Chit Chat Thread

[karthik.subraveti]

Typically configurations are stored in yaml files and passed into the scripts. Passing in kwargs is kind of ugly cause of two things

1. Non intuitive to the programmer writing the code to know what to expect in kwargs.

2. Validation. The key and value might not conform to programmers expectation and error checking has to be built into every method.

[drscotthawley (Scott H Hawley)]

Ok, but I don’t want my inner subroutine to have to parse a yaml file just know whether to set a particular flag.

So the same question still remains: how best to propagate the information “all the way in” to subroutine from either the command line or a yaml file?

I ask this because I’m doing lots of runs trying different ideas, and so far just editing the source code directly hasn’t been a good practice as far as keeping a record of which runs have which settings.

[jeremy (Jeremy Howard)]

@drscotthawley

So the same question still remains: how best to propagate the information “all the way in” to subroutine from either the command line or a yaml file?

Use Google Fire and add the params to your function signature. To have stuff propogate everywhere, you can use a singleton like we do in fastai (ours is called fastai.defaults) and dump config stuff there.

[karthik.subraveti]

You can have a single config class loading the yaml file and the config instance can be a global instance, much like environment variables. I think however the solution depends on what you are actually trying to implement.

[karthik.subraveti]

I heard of google fire for the first time in yesterday’s class :-).
It sounds like a very neat way of adding command line args. The current args(parse args …etc ) lack the application context.

[tanyaroosta]

@magiclantern

I’m also interested in NLP. I’m also interested in transfer learning - as an example in the medical field we often only have access to a small set of clinical notes - which makes it hard to get a good model from scratch. Other areas of interest are sentiment classification (not just positive vs negative). Language models and the ability to generate text is also of interest.

I am in South Bay too and very interested in study group/Kaggle. My email is tanya.roosta@gmail.com

[kai (Kai Lichtenberg)]

Hey @maxim.pechyonkin I remember you’re name from your blog, awesome series on Capsule Networks! Did you ever use a capsule based network or have seen one in the wild so far?

[axelstram (Axel Straminsky)]

Anyone from Argentina?

[init_27 (Sanyam Bhutani)]

I just realised this is worthy a share here too:

I’ve been spending much time getting better at Python.

I highly recommend following James Powell twitter: @dontusethiscode and watching any (yes, literally any-they are that good!) of his talks on YouTube for some great Python insights.

Do you know of any other great people to follow for getting better at Python?

[markp (Mark)]

Just came across this, shared by Ian Goodfellow on social media. Really cool!

“Semantic Image Synthesis with Spatially-Adaptive Normalization”

https://nvlabs.github.io/SPADE/?fbclid=IwAR3zbsP1ETOQSVjMI2c48GmP-uLpDvkcNKOAJQRmzzZ6GbOLs7AZrVn4vSo

[wittmannf (Fernando Marcos Wittmann)]

Uber Open-Sources Ludwig Code-Free Deep-Learning Toolkit

[jcatanza (Joseph Catanzarite)]

Please fix your twitter URL, which currently links to Hortonhearsawho!

[magiclantern (Seth)]

I realize now that I didn’t answer part of rkingery’s question: I’m a remote participant. I’m in the Denver area. A remote study group would be great for me.

[tomsthom (Thomas Chambon)]

Hi everybody, I’m following the course from France, @tomsthom on twitter.
Is there anyone following from France?

I find the development process in Jupyter notebook to be very interesting!
I also feel very productive in Jupyter but didn’t think it was possible to develop “real” software using it!
Very happy to learn how to do it with the techniques used for fastai development.

[NandoBr (Fernando Melo)]

Hi, thanks for pointing that out.
It has a capital letter F in the beginning.
The correct is @Fmelobr on Twitter.
Thanks.

[PierreO (Pierre Ouannes)]

I agree James Powell is wonderful ! I even made a blog post about how decorators work in Python based on one of his talk

[jcatanza (Joseph Catanzarite)]

@NandoBr

@Fmelobr

Thanks, Fernando. You have (at least) one new twitter follower;-)

[chans.best (chandan)]

HI folks i would like to hear from you how do you manage your time studying for this course.
How many hours do you put in , how do you manage etc, perhaps this could be topic for another thread.
I have a demanding full time job i only find time to read or watch videos during weekdays commute. Any actual coding i need to wait for weekend and its kind of challenge. I would like some inspiration or advice form others if i can get.

[Interogativ (Bart Fish)]

Don’t know how many of you have seen this yet, but they’re $99 and I just ordered two of them:Jetson Nano by Nvidia.

It runs python and pytorch according to their post. Has 4 USB ports, Gig Ethernet, 128 CUDA cores, 4GB DDR 4 RAM

[AmanDaVinci (Aman)]

Hi Chandan,

I think you will find this thread very helpful:

How to do fastai - Study plans & Learning strategies

I have been long searching for the most optimal way to study the fastai courses and make the best out of this transformative opportunity. I have read the forums to see what others have been doing, researched the web for tips & tricks and kept Jeremy’s teaching philosophy in mind while formulating my on-going learning strategy. My Fastai Study System My learning strategy has always been evolving and the current iteration looks something like this: 1. Lecture first pass: The first time where I …

Many Fastai veterans have posted their own study system, goals and experiences. This reply by Theodore might be something close to what you are looking for:

@Gabriel_Syme

My way of learning has been a sort of transfer knowledge process where I take insights from the awsome work Jeremy and the team are doing, trasnfer it over to my domain and its specific problems that move me, and develope ‘unique’ (to my domain at least) solutions to all that.

Happy learning!