Citing the fastai library?

(Even Oldridge) #1

I’m currently in the process of writing a paper, and would like to cite the library in order to direct people there as the paper builds off of the library. What’s the best way to do so? @jeremy @sgugger do you have a bibtex entry setup for the library itself?

0 Likes

(Fabrizio) #2

Hi, libraries and frameworks you use should not be cited in the bibliography. You simply insert a footnote with the link the first time you mention it inside the experiment section. However the bibliography should have an entry for each dataset you tested your model on, unless you made it up. Hope it helps

1 Like

#3

Why not? Because all the work people went into making those isn’t worth of academic recognition?

To answer the first question, we will write an article about fastai soon-ish that is going to be the thing to cite, but in the meantime, you should just use a link pointing to docs.fast.ai when citing fastai.

4 Likes

(Dmytro Mishkin) #4

@fabris I beleive, you are wrong. Everyone cites framework in bibliography now. Pytorch was cited as url, before publication come, for example. And (almost) all frameworks have their associated papers.

0 Likes

(Fabrizio) #5

I’m sorry, I meant no offense, Eventually, of course, it is up to the authors choose what put inside the references.
Even Elsevier’s guide for authors says that every used piece of software must be cited inside the references.
So you are completely right. However, let me explain why I wrote that.
While I was replaying I had in mind this old advise “In the paper are you gonna cite the maker of the pen you used to write it?”. I think that the ultimate goal of a library like fastai is to be widely used, while a paper to be cited. I also strongly believe that fastai already succeed in that and you guys should be very proud of what you have done. But of course this is just my point of view.
Finally, let me say again, I’m sorry if I have offended you or your work with my words. It wasn’t my intent!

0 Likes

(Dmytro Mishkin) #6

No offence taken :slight_smile: I think, the reason of citing libs is that because lots of them involve work of researchers, e.g. caffe. And researchers got evaluated on number of citations and papers. If one spent year of work and cannot even justify it to faculty board, it is sad.

1 Like

#7

No offense take, you caught me in a grumpy moment so my response was a bit dry.
I understand your point but in the field of ML you have very good articles that could be summarized as a bunch of tricks to make training work, so if those exist and are widely cited, why not a library? :wink:

2 Likes

(Fabrizio) #8

Now I see and I agree libraries like fast.ai deserves it!!:wink:

0 Likes

(Kaspar Lund) #9

From a scientific point of view its more a matter of enabling reproducibility . A paper is stronger/ more credible if the reader can reproduce the results. A notebook and/or af github repository with reference to essential libraries is a bug plus for a paper.

0 Likes

(Bobak Farzin) #10

I want to bump this again because I am working on a paper for the NLP challenge and would like to cite Fast.ai. Still link to the docs.fast.ai page?

1 Like

(Zachary Mueller) #11

I am writing an academic paper and I’d like to cite the library as well :slight_smile:

0 Likes

#12

We’re fully focused on the development of v2 right now, but will write that fastai article once it’s released. In the meantime, please cite with a link to docs.fast.ai

3 Likes

#13

Does something like this work as bibtex?

I feel like something should be put in author. Could you clarify or confirm?

@misc{fastai,
author = {},
title = {Fastai Python Library v1.0.57},
howpublished = “\url{http://docs.fast.ai/}”,
year = {2019},
note = “[Online; accessed 1-Sept-2019]”
}

0 Likes