Lesson 1 In-Class Discussion

(Santhanam Elumalai) #188

Any idea about this issueipynb?

(Chris Palmer) #189

OK, thanks very much - I will write later today - right now I am still at work :wink:

(Soumyo Dey) #190

Which version of Python are you using? fstrings are only available in Python3.6

(Phani Srikanth) #191

f-strings have been introduced in Python 3.6. Most likely, you are using an earlier version.

(James Requa) #192

One thing you can try is to check the packages that are available for download on Anaconda’s website, for OpenCV as an example here

I have found that condaforge is usually a more reliable source than menpo.

(Santhanam Elumalai) #193

@ace139 @binga I just saw the python version which 2.7, Updating it now to 3.6. Will let you know the result

(Ravi Sekar Vijayakumar) #194

Do you know of any notebooks that do this math ? As a hardware engineer, I am interested in it.

(Ravi Sekar Vijayakumar) #195

The idea is for us to plot and visually pick the learning rate giving low enough loss but fast enough. Since you visually look at it, any noise due to mini batch local optima can be ignored and we can pick a value from a smooth plot.

(Anand Saha) #196

Please note that the actual size of the model in memory will be more than this. We are not considering many things here. I doubt if anyone calculates this other than on small models for learning purpose.

(Hafidz Zulkifli) #197

I normally listen to TWIML (this week in machine learning) and Linear Disgression. I enjoy listening to the former since it often goes into applications of ML. I used to listen to Data Skeptic, but I get turned off listening to the female podcaster (the way she talks sounds rather sarcastic/disgusted. Probably just me though).

(Vikrant Behal) #198

@jeremy I added a section ‘Resources shared during class’ at the end of your post. Pls. feel free to update/delete unwanted information.

(Ismaël Koné) #199

At first glance I think it should be
load_model(m, f+'wgts/{fn}.pth')
But please print the content of f:
so that we can give a correct answer.


The AI podcast from nvidia is excellent. They regularly feature experts who work in the different aspects of AI and more importantly discuss real world applicability


hi @jeremy, what is the purpose for dividing us into groups ? Are we supposed to work on a team project or simply to aid learning during the course? thx!

(Vikrant Behal) #203

Jeremy did mention this. I think he idea is to identify project and complete it with assigned team.

(Vikrant Behal) #204

Is this the same podcast mentioned in wiki above?


no it’s a different one

The AI Podcast

(Vikrant Behal) #206

Wow. Thanks! Do you wanna add this to podcast list above?


yep will do. :slight_smile:

(helena s) #209

If i recall correctly his argument is that backprop is not present in nature…