Study group Polska

(Jeremy Howard (Admin)) #37



Didn’t notice that one. That was fast. Thanks!


I set up fastai (previous version) on ubuntu 16, ubuntu 16 gnome, ubuntu 18 without any problems.
I think the trick is to NOT install CUDA on your own, let env do everything. And avoid pip when possible.

I just have lots on env right now :frowning:
one for keras, other for vanilla pytorch, other for fastai, another for some DL style transfer demo. Because each requires different CUDAs and diff ver of pytorch etc

When I tried installing CUDA following instructions from nvidia, I ended up re-installing everything :confused:

If you don’t menage to make it work, follow Jeremy advice, and just use Google Cloud. It takes too much time and attention, that would be better spend on focusing on lessons. And tackle setting up own rig after the course.


@radek użył kodu z pierwszej lekcji w konkursie kaggle Quick, Draw!:

a co Wy robicie jako prace domową? :slight_smile:


z wyników głosowania wynika że będziemy spotykać sie wirtualnie w czwartki o 20.00

proponuje platforme jitsi meet, jest to darmowa aplikacja do wideokonferencji, open source, nie potrzeba loginu

wystarczy kliknac na ponizszy link

(Piotr Czapla) #42

Thank you :smiley: , It was really nice and super motivating gesture. Thank you @jeremy :blush:.
The work on German language was done in cooperation with @mkardas but it happened that I’ve created the thread so I ended up on the slides.

As @jeremy said the most important part is to focus on one project and do it well.

I really hope we can make more awesome projects this semester!

Update: I’ve just noticed that @aayushy got a even better model with accuracy of:

(Piotr Czapla) #43

Two sources of conversations & notifications make things more complex to follow and it is important we share as much as we can on the forum. It does not mean that I would not want to join Women in Machine Learning for other conversations than, can you let me in :)?

Me :slight_smile:

I can help you if you want.

Give it a try few times. You get use to it after third lecture and the experience is awesome because you don’t skip any parts and you feel like you are there with a bit of jet lag :).

About 8x times. V100 is 2x faster than 1080ti and 1080ti is 2x to 4x faster than K80.
(k80 have 2 gpu cores but you get only one when you start a vm hence 2x to 4x slower)

I went to sleep before I finished. But since Jeremy is saying that it is better to keep the conversation in public in polish than in private let’s do that. I see the point that it might be easier better for SEO etc. Although I do feel the same as @Blanche, talking polish on mostly english forum does feel a bit odd :slight_smile:

(Michal Wawrzyniuk) #44

Ok in the matter of language two are accepted Shespire and Slowacki
What is most important how we can progress together in ML ?
Personally, I’m the beginer but I see Piotrek is more advanced not sure about rest of group :slight_smile: But doesn’t really matter we are joined by one goal :slight_smile: What would you say to pick one ML challenge and work on it together and all the time follow mainstream of Fastai V3. Is this has sense ?? What you think ?

Astalavista :slight_smile:


i guess there are pros and cons of using Polish language here
pros: it feels nice :slight_smile:
cons: people who can’t speak Polish (most at this forums) cannot join the discussion to learn or help

i am fine with either Shakespeare or Słowacki :slight_smile:


great work @piotr.czapla @mkardas and @aayushy congrats!


yeah it does. do you have any project ideas? maybe kaggle competition?

(Michal Wawrzyniuk) #48

I was thinking about hart EKG interpretation by ML not sure will we get data.
There was some kagle about pneumonia X-ray classification that could be another topic

Anyone any other ideas?



perhaps not what you were looking for but maybe good enough to start?

sounds interesting

(Michal Wawrzyniuk) #50

For me we can start on this

We have data :slight_smile: :slight_smile:

(Michal Wawrzyniuk) #51

This is what is sleep APNEA

it serious sleep disorder can affect anyone.
It visible on ECG.

My plan is to convert ECG to IMAGES label them and use ConvNet to diagnose those
episodes :slight_smile:


I’m going to make simple web app to recognize polish currency bills. DL part will be very easy. The challenge will be working with “edge” models, because I want everything to be on client side. Other problem would be, that I have to learn js :smiley:
I bet I can find some tabular data problem at work also.

If someone has hard time picking some problem, has nice challenges to choose from.


I just installed fastai on my “DL” box (I wouldn’t call my PC with 1060 6gb a DL box, more a Witcher3 box). Was pretty straight forward.
step 1: install nvidia 410 driver (for cuda 92): <- but with
sudo apt install nvidia-410
step 2:
create new envirement
conda create --name mynewenv
source activate mynewenv
and in new env I followed instructions from here

just don’t know why nvidia-smi is showing cuda 10. oh, well…
I’ll guess it will bite me in the ass later, heh.

and little tip when working with notebooks- I copy them and add “tmp_” before editing, so git ignores changes.


Are you planning on building a mobile app or something that will live in the browser?


yeah, something like

this is inspiration for me:
I will be trying to recreate functionality of this app.
The thing is, that it will be more proof-of-concept, because polish bills are different sizes, so blind ppl can recognize different bills without app :slight_smile:

But I will try to keep it simple, because it will also js-learning project for me.
I will be reling on this article:
yeah, something like
But I will try to keep it simple, because it will also js-learning project for me.
I will be relying on those articles:


This is a really cool project :slight_smile: I found this article on deploying pytorch models but it targets IOS.