I have free GPU available from work. So we can share resources and I can share my screen. Needless to say if we do end up writing a blog, due credit will be shared
Personally I think this will be a great project because there is no central place for DataBlocks API in Fastai2 with multiple examples at the moment.
Not sure if we should move it to the non-beginner part. may be in the Mid-Level API discussion.
I agree with @VishnuSubramanian but I’m not building a top 5 solution. I’m just trying to build a full pipeline. Working with https://www.kaggle.com/c/plant-pathology-2020-fgvc7/data
I was talking about the GPUs used for submitting to Kaggle. Definitely we can try. If you have any ideas on how we an colloborate, we can srart.
If in doubt - the answer is probably “yes”
Thanks Jeremy. @VishnuSubramanian @barnacl
Let’s take this discussion to https://forums.fast.ai/t/datablocks-api-examples/66204
Count on my, I am also interested in learning datablock API
I was going thru documentation for untar_data and had question for below line
“Internally, if files are not available at fname
location already which defaults to ~/.fastai/archive/
, the files get downloaded at ~/.fastai/archive
and are then extracted at dest
location. If no dest
is passed the default_dest
to download the files is ~/.fastai/data
. If files are already available at the fname
location but not available then a symbolic link is created for each file from fname
location to dest
.”
Q. why dest defaults to ‘~/.fastai/data’ and fname defaults to ‘~/.fastai/archive/’ ?
As the function does download files from url
to dest
, then why it does lookup at fname
location. we can simply check whether it’s present at dest.
Might be a silly question but why is it called the head
of a model? These are the last layers, so I would think of it more as a tail.
how i read the func - when downloading, we’re usually downloading a zip. the zip files are archived, and unzipped to fname. we check the archive when downloading because we’re comparing the downloaded zip to the zip at the url, instead of the extracted file to the zip at the url.
I’ve created a summary and review thread containing: [Podcast + Writeup] Summaries + Things Jeremy Says to do + Qs and Unofficial Lecture 1 Notes
Both are Added & Linked in the Wiki.
When thinking of models reverse how you would normally describe them. The body is the bottom and the head is the top. Sorta like a tree and it’s roots, or even a person.
I think of it as a pyramid. Base is the largest (your input image), head is the top
Can you share tips for navigating the fastai code?
Do you have a local checkout of fastai and open notebooks from there? Did you load the generated python code into an IDE? Do you rely on doc() function and links to the source on github?
I tried the last option but I quickly end up with a mess of tabs open to github, dev.fastai and fastcore.fast.ai.
Also I’d like to take an advantage of the the fact that everything is a notebook, but I couldn’t find a good way for navigating through the library, other than grepping for the symbol I want, or matching the doc page with the notebook filename. And again an explosion of open tabs, this time to jupyter notebooks.
https://forums.fast.ai/t/source-code-mid-level-api/65755/181 some discussion over here. VSCode, vim … options
Hi,
I don’t seem to have the questions listed in the Questionnaire section. Can someone please share the link with the correct notebook? ---- I will pick it up from the github Fastbook page.
Check the GitHub repo when in doubt (it has the questions)
Yup. Will pick it up from there. Thanks