Andrei Zhabinski
Andrei Zhabinski
The [`update!()`](https://github.com/dfdx/Avalon.jl/blob/main/src/optim.jl#L37-L48) function takes an optional argument `ignore` - a set of field paths that should not be updated. A field path is a tuple of symbols representing path to...
Great, pretraining is a very important use case for Avalon, so we will definitely have a more concise syntax for freezing parameters, but exact API will arrive later, perhaps shortly...
I think momentum for other parameters didn't make it any better for my use case at that time, so I just decided to not include unchecked feature. However, if in...
Totally agree. I would even expand it to a detailed documentation that includes: 1. Brief overview of theory behind RBMs, so that people could get into it quickly. 2. Usage...
Sounds good to me. Do you have an idea how it should look like? I thought about including formulas as a part of mini tutorial on restricted Boltzmann machines, but...
Indeed, here's MWE: ```julia using Espresso ExGraph() ``` Diving deeper, the issue is due to invocation of `get_caller_module()` (commenting it out fixes the error). This function deals with Julia's internals,...
Yeah, looks like `while true` breaks this function. Intuitively, we need to return `nothing` in this case because there's no block designating the condition, but we need to test it...
Busy right now, but seems to be the same as dfdx/Yota.jl#84
Yes, it's the same issue as the one I linked. Unfortunately, I don't have an immediate fix for it - IRTools adds some magic that I don't know how to...
Agree. My current plan is to take a look at various alternatives to IRTools, which is bottleneck in this issue and to be deprecated anyway, and see what would be...