search

rstudio / ai-blog

Repository for the RStudio AI Blog (formerly: TensorFlow for R Blog)
https://blogs.rstudio.com/ai/
48 stars 36 forks source link

add torch post #103

Closed skeydan closed 4 years ago

javierluraschi commented 4 years ago

I think we also need to add fix the dimensions of the Torch thumbnail, I think adding white space to the left/right of the logo would change the aspect ratio enough to make it look closer to the other posts we have.

Post is looking great already! Really love the Collab notebook!

skeydan commented 4 years ago

thanks for the reviews! Yeah, for the thumbnail, that is the only thing I can think of too (was going to ask if anyone knew an easier way...)

topepo commented 4 years ago

Can you show train_ds[1][[1]]$size() before mentioning the effect of transform_to_tensor? People are going to have issues understanding the problem before then.

Also, can you set the session seed to that the post is reproducible?

Also also, it would be helpful to explain the format/class of train_ds and test_ds since they are specialized data objects.

skeydan commented 4 years ago

@topepo thanks for commenting.

Can you show train_ds[1][[1]]$size() before mentioning the effect of transform_to_tensor? People are going to have issues understanding the problem before then.

I see what you mean, but I want to keep this post as straightforward and legible as possible -- it's really the introductory post on torch! On the other hand, as I totally understand people who want to look under the hood and peek at what's inside, I've created the notebook. In the colab, people can easily create new cells evaluating things differently, e.g., take a look what'd be returned if there was no such transform.

Also, can you set the session seed to that the post is reproducible?

I'd rather not, for similar reasons. I really just want to show the absolute essentials; and it is really not about reproducability here.

Also also, it would be helpful to explain the format/class of train_ds and test_ds since they are specialized data objects.

Again, same reasoning. This really is just the introductory post, whose main purpose is to get people interested! There already is a lot of documentation available, at once we've gone "official" with torch there will be many more posts using it. (It would be a bit different if we were just binding to Python, so I could mention, in passing, that something "is a Python generator", or similar. But since torch is built in R, explaining data structures really seems out of scope for this post.)

skeydan commented 4 years ago

hey @batpigandme thanks so much!

happy it's not as many as for the other one hehe