Ryan Curtin

@DamienCassou @egh this seems to be a result of `ofxclient`. I opened an issue there: https://github.com/captin411/ofxclient/issues/78 My workaround is to comment out the `import keyring` and `keyring.get_password()` lines from `ofxclient/cli.py`,...

Should I hold out for actual support, or use that workaround in the finished paper? To me it seems strange to explicitly point out a "None" affiliation.

My memory from years ago is that it can be tricky to make this implementation of the augmented Lagrangian method converge, but, that is only in the context of what...

There is a tradeoff between `penaltyThresholdFactor` and `sigmaUpdateFactor`. Basically, the way the augmented Lagrangian optimizer works is that it adds a "soft" penalty corresponding to the constraints to the objective,...

Hi @vikram2000b, you're more than welcome to. When you have a PR ready, someone will review it when they're able to. :+1:

@dB2510 there is some code in that issue that you can directly compile and use in a program to reproduce the issue. You will probably have to find some datasets...

@gaurav-singh1998 I have yet to see any PR related to the issue, so yes, if you can solve the issue, please do. :+1:

Hey @gaurav-singh1998, the bit of code in `BeginOptimization()` terminates the optimization after a callback returns `false`. We actually just submitted a paper to arXiv today that describes the details; you...

`epochSize` is defined as `function.NumFunctions() * optimizer.BatchSize()`. Thus effectively `epochSize` is equivalent to the number of times the `StepTaken()` callback is called. It does seem to me like `epochs` should...

Do you want me to wait to release the next version of ensmallen until #88 is merged?