pytorch-ignite / pytorch-ignite.ai Goto Github PK

Hugo highlighted line numbers are not preserved as shiki override the code blocks.

This can be applied to blog posts and tutorials

Once we solve issue #10 we can work on how to display the output of all tutorials (see picture below for reference) while also caching them so they aren't expensive for each new build.

@Priyansi can you fix bad links here, please

• tutorial goes to a wrong location: https://pytorch-ignite.ai/docs/tutorials
• how-to guides goes similarly to : https://pytorch-ignite.ai/docs/how-to-guides

@ydcjeff can we make some CI links checks to ensure that they are correct.

The index page of posts/ is automatically generated with an image, preview text and titles of the available blog posts. However, there is nothing like this for docs and we have to manually list down all the topics. Any ideas and suggestions on how to display various topics on the index pages will be appreciated!

Ideas List:

• Grid like structure - @trsvchn
• Tag wise filtering of topics (for ex- computer vision, nlp, etc)

While copying code snippets via the Copy button, an alert pops up every time like this -

It would be a better user experience if we remove this.

So directory based sidebar is added in ff3e6ce.

Questions is should we show in all pages and do not show in some pages we don't want?

They have to be same size.

• Link examples repo as a submodule
• Add a script based on nbconvert to:
  • convert ipynb notebook from examples to markdown files
  • prepend hugo related header (post title, footer...)

Almost recopy https://pytorch.org/ignite/master/about.html and update the info.

If we limit a page by a certain amount of tutorials or guides, we might need to have pagination pages for older content.
It doesn't make sense to have infinitely scrolling pages.

https://pytorch-ignite.ai/api/

As of now, if I need to copy a code snippet from the site I need to highlight the entire area before copying. This is inconvenient and would be better if the users have access to a quick Copy to Clipboard button. Ideas on how to implement this: https://www.dannyguo.com/blog/how-to-add-copy-to-clipboard-buttons-to-code-blocks-in-hugo/

https://gohugo.io/tools/search/#readout

https://gohugo.io/templates/taxonomy-templates/

We would like to have the front page look like numpy.org

• Header (we already have it and it is nice)
• our logo (logo mark + text) on the left
• "get started" on the right as numpy has
• horizontal tape with our latest news

@vfdev-5 mentioned to add sidebar so that we can shorten the link clicks for nested contents.

I wish to make sidebar to show the link to the files (posts/*.md) in a directory (e.g, posts/). And show the h2 of the current visiting page.

Sidebar draft design:

| posts
|-- a.md (current viewing page)
    |-- h2
    |-- h2
    |-- h2
    |-- h2
    |-- h2
|-- b.md
|-- c.md
|-- d.md

WDYT? @vfdev-5 @trsvchn @Priyansi

This issue lists Renovate updates and detected dependencies. Read the Dependency Dashboard docs to learn more.

Open

These updates have all been created already. Click a checkbox below to force a retry/rebase of any.

• chore(deps): update all non-major dependencies (prettier, shiki)
• chore(deps): update actions/setup-python action to v5
• chore(deps): update dependency jsdom to v24
• chore(deps): update dependency prettier to v3
• chore(deps): update dependency shiki to v1
• chore(deps): update peaceiris/actions-gh-pages action to v4
• chore(deps): update peaceiris/actions-hugo action to v3
• Click on this checkbox to rebase all open PRs at once

Detected dependencies

• Check this box to trigger a request for Renovate to run again on this repository

@ydcjeff I would like to put somewhere the links on our previous public presentations and other resources like posters etc.
Here is the content:

• PyData Global 2021 Sprint slides: https://pytorch-ignite.github.io/pydata-global2021-slides/
• Python Bucaramanga Meetup: https://vfdev-5.github.io/pybucaramanga-pytorch-ignite-slides/
• PTCV21: https://vfdev-5.github.io/ptcv21-pytorch-ignite-slides/
• PTED2021: https://drive.google.com/file/d/1YXrkJIepPk_KltSG1ZfWRtA5IRgPFz_U/
• PTDC2019: https://drive.google.com/file/d/1bqIl-EM6GCCCoSixFZxhIbuF25F2qTZg
• PTDC2018: https://drive.google.com/file/d/1_2vzBJ0KeCjGv1srojMHiJRvceSVbVR5
• ...

It can be a new page or a section somewhere, I do not know yet where it would be the best place.

As I recall from our last meeting, Code Generator has to be included under Ecosystem on the main page (pytorch-ignite.github.io).

The Learn More button under Unparalleled flexibility of Handlers in the Places Ignite Shines section is out of its designated box and not horizontally aligned with other boxes.

Screenshot -

When I click on documentation as of now, it redirects back to the old documentation.
A new page divided into 4 sections with links that lead to tutorials, how-to's, explanation and API reference is needed.

It's a minor inconvenience but a link in the rendered markdown table isn't highlighted or changes colour. A user has to hover over it to figure out if it's a link.
Left text is not a link and the right text is.

Using iOS safari, some code cells rendering problems appear. The font size seems not coherent.

Some screenshots below

Job "Run ./bin/htmltest dist" is currently failing

concepts/02-events-and-handlers/index.html
  href blank --- concepts/02-events-and-handlers/index.html --> 
  href blank --- concepts/02-events-and-handlers/index.html --> 
========================================================================

https://github.com/pytorch-ignite/pytorch-ignite.ai/runs/4915145002?check_suite_focus=true

Numpy tutorials have easy to use download links (.ipynb, .py or .md) at the top of every page, courtesy of the Sphinx Book theme. Is it possible to recreate something identical for this site? Ideally, there should be 4 buttons - Open in Colab, Download as a Jupyter notebook, View in GitHub and Download a python file.
When Numpy renders its tutorials, they create a Jupyter Notebook and other necessary files that are linked to these Download buttons. I think we have to implement something like that.

Related pytorch/ignite issue: pytorch/ignite#2415

I am not sure if this is possible with Hugo but it'd be great if instead of us formatting code, Hugo could automatically do it while rendering the site as it happens right now in Ignite with Black.

Let's replace

with a text like "Latest release: v0.4.6" and the link to the latest release.

@gucifer please update the notebook and generate its markdown with the following fixes:

1. Use # only once for the title and ##, ### etc for others. As title is specified in tags as title: ..., you have to update all headers by adding one more # to each of them.

2. Fix this :

# Models for GAN

## Generator

The latent space dimension of input vectors for the generator is a key parameter of GAN.
....

## Generator

You have twice the header ## Generator

3. Fix typos:

- dataset, emmiting events as it goes.
+ dataset, triggering events as it goes.

- Events - Events are emmited by the Engine
+ Events - Events are emitted by the Engine


- Event is emmited by the Engine. 
+ Event is emitted by the Engine.

By the way, the following sentence is a bit incorrect

Handlers - These are functions which can be configured so that they are triggered when a certain 
Event is emmited by the Engine. Ignite has a long list of pre defined Handlers such as checkpoint, 
early stopping, logging and built-in metrics.

=> Handlers are any functions and not only the ones under the link. Please fix the wording.

4. Let's merge neighbouring cells together for better MD rendering.
   For example

In [ ]:
image_size = 64

In [ ]:
data_transform = transforms.Compose(
    [
        transforms.Resize(image_size),
        transforms.CenterCrop(image_size),
        transforms.ToTensor(),
        transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
    ]
)

In [ ]:
train_dataset = ImageFolder(root="./data", transform=data_transform)

In [ ]:
test_dataset = torch.utils.data.Subset(train_dataset, torch.arange(3000))

=>

In [ ]:
image_size = 64

data_transform = transforms.Compose(
    [
        transforms.Resize(image_size),
        transforms.CenterCrop(image_size),
        transforms.ToTensor(),
        transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
    ]
)

train_dataset = ImageFolder(root="./data", transform=data_transform)
test_dataset = torch.utils.data.Subset(train_dataset, torch.arange(3000))

Please apply that for all cells that are not separed with any text or output.