Study Group for Fast AI Part 1 Version 2 (Seattle area or online?)

(Vikas Bahirwani) #1


I am wondering if there are people either at Microsoft (Greater Seattle Area) or in Seattle area that have recently started Fast AI Part 1 Version 2 recently and would like to be a part of a study group?

I am also open to folks joining online for mutual accountability :slight_smile:


(Dan Goldner) #2

Hi - I just started part 1 version 2. If you end up with an online contingent, let me know. -Dan (not in Seattle :))

(Vikas Bahirwani) #3

Hello Dan. Not a problem. We can still hold each other accountable :slight_smile:

Do you plan to do one lecture a week? What does your study availability look like? We can also Skype to ask questions / compare notes if you like.


(Dan Goldner) #4

I’m aiming for 1 lecture per week. I’m half-way through lecture 1.
Study hours are likely to be evening/weekends, but I can get free for a Skype call most days.
If you can get a live local group going in Seattle, I can adapt as far as schedule …

(Vikas Bahirwani) #5

sounds good. Keep you posted. And let me know when you finish lecture 1, we will have a check in.

(Dan Goldner) #6

Lecture 1 done! Played with running the fit multiple times; results changed slightly each time – perhaps due to the data being randomly subsampled in minibatches? Also started training a roses-vs-tulips model on a small data set. Have run it but not yet tried to improve the fit, which varies from run to run between accuracies of 50-80%.

(Vikas Bahirwani) #7

yes that is correct. The fit will change slightly because data is randomly sampled.

I tried my code with TajMahal vs. Pyramids and got 100% accuracy (some type of over fit perhaps). I am going to play more tonight but let’s Skype tomorrow afternoon if you are free?

I want to see how you played with the code and share what all I did (basically following the idea in this post).

My skype id is feel free to add me.

(Dan Goldner) #8

Great - I’m dangoldner on Skype, and @gmail. But tomorrow doesn’t work - let’s connect by email for logisitics.