david graff
david graff
The most recent change of spinning each parameterization into it's own separate is the wrong design. Each of the three choices (v1, default, organic) differ **only** in their parameterization **not**...
> > I hate RST solely because of the need for double-tick monospace formatting. > > I `pandoc` my markdown documents to other formats for this very reason I feel...
> > I feel like it must be possible to write our docs in markdown and compile them into RST at deploy-time. Do you know if something like that is...
> 1. Do we use TOML instead of JSON? TOML or YAML. I prefer TOML as a more modern alternative to YAML that has support in the standard library. Writing...
IMO (which I assume is one you share): 1. we should remove the `loc` and `scale` buffers from the `RegressionFFN`, then 2. one of: - tell users that they need...
FWIW, the whole object can be removed and we can switch to just using a native `torch.data.DataLoader`
The second argument sets `requires_grad=False`. While I'm sure there _are_ differences between the two, a non-learnable parameter and a buffer are functionally the same in most cases, which is why...
You're welcome to tackle this, but the mixin class is only used in two spots of the code. It's also not a formal subclass of the sklearn `BaseEstimator`, so I...
should be fixed by #467 or added later. Also linked to #448
a suggestion: we this section in the docs should be called "cookbook"