
OverflowError: int too big to convert
This seems like it's probably due to parameters causing the lms operation to runaway which is reasonable, however it only shows up in the rogue output, and that is saturated with warnings like this so it is very easy to miss. Also this isn't handled at all in the serialGradientDescent command, so it leaves the system in a bad state with the amplitude-scale-arrays zeroed out, and the etaScanInProgress variable left at 1, so it won't run anymore eta scan operations.
Describe the bug
I'm playing around with tuning serial gradient descent, and fairly often it seems to fail with this error:
This seems like it's probably due to parameters causing the lms operation to runaway which is reasonable, however it only shows up in the rogue output, and that is saturated with warnings like this so it is very easy to miss. Also this isn't handled at all in the serialGradientDescent command, so it leaves the system in a bad state with the amplitude-scale-arrays zeroed out, and the
etaScanInProgressvariable left at 1, so it won't run anymore eta scan operations.