Remote NLP Study Group meets Saturdays at 8 AM PST, starting 12/14/2019

Deep Learning
45

I think they added TPU support from v1.3 onwards

46

Hello! I’m following the instructions in the ppt document from Saturday but I get the following import error when I try to open the notebooks. Do you know why? :cold_sweat:

  (fastai) C:\Users\Silvia\fastai\course-nlp>jupyter notebook
    Traceback (most recent call last):
      File "c:\users\silvia\appdata\local\programs\python\python36\lib\runpy.py", line 193, in _run_module_as_main
        "__main__", mod_spec)
      File "c:\users\silvia\appdata\local\programs\python\python36\lib\runpy.py", line 85, in _run_code
        exec(code, run_globals)
      File "C:\Users\Silvia\AppData\Local\Programs\Python\Python36\Scripts\jupyter-notebook.EXE\__main__.py", line 4, in <module>
      File "c:\users\silvia\appdata\local\programs\python\python36\lib\site-packages\notebook\notebookapp.py", line 81, in <module>
        from .services.kernels.kernelmanager import MappingKernelManager
      File "c:\users\silvia\appdata\local\programs\python\python36\lib\site-packages\notebook\services\kernels\kernelmanager.py", line 19, in <module>
        from jupyter_client.session import Session
      File "c:\users\silvia\appdata\local\programs\python\python36\lib\site-packages\jupyter_client\__init__.py", line 4, in <module>
        from .connect import *
      File "c:\users\silvia\appdata\local\programs\python\python36\lib\site-packages\jupyter_client\connect.py", line 35, in <module>
        from jupyter_core.paths import jupyter_data_dir, jupyter_runtime_dir, secure_write
    ImportError: cannot import name 'secure_write'
47

you will save yourself some environmental work if you give google colab a shot. (I switch back and forth all the time) for that error, something to do with your pip or conda env. you might want to review the env setup instructions.

48

Welcome, @Silvia! Looking forward to having you with us this Saturday!

49

I found this:

The recommended fix is:
pip install --upgrade jupyter_client

Try this and see if it works?

50

I tried it but the problem persists
 :S I tried some other options but anything seems to work, so I’m launching the notebooks from Anaconda Navigator.

1 Like
51

Great! If you can run the notebooks, you’re all set.

52

Thanks for this.

1 Like
53

The Norvig-Chomsky discussion referred to in the first notebook parallels Richard Feynman’s much earlier comments on Greek vs. Babylonian Mathematics (in this 10 minute clip from one of his Messenger Lectures, given at Cornell University in 1964) Here, Greek mathematics represents the axiomatic approach where all knowledge must be systematically derived and proven from a set of postulates. Babylonian mathematics represents the approach of learning about nature by looking for patterns, not necessarily understanding why the world is that way, but being able to use that knowledge to predict real things about nature (such as lunar eclipses!) Feynman concluded that Physics needs both approaches.

1 Like
54

This week’s assignment (complete before Saturday 12/21/2019 class)

2 Likes
55

BTW the top post is a wiki post, so feel free to put the most up to date info directly in there, if it’s helpful to people. (I normally both post new info in a reply, and also update the top post, FYI.)

3 Likes
56

Great! I had not noticed that. Good to know!

57

San Jose. (Moved from Santa Monica years ago)

1 Like
58

Ciudad de MĂ©xico

1 Like
59

The Fastai NLP Study Group
will meet Saturday Dec 21st, at 8 AM PST, 11 AM EST, 5 PM CET, 9:30 PM IST

Topic: Topic Modeling with Non-negative Matrix Factorization (NMF) and Singular Value Decomposition (SVD)

Suggested homework / preparation:

  1. Watch NLP video #2
    Video playlist is here
  2. Read and work through the notebook Topic Modeling with NMF and SVD
    Course notebooks are in github repo

Join the Zoom Meeting when it’s time!

To join via phone
Dial US: +1 669 900 6833 or +1 646 876 9923
Meeting ID: 832 034 584

The current meetup schedule

Sign up here to receive meetup announcements via email.

60

I was looking for a little more context on SVD. Came across this https://www.youtube.com/watch?v=P5mlg91as1c which I found useful.

1 Like
61

Great meeting today! Thanks to all who contributed questions and comments. Here is the chat text.

08:00:43 From kodzaks : Yes, thank you very much for doing it!
08:01:16 From Mutum : Agreed,Yes much appreciated
08:14:04 From VĂ­ctor Peinado : Let me share this website with awesome animations of matrix multiplications: http://matrixmultiplication.xyz/
08:14:47 From csaroff : Thanks!
08:15:03 From Hans Paul P : nice
08:35:00 From Surya : For R dig into tm related libraries ,also word2vec
08:36:27 From Chris Palmer : These are just historical - for instance removing stop words and lemmatization is still needed for topic modelling - no matter what technique you use
08:40:07 From Chris Palmer : Should have types NOT just historical

08:41:35 From Mash Zahid : Thanks, Surya.
08:42:30 From VĂ­ctor Peinado : Regarding stemming/lemmatization, in general terms, traditional ML algorithms work better when you reduce the vocabulary by stemming or lemmatizing, while modern DL approaches are powerful enough to leverage these variations
08:43:22 From VĂ­ctor Peinado : Also, lemmatizing is more convenient with languages with rich morphology (eg. roman languages) while stemming is ok for language as English
08:44:43 From mattc : That multiplication R X Q is not a cross product but normal matrix multiplication right?
08:44:51 From Abhinav Verma : yes
08:45:15 From mattc : cool. I always get that confused.
08:45:31 From Abhinav Verma : hehe, no worries
08:47:40 From csaroff : Isn’t the cross product the same as “normal matrix multiplication”?
08:48:19 From Abhinav Verma : yes , in a way. you can say that
08:48:52 From Abhinav Verma : I would have to look up the exact difference between the two though
08:49:04 From Mash Zahid : This may help the intuition between factorization and multiplication: http://nicolas-hug.com/blog/matrix_facto_2
08:49:45 From mattc : I bookmarked that Mash, thanks.
08:49:55 From Mash Zahid : But linear algebra is a major undergrad course, not easy to distill into just a few slides as we do here.
08:54:27 From jekaterina : Another great linear algebra course is here: http://immersivemath.com/ila/learnmore.html
08:54:54 From jekaterina : it has all the interactive graphs and diagrams that help visualize main concepts
08:56:23 From VĂ­ctor Peinado : Another resource: Essence of Linear Algenbra, by 3Blue1Brown https://www.youtube.com/watch?v=fNk_zzaMoSs&list=PLZHQObOWTQDPD3MizzM2xVFitgF8hE_ab
08:56:29 From Mash Zahid : I like “Coding the Matrix” by Philip Klein of Brown Univ, delivered via Coursera. It’s a deep content intro to linear algebra (and more), with a focus on applications in computer science. The course is accompanied by a textbook written by Klein, which makes the course material better organized and more in-depth than slides and videos alone would allow.
http://codingthematrix.com/
08:58:48 From Mehul’s iPhone : thanks everyone for the links :slight_smile: has anyone done the undergraduate linear algebra course at MIT
08:58:50 From Mehul’s iPhone : https://ocw.mit.edu/courses/mathematics/18-06sc-linear-algebra-fall-2011/
08:59:15 From Mehul’s iPhone : any thoughts about this one ? it is by Gilbert Strang
09:00:17 From Mash Zahid : Gilbert Strang’s linear algebra course at MIT is legendary!
09:00:32 From Srinivas : The gilbert strang course is the very best. Other that is good although at a more basic level than Strangs is LAFF from UT Austin at edx https://www.edx.org/course/linear-algebra-foundations-to-frontiers
09:01:13 From Mash Zahid : Mastering his course will give you a special superpower :grin:
09:02:24 From Mash Zahid : In my not-so-humble opinion, #1 Strang at MIT, #2 Klein at Brown because his teaching style is a bit more basic/simpler.
09:02:46 From mattc : I want superpowers. Is it the edx class you’re referring to?
09:02:48 From Mash Zahid : (I don’t know the UT Austin prof.)
09:04:02 From Mash Zahid : MIT Strang course is link above from MIT OCW. Klein’s materials are on website & YouTube. All free, so everyone can have superpowers if they work for it!!!
09:05:28 From mattc : great. Maybe I’ll ruin my holiday break and work through some of that.
09:08:13 From Chris Palmer : Looks like Coding the Matrix on Coursera has been deprecated. I found a link to lectures and class content here: https://www.reddit.com/r/Python/comments/5j042r/coding_the_matrix_this_course_for_anyone_need/
09:15:30 From kodzaks : GitHub for that courser course https://github.com/pauldevos/Coding-the-Matrix
09:16:23 From Hans Paul P : can you please jot down the mathematical understanding of it pls?
09:16:56 From Hans Paul P : the more groups the more we understand?
09:17:09 From Mash Zahid : Yes
09:17:19 From Mash Zahid : Or “topics"
09:17:37 From Mash Zahid : Yes, clusters
09:17:54 From Hans Paul P : ok thanks
09:18:32 From Surya : Yes but the trade off is how minimal you keep your clusters and yet retain most of the information 

09:19:51 From csaroff : Is this the same type of problem that variational auto encoders aim to solve?
09:20:56 From Mutum : so just like giving k values in K-means
09:27:25 From Alvaro G : Mashad, can you please post the link to such documentary?
09:27:48 From VĂ­ctor Peinado : is this the documentary? https://www.youtube.com/watch?v=5dZ_lvDgevk
09:28:55 From Chris Palmer : That youtube link is not available to me here in Australia

09:30:21 From csaroff : Vpn?
09:31:53 From Yuri S : Want to clarify that Google is not working on identifying Uighers in China.
09:32:08 From Chris Palmer : I guess so :slight_smile: - keep meaning to get around to using a VPN

09:36:29 From Pazhanivel : is it not that both autoencoders and PCA/SVD are used to reduce the number of features?
09:38:05 From Mash Zahid : PBS Frontline “In the Age of AI” https://www.pbs.org/wgbh/frontline/film/in-the-age-of-ai/
09:50:51 From mattc : Thank you, Mash, that makes sense
09:51:27 From Rey Farhan : Does anyone have a link to a good intro to autoencoders for this kind of task?
09:51:31 From Mash Zahid : You are most welcome and kind. (My family always remind me that I know nothing! :grin:)
09:52:43 From Mash Zahid : Autoencoders, refer to Aurelion Geron’s book and code https://github.com/ageron/handson-ml2
09:52:51 From Rey Farhan : thanks
09:53:06 From Pazhanivel : Thank you
09:53:37 From Team TWiML : Thanks everyone!

62

Recent Forbes article a propos of our brief class discussion on social responsibility: “Tech firms in the United States are lending expertise, reputational credence, and even technology to Chinese surveillance companies, wittingly or otherwise.”

63

@jcatanza Hi Joseph, is there a meetup this week?

64

The Fastai NLP Study Group will meet
Saturday January 04, at 8 AM PST, 11 AM EST, 5 PM CET, 9:30 PM IST

Join the Zoom Meeting when it’s time!

Topic Modeling with Non-negative Matrix Factorization (NMF) and Singular Value Decomposition (SVD), revisited

Suggested homework / preparation:

  1. Watch NLP video #3

Video playlist is here

  1. Read and work through the notebook 2b-odds-and-ends

  2. Review the notebook we covered last meetup Topic Modeling with NMF and SVD

Course notebooks are available on github

To join via Zoom phone
Dial US: +1 669 900 6833 or +1 646 876 9923
Meeting ID: 832 034 584

The current meetup schedule is here

Sign up here to receive meetup announcements via email.