Hi @hiromi,
Nice to see you here!
Your notes have helped me a lot in making sense of the lectures from Part 1 and Part 2 2019.
I hope this encourages you to keep doing the same this time around!
Thanks for sharing!
Best regards,
Butch
2 Likes
Wonderful to hear 
Hi there, I’m Justin! You can find me on twitter @justinJDN
I work on the Data Science team at Yelp currently, and in previous roles I’ve had a bit of experience building DL models for both personal and professional reasons. Professionally, I’ve mostly focused on using RNNs to model time-series supply chain data and more recently, in some image analysis tasks related to content moderation. Personally, I’m very interested in both computer vision and macroeconomic applications. Most of my work has been done in TF 1.0 so I’m super excited to learn more about PyTorch and Fast.ai.
4 Likes
@jeremy Just curious but is there any backend functionality on this platform to export the posts and extract twitter handles? It might be kind of niffy to have a centralized list of twitter handles for fast.ai users. Also, if you give me the csv I’d be happy to parse it (I imagine a regex would be good enough as long as people use the @ marker).
@bibsian the forums are “Discourse”. It has an API. Have a look and see if the API has something like what you want. Let us know how you go!
Pleased to meet you! Hopefully we can soon move on to shaking hands the old fashioned way.
Speaking of old, I started out as a games programmer back in the 90’es. Thanks to the tech boom(s) that followed, I have since played software engineering roles in aviation, telecom, container shipping, banking, pharma and medical devices.
Two topics of particular interest to me include: (1) Design control of safety critical AI/ML systems, such as the FDA’s guidance on AI/ML in SaMD (2) Sequence modeling for time series prediction. I would love to hear from anyone doing related projects.
I am not active on twitter.
Stay healthy,
Rene
3 Likes
Welp, I gave it a go but I’m hitting a snag with permissions/cookies. Feel free to check out the code in this notebook: https://colab.research.google.com/drive/1Rr2o90HP8UV6q5GGhBkOufwWc2mc_BHT (anyone’s welcome to edit it).
Overall I think it’s definitely possible if I could figure out the user verification process using either an API key or signing in through the API but the docs weren’t to clear to me on how to go about it.
1 Like
Hey Andrew,
Great idea! This seemed like a fun little project, so I hacked something together this evening. Here is a repository with some code to retrieve this post and all it’s replies.
I wrote a basic bash script / some regex to pull out Twitter and LinkedIn URLs from the HTML content, but it’s worth noting that this isn’t catching everything. Some of the Twitter handles aren’t encoded as hyperlinks, so my baseline script isn’t catching them. In the interest of preserving user privacy, I didn’t upload the actual extracted data to the repository. You can either run the code locally, or grab it from here (https://send.firefox.com/download/6de87e85033b835b/#8feQHX4iRW1elva3T22x7Q) (password is “^h@w5Ge9PTM7*4LHYp”, and the link will expire in one week).
It might be a fun extension of this to train a NER to catch the missing handles. It doesn’t seem horribly difficult, but I could see it being a little challenging to differentiate between the “@FastAIUsername” vs. the “@TwitterHandle”. Let me know if you want to work on it together
Thanks!
Daniel
2 Likes
I looked into the Discourse API documentation, but they don’t really have support for “exporting” a full post and all of it’s replies.
What I ended up doing was visiting the main posts URL and just adding “.json” to the very end of it (i.e. “https://forums.fast.ai/t/introduce-yourself-here/62445.json”). You can then access the “post_stream”.“streams” element to get an array of all the post ids. Then you would visit each of the posts individually to download (i.e. “https://forums.fast.ai/posts/post_id”). To speed it up, I did this with the requests library, and just added the _forum_session cookie once I was authenticated.
4 Likes
Now that’s cool! Great find!!! Will it tell you information such as “likes” etc?
This is great! I like how you used headless chrome; definitely the quickest way to get around that verification issue 
.
And yeah, I’d love to work with you on this. We could probably do it pretty quickly with spaCy and scraping some random handles from twitter and create some rules for discarding forum names (I thought it was pointed out in the json coming back but I’m on my iPad right now so I can’t verify that).
I’ll DM you tomorrow; I still need to setup my paper space so wondering if this might be a good reason to give that a go in conjunction with developing something in nbdev… if you’re up for that.
Hi everyone, I am Chris, I live in Melbourne and work as a Data Engineer at Ahpra - the Australian Health Practitioners Regulation Agency, but I am passionately interested in AI and DL, and fast.ai gave me a great start into this interesting world some years ago now! I faithfully apply myself to improving my skills, one of the most interesting other things I did was a Udacity Deep Learning nanodegree, which I was given a Facebook Pytorch scholarship for. Doing Pytorch on a low level gave me more confidence to handle the top-down approach here at fast.ai.
I’m really looking forward to the unfolding information we will enjoy in the latest iteration, Jeremy is such a great teacher, he is much loved and appreciated by many in our community of AI practitioners, and I am one of them! Currently I am working from home - a COVID-19 related directive from our company, so I think I’ll have more opportunity to keep up with this course 
6 Likes
Hi all, I am Vishnu Subramanian from India. I am happy to be part of fastai journey and I am doing this for the 4th year. I am so excited and would love to put more effort than ever this time. Till a few months back I was building a product for the travel domain in a startup. Recently I launched my own company called jarvislabs.ai focused on building products for accelerating AI adoption. Thanks to fastai it keeps me motivated, won few medals in Kaggle, spoke in conferences, authored 1st book on PyTorch(outdated). Some of the goals for this year is to write more blogs on what I have been and will be doing. I am currently working on creating solutions for old Kaggle competitions if anyone is interested to work together let me know.
6 Likes
Hello everyone
I am Psychologist, associate professor at Universidade São Francisco, Campinas, Brazil. I am currently visiting scholar at University of California Berkeley at Institute of Personality and Social Research. I study assessment of intelligence, personality and psychometrics. Since 2008 I dive into R and define myself as a data scientist with focus on psychometrics.
My journey in deep learning started in 2017 when I attended two courses with Roberto Lotufo from NeuralMind where he mentioned fast.ai. I watched 2017 DL I videos as many of you I was mesmerized by the classes (thanks Jeremy !!) As an outsider from CS, math and statistics I always learned via top-down/whole-game-first approach. So those classes was exactly what I needed to learn DL. Since then I dreamed to attend fast.ai DL in person. As this year I am living on the other side of the bay bridge, on the right time, now locked in my home, it is true, but from the cloud I am here meeting you
I am interested in applying DL to psychology, psychometrics and assessment on topics like use of word embeddings to understand how people describe their personality traits, NLP for evaluating quality of creative metaphors and verbal creative productions, automated scoring of verbal constructed responses to psychological tests. Just for fun, while learning baby steps in DL I was curious to see what VGG sees on a very old traditional projective test, the Rorschach inkblot test (http://www.labape.com.br/rprimi/ds/vgg_rorschach.html )
As many of you already commented, I also felt overwhelmed by the incredible talented group is here. I am proud to be part of it and looking forward to learn and evolve and see what amazing new experience and creations will be done the following months.
Be safe
Ricardo
@rprimi
3 Likes
So we’re nearly colleagues, at least in name
1 Like
Hi
I am Dina, I work in the Bioinformatics world. My background is in computer science and machine learning. This is my first hands-on DL experience, so glad to be part of the course, and learn from all your experience. I would love to be able to apply it on genomic data and get fancy results with!!
5 Likes
Hey! 
I have not introduced myself earlier as I suspected I would not benefit too much from being on the forums for this course. But here I am, learning a lot as always, and as always underestimating the magic of these forums! 
My story is quite simple - I used to have a corporate job that was crushing my soul (offshoring / outsourcing is not a fun business to be in as an employee). I accidentally ended up learning web development and worked for a while as a Ruby on Rails dev. Sailing the perilous ocean that is Internet I came across fast.ai, learned a little bit about machine learning, and have been doing that for over a year now. I now work as an AI Research Engineer at the Earth Species Project and am enjoying every bit of if!
So grateful to continue to be part of the fast.ai journey. The last couple of days have not been what I anticipated, but have definitely been amazing! So many new faces and so much positive energy on these forums and in the zoom chat! 
Thx for an amazing week and see you all around!
13 Likes
Hi everyone,
Sorry for the late introduction, I’m Hervé from Paris, France. Happy to meet you all.
My background is in investment/fund management, and I currently work for a fintech that endeavours to promote data science in finance.
As many of you here, I have followed fastai since v1 and I am very grateful to Jeremy, Rachel and Sylvain for their fantastic work.
I’m very excited to take part in v4. I am looking forward to learning and collaborating with the fastai community!
4 Likes
Hi,
I am Issa and currently a data science postdoctoral fellow at the Data Institute,University of San Francisco. My background is in mathematics(PhD), probability and statistics (master) and computer science (undergraduate).
I started practicing and learning deep learning through fastai since August 2020. I am very grateful to Jeremy, Rachel, Sylvain, and many others who made and continue to make my transition(of my many other) so smooth. I am very interested in AI applications in healthcare specially improving access to healthcare(doctors) in developing countries in Africa (like Chad) and hope to teach deep learning through fastai there in the future.
I am very excited about this course (I really like Jeremy teach style) and looking forward to learn deep learning by practice, understand better fastai, and collaborate with you all.
6 Likes
Hi, I’m Ben Mainye from Kenya. I came for the course in person in fall 2018 and spring 2019 as a diversity fellow. I’m honored to be chosen again to participate in the course. I was formally trained as a microbiologist but I have many ideas on research and I had limited skill to do them since my undergraduate education didn’t empower me much. It feels like it was leading me to go for a masters degree. Besides, the course was leaning into being a researcher/Academic.
Anyways, where am I in my deep learning journey? After the previous iteration of the course I took a step back to learn and try to grok deep learning by trying to build a deep learning library from scratch using Andrew Trask’s book Grokking Deep learning and notebooks from part 2 last year(ongoing). Before I can continue adding more features to my immune classifier and Mosquito classifier. At the same time, researchers from a laboratory here in Kenya called the Institute of Primate Research really liked it and they’ve collaborated with me to write a research paper about it with the extension to look into neglected tropical diseases. Therefore, at the moment I’m making products around diagnostics and surveillance of intracellular and extracellular parasites. On the other hand, I’m trying to collaborate to finish up another paper about Open science and how it has evolved over the years. Which you can read about here. Plus, I have been collecting data about my skateboarding tricks and I’m gonna develop something interesting as soon as I finish collecting data to help me learn skateboarding better with deep learning !
What I’d like to get out of the course. I’d like to continue fixing my previous projects to include interesting techniques on data augmentation like Mixup and learn how to use jupyter widgets better especially to edit multiple images at a time by reading the fastai docs and probably contribute to the module dedicated to medical imaging – I think one of the apps I’m making could be useful since it heavily uses ipywidgets. Lastly get better a understanding of Image Segmentation.
If you want to know more about me. You could check out my upcoming blog or portfolio site here. If you can, please give me feedback.
Thanks again for the opportunity Jeremy, Rachel , Sylvain and USF!
Twitter @Shuyin_ben.
8 Likes