Move 5 tutorials from fluxml.github.io #2125
Conversation
|
Once the build has completed, you can preview any updated documentation at this URL: https://fluxml.ai/Flux.jl/previews/PR2125/ in ~20 minutes |
This comment was marked as off-topic.
This comment was marked as off-topic.
| "Linear Regression" => "tutorials/linear_regression.md", | ||
| "Julia & Flux: 60 Minute Blitz" => "tutorials/2020-09-15-deep-learning-flux.md", |
There was a problem hiding this comment.
What do you think about adding a 🚧 or some admonition in the tutorial itself so that users know this might be out of date? Or is the hope to have those sorted out before these hit a stable release?
There was a problem hiding this comment.
Haven't seen one of those in a while! I don't think these are terribly out of date in fact, nothing looks like it clearly won't run.
There was a problem hiding this comment.
Found a few places they needed to import params
There was a problem hiding this comment.
Re updating, some of these link to the model zoo:
https://github.com/FluxML/model-zoo/blob/master/vision/mlp_mnist/mlp_mnist.jl
https://github.com/FluxML/model-zoo/blob/master/vision/conv_mnist/conv_mnist.jl
https://github.com/FluxML/model-zoo/blob/master/vision/dcgan_mnist/dcgan_mnist.jl
IDK what the plan ought to be for keeping things in sync. These all look like they qualify Flux.params
There was a problem hiding this comment.
If we make them literate files, then copying over the model zoo files and deleting them from that repository would work. In the meantime, things can stay as they are.
| save("./output.gif", gif_mat) | ||
| ``` | ||
|
|
||
|  |
There was a problem hiding this comment.
We might as well move the assets over so that everything is in one place. @ref links can be cleaned up later.
There was a problem hiding this comment.
I tried a bit locally, and gave up... IDK why I couldn't get relative links to work. Note that they are broken in the current website version too:
https://fluxml.ai/tutorialposts/2021-10-08-dcgan-mnist/#output
There was a problem hiding this comment.
Ok, no use fighting them for a quick PR. I suspect the website links were also a victim of the Franklin migration.
There was a problem hiding this comment.
Thanks, @mcabbott!!
I'll look into using Literate.jl for the tutorials.
| * `augment` augments the data by adding gaussian random noise to our image to make it more robust: | ||
|
|
||
| ```julia | ||
| augment(x) = x .+ gpu(0.1f0*randn(eltype(x), size(x))) | ||
| ``` | ||
|
|
||
|
|
||
| * `anynan` checks whether any element of the params is NaN or not: | ||
|
|
||
| ```julia | ||
| anynan(x) = any(y -> any(isnan, y), x) | ||
| ``` | ||
|
|
||
| * `accuracy` computes the accuracy of our ConvNet: | ||
|
|
||
| ```julia | ||
| accuracy(x, y, model) = mean(onecold(cpu(model(x))) .== onecold(cpu(y))) | ||
| ``` |
There was a problem hiding this comment.
I think the indented julia blocks are making the output look weird. Maybe unindent them for now?
IDK. What else goes in there? Lots of the reference sections do start with some text, to supplement the docstrings. The logical place to add something sounds like the MLUtils page. But probably not the exact text, since it's long and doesn't say much? https://fluxml.ai/Flux.jl/stable/destructure/ Alternatively, a bigger "data handling" or something tutorial, which introduces how to read CSV files and image files and so on, as well as data-handling tools, might be nice to have if someone wants to write it. Maybe this is similar to what you picture the "how to" section being -- tutuorial-style pages, but not data-model-train-test examples, instead walk-throughs of other things? |
This moves 5 pages from https://fluxml.ai/tutorials/ to the docs at https://fluxml.ai/Flux.jl/dev/, each of which builds & trains some model. Xref #2105 (comment) about moving them.
I have made no attempt to update them, the main change is replacing the blog's title + author format with an H1 and an
!!! infoblock.I have not checked that they run. Nor does this PR make any attempt to make CI run them.
They should certainly be updated to explicit parameters etc. (I presume that by having them in the docs, we are undertaking to keep them up to date, rather than being fixed-in-time blog posts.) But better to make the update a separate PR.